Tesi Daghia Federica
Active fibre-reinforced composites
with embedded shape memory alloys
by
Federica Daghia
Relatore: Prof. Ing. Erasmo Viola
Correlatore: Prof. Ing. Francesco Ubertini
Dottorato di ricerca in Meccanica delle Strutture – XX ciclo
Coordinatore: Prof. Ing. Erasmo Viola
Keywords:
adaptive structures, fibre-reinforced composites, shape memory alloys
Settore scientifico disciplinare:
Area 08 – Ingegneria Civile e Architettura
ICAR/08 – Scienza delle Costruzioni
ALMA MATER STUDIORUM • Università di Bologna
Bologna, 23 marzo 2008
In theory, there is no difference
between theory and practice.
But, in practice, there is.
— JAN L. A. VAN DE SNEPSCHEUT
Contents
List of Figures
v
List of Tables
ix
Sommario
1
Introduction
3
Capitolo 1 – Sommario
7
1 Active fibre-reinforced composites
1.1 Reinforcement . . . . . . . . . . . .
1.2 Matrix . . . . . . . . . . . . . . . .
1.3 Manufacturing process . . . . . . .
1.3.1 Hand lay-up . . . . . . . . .
1.3.2 Spray-up . . . . . . . . . .
1.3.3 Moulding . . . . . . . . . .
1.3.4 Pultrusion . . . . . . . . . .
1.3.5 Filament winding . . . . . .
1.4 Active materials . . . . . . . . . .
1.4.1 Electrical coupling . . . . .
1.4.2 Thermal coupling . . . . . .
1.4.3 Magnetic coupling . . . . .
1.4.4 Chemical coupling . . . . .
Capitolo 2 – Sommario
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
11
15
16
17
18
18
20
20
20
21
22
22
23
25
ii
Contents
2 Modelling of fibre-reinforced composites
2.1 Analytical modelling of laminated plates . . . . . . . . . . .
2.1.1 Equivalent Single Layer theories . . . . . . . . . . .
2.2 A hybrid-stress finite element for laminated plates . . . . .
2.2.1 Hybrid stress formulation . . . . . . . . . . . . . . .
2.2.2 Finite element assumptions . . . . . . . . . . . . . .
2.2.3 Finite element equations . . . . . . . . . . . . . . . .
2.2.4 Numerical testing of the finite element performance
2.3 Reconstruction of the transverse stresses . . . . . . . . . . .
2.3.1 Recovery by Compatibility in Patches . . . . . . . .
2.3.2 Reconstruction of the transverse shear stresses . . .
2.3.3 Reconstruction of the transverse normal stress . . .
2.3.4 Numerical testing of the reconstruction procedure .
2.4 Bayesian estimation of laminates’ elastic constants . . . . .
2.4.1 Bayesian estimators comparison . . . . . . . . . . . .
2.4.2 A modified procedure . . . . . . . . . . . . . . . . .
2.4.3 Testing of the estimator procedures . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
27
28
30
52
53
54
57
59
66
68
71
71
72
78
80
92
95
Capitolo 3 – Sommario
103
3 Shape memory alloys
3.1 Shape memory alloy properties . . . . . . . . . . . . . . . . . . .
3.1.1 Martensitic phase transitions . . . . . . . . . . . . . . . .
3.1.2 Micromechanical interpretation of shape memory alloy
properties . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Shape memory alloy applications . . . . . . . . . . . . . . . . . .
3.2.1 Couplings and fasteners . . . . . . . . . . . . . . . . . . .
3.2.2 Actuators . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.3 Biomedical devices . . . . . . . . . . . . . . . . . . . . . .
3.2.4 Civil engineering applications . . . . . . . . . . . . . . . .
3.3 Modelling of shape memory alloys . . . . . . . . . . . . . . . . .
3.3.1 Outline of phenomenological thermodynamics . . . . . . .
3.3.2 Outline of plasticity theories . . . . . . . . . . . . . . . .
3.3.3 One-dimensional models . . . . . . . . . . . . . . . . . . .
3.3.4 Three-dimensional models . . . . . . . . . . . . . . . . . .
3.4 Experimental characterisation and modified phase diagram . . .
3.4.1 Detwinned martensite formation at low stress levels . . .
3.4.2 Influence of the loading path on the reverse phase transition temperatures . . . . . . . . . . . . . . . . . . . . . .
105
106
110
113
115
116
116
117
118
120
121
124
125
141
158
159
165
iii
Capitolo 4 – Sommario
4 Active composites with shape memory alloy wires
4.1 Modelling SMAHC . . . . . . . . . . . . . . . . . . .
4.1.1 Finite element formulation . . . . . . . . . .
4.1.2 Host structure plate element . . . . . . . . .
4.1.3 Shape memory alloy wire element . . . . . . .
4.1.4 Numerical validation . . . . . . . . . . . . . .
4.2 Building and testing shape control SMAHC . . . . .
4.2.1 SMAHC beam . . . . . . . . . . . . . . . . .
4.2.2 SMAHC plate . . . . . . . . . . . . . . . . . .
179
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
181
184
187
188
190
192
193
194
205
Conclusions and perspectives
219
Bibliography
221
iv
Contents
List of Figures
1.1
1.2
1.3
1.4
1.5
1.6
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
Typical stress-strain curves for various fibres . . . . . . . . . . .
Specific properties of fibres and bulk materials . . . . . . . . . .
Types of reinforcement . . . . . . . . . . . . . . . . . . . . . . . .
Production rate versus quality of the composite for some manufacturing processes . . . . . . . . . . . . . . . . . . . . . . . . . .
Laminae fibre orientation . . . . . . . . . . . . . . . . . . . . . .
Classification of some active materials according to the stressstrain response . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
12
14
Plate-like body . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Material and problem coordinate systems . . . . . . . . . . . . .
Generalised strains for the CLPT . . . . . . . . . . . . . . . . . .
Generalised strains for the FSDT . . . . . . . . . . . . . . . . . .
Generalised strains for the HSDT . . . . . . . . . . . . . . . . . .
Coordinate systems for the plate element . . . . . . . . . . . . .
Simply supported square plate, regular and distorted mesh patterns
Convergence in energy norm for sinusoidal load and both the
stacking sequences . . . . . . . . . . . . . . . . . . . . . . . . . .
Convergence of the transverse displacement at point A for different loads and stacking sequences . . . . . . . . . . . . . . . . . .
Convergence of the in-plane displacement at point B for the
(0/90) stacking sequence and both uniform and sinusoidal load .
Convergence of the stress resultants at point C for the (0/90/0)
stacking sequence and uniform load . . . . . . . . . . . . . . . . .
Convergence of the stress resultants at point C for the (0/90)
stacking sequence and uniform load . . . . . . . . . . . . . . . . .
Element patch for the RCP . . . . . . . . . . . . . . . . . . . . .
31
32
34
39
44
54
59
16
18
21
60
61
61
62
63
69
vi
List of Figures
2.14 Convergence in energy norm of the shear stress profiles for sinusoidal load and both the stacking sequences, using HQ4 with or
without RCP recovery . . . . . . . . . . . . . . . . . . . . . . . .
2.15 Convergence of the moments at point D by HQ4 with or without
RCP recovery, for the (0/90/0) stacking sequence and sinusoidal
load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.16 Convergence of the moments at point D by HQ4 with or without
RCP recovery, for the (0/90) stacking sequence and sinusoidal load
2.17 Shear stress profiles at point C using RCP recovery on various
regular meshes, for the stacking sequence (0/90/0) and sinusoidal
load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.18 Shear stress profiles at point C using RCP recovery on various
regular meshes, for the stacking sequence (0/90) and sinusoidal
load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.19 Shear stress profiles at point C with and without RCP recovery
on 9×9 meshes for the stacking sequence (0/90/0) and sinusoidal
load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.20 Shear stress profiles at point C with and without RCP recovery
on 9 × 9 meshes for the stacking sequence (0/90) and sinusoidal
load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.21 Forward sensitivity analysis for Plate 1 . . . . . . . . . . . . . . .
2.22 Forward sensitivity analysis for Plate 2 . . . . . . . . . . . . . . .
2.23 Natural modes corresponding to the frequencies considered in the
analysis (Plate 2c) . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1
3.2
3.3
3.4
3.5
3.6
73
74
74
76
76
77
77
90
91
96
Shape memory effect . . . . . . . . . . . . . . . . . . . . . . . . . 107
Superelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Shape memory effect and superelasticity . . . . . . . . . . . . . . 108
Shape recovery under applied load . . . . . . . . . . . . . . . . . 109
One-way and two-way shape memory effect . . . . . . . . . . . . 110
Formation of different martensitic variants in a single grain of
polycrystalline CuZnAl alloy during uniaxial tensile loading (Patoor et al., 2006) . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.7 Crystal structures and arrangements in the stress-temperature
plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.8 Free energy of austenite and martensite as a function of temperature112
3.9 Loading path and phase transitions for the shape memory effect 114
3.10 Loading path and phase transitions for superelasticity . . . . . . 114
vii
3.11 Phase diagrams for Tanaka (1986), Liang and Rogers (1990) and
Brinson (1993) models . . . . . . . . . . . . . . . . . . . . . . . .
3.12 Phase diagrams for Auricchio and Sacco’s models . . . . . . . . .
3.13 Internal variables for Popov and Lagoudas (2007) model . . . . .
3.14 Phase diagram for Popov and Lagoudas (2007) model . . . . . .
3.15 Schematic representation of a DSC measurement for shape memory alloys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.16 Uniaxial phase diagrams . . . . . . . . . . . . . . . . . . . . . . .
3.17 Constrained recovery sketch in Brinson’s phase diagram . . . . .
3.18 Nitinol wires in constrained recovery . . . . . . . . . . . . . . . .
3.19 Comparison between original and modified Brinson’s models . . .
3.20 DSC for shape memory alloy wire with 4% elongation . . . . . .
3.21 DSC test for the virgin specimen . . . . . . . . . . . . . . . . . .
3.22 First heating curve for specimens strained at room temperature .
3.23 Strain versus peak temperature values for test 1 . . . . . . . . . .
3.24 First heating curve for specimens strained at room temperature
- detail of 2% strain . . . . . . . . . . . . . . . . . . . . . . . . .
3.25 First heating curve for specimens thermally cycled under 200
MPa constant stress . . . . . . . . . . . . . . . . . . . . . . . . .
3.26 Applied stress versus peak temperature values for test 2 . . . . .
3.27 First heating curve for specimens thermally cycled at 4% constant
strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
4.10
4.11
4.12
4.13
4.14
4.15
4.16
SMAHC plate and reference systems . . . . . . . . . . . . .
Rigid body connections . . . . . . . . . . . . . . . . . . . .
Numerical validation: stiffness and shape control . . . . . .
SMAHC specimen . . . . . . . . . . . . . . . . . . . . . . .
Experimental setup for the SMAHC beam . . . . . . . . . .
SMAHC beam response - cycle 1 . . . . . . . . . . . . . . .
SMAHC beam response - cycles 1, 3 and 5 . . . . . . . . . .
SMAHC beam response - cycles 1, 3, 5 and 10 . . . . . . . .
Simulation of SMAHC beam response - cycle 1 . . . . . . .
Simulation of SMAHC beam response - cycles 1, 3 and 5 . .
Simulation of SMAHC beam response - cycles 1, 3, 5 and 10
VARTM lay-up . . . . . . . . . . . . . . . . . . . . . . . . .
VARTM composite making . . . . . . . . . . . . . . . . . .
VARTM plate actuation . . . . . . . . . . . . . . . . . . . .
DSC measurement on trained wires . . . . . . . . . . . . . .
Pre-preg SMAHC plate making . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
128
132
144
149
151
158
160
161
164
166
167
170
171
171
172
173
175
189
191
192
197
198
200
200
201
203
203
204
206
206
207
209
211
viii
4.17
4.18
4.19
4.20
4.21
List of Figures
Pre-preg SMAHC plate . . . . . . . . . . . . . . . .
Experimental setup for the pre-preg SMAHC plate .
Pre-preg SMAHC plate actuation . . . . . . . . . . .
Pre-preg SMAHC plate response for different cycles
Simulation of pre-preg SMAHC plate response . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
212
213
214
215
217
List of Tables
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14
2.15
2.16
2.17
2.18
2.19
2.20
2.21
Transverse displacement w̄ at point A for simply supported (0/90/0)
laminate subjected to uniform load . . . . . . . . . . . . . . . . .
Transverse displacement w̄ at point A and in-plane displacement
ū1 at point B for simply supported (0/90) laminate subjected to
uniform load . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Stress resultants at point C for simply supported (0/90/0) laminate subjected to uniform load . . . . . . . . . . . . . . . . . . .
Stress resultants at point C for simply supported (0/90) laminate
subjected to uniform load . . . . . . . . . . . . . . . . . . . . . .
Characteristics of Plate 1 and Plate 2 . . . . . . . . . . . . . . .
Input frequencies of Plate 1 and Plate 2 . . . . . . . . . . . . . .
Test 1 on Plate 1: final estimates and iterations . . . . . . . . . .
Test 2 on Plate 1: final estimates and iterations . . . . . . . . . .
Test 3 on Plate 1: final estimates and iterations . . . . . . . . . .
Test 1 on Plate 2: final estimates and iterations . . . . . . . . . .
Test 2 on Plate 2: final estimates and iterations . . . . . . . . . .
Test 3 on Plate 2: final estimates and iterations . . . . . . . . . .
Test 1 on Plate 1: MVE-mod final estimates and iterations . . .
Test 2 on Plate 1: MVE-mod final estimates and iterations . . .
Test 3 on Plate 1: MVE-mod final estimates and iterations . . .
Test 1 on Plate 2: MVE-mod final estimates and iterations . . .
Test 2 on Plate 2: MVE-mod final estimates and iterations . . .
Test 3 on Plate 2: MVE-mod final estimates and iterations . . .
Characteristics of Plate 1c and Plate 2c . . . . . . . . . . . . . .
Input frequencies of Plates 1c and 2c . . . . . . . . . . . . . . . .
Final estimates and iterations for Plates 1c and 2c . . . . . . . .
63
64
64
65
87
87
88
88
88
89
89
89
93
93
94
94
94
95
97
97
97
x
List of Tables
2.22 Final estimates and iterations for Plate 2c - pseudoexperimental
frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.23 Characteristics of cross-ply and angle-ply plates . . . . . . . . . . 99
2.24 Input frequencies for cross-ply and angle-ply plates . . . . . . . . 99
2.25 Final estimates and iterations for cross-ply and angle-ply plates . 99
2.26 Characteristics of carbon-epoxy plate . . . . . . . . . . . . . . . . 101
2.27 Measured frequencies for carbon-epoxy plate . . . . . . . . . . . 101
2.28 Estimates for carbon-epoxy plate . . . . . . . . . . . . . . . . . . 101
3.1
3.2
4.1
4.2
4.3
4.4
4.5
Comparison between the different models proposed by Auricchio
and Sacco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Nitinol material properties . . . . . . . . . . . . . . . . . . . . . . 163
Axial and flexural rigidities of laminate strips . . . . . . . . . . .
Constitutive matrices for the finite element model, units N and
mm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Nitinol wires: material properties . . . . . . . . . . . . . . . . . .
Pre-preg glass fibres composite lamina: material properties . . .
Trained Nitinol wires: material properties . . . . . . . . . . . . .
195
196
196
209
209
Sommario
1
Compositi fibro-rinforzati attivi
con leghe a memoria di forma
La tesi riguarda i compositi fibro-rinforzati contenenti fili a memoria di forma.
L’applicazione strutturale dei materiali attivi permette lo sviluppo di strutture adattive, in grado di rispondere attivamente a cambiamenti dell’ambiente
circostante. In particolare, attuatori a memoria di forma inseriti in un composito ne controllano la forma o la rigidezza, influenzando le proprietà statiche
e dinamiche del composito stesso. Le applicazioni possibili sono molteplici ed
includono la prevenzione dell’instabilità termica di pannelli esterni di aeromobili, il controllo della forma per il miglioramento aerodinamico e l’apertura e il
posizionamento di strutture spaziali espandibili.
Lo studio dei compositi attivi è un argomento complesso e multidisciplinare,
che richiede una profonda comprensione del comportamento accoppiato tipico
dei materiali attivi, oltre che dell’interazione tra i diversi costituenti del composito. Sia i compositi fibro-rinforzati che le leghe a memoria di forma sono argomenti di ricerca molto attivi, che presentano una serie di problemi aperti di tipo
sia sperimentale che di modellazione. Per questo, nonostante la tesi si concentri
sui compositi attivi, alcuni risultati qui presentati possono essere utilizzati nel
contesto dei compositi fibro-rinforzati tradizionali o di altre applicazioni delle
leghe a memoria di forma.
Il corpo della tesi è suddiviso in quattro capitoli.
Nel primo capitolo sono introdotti i compositi attivi fibro-rinforzati, presentando una breve panoramica sulle principali tipologie di rinforzo, matrice e
processo produttivo per i materiali compositi, insieme ad una introduzione e
classificazione di alcuni materiali attivi.
Nel secondo capitolo sono esaminate alcune problematiche relative ai compositi fibro-rinforzati tradizionali. I principali modelli bidimensionali per piastre
composite vengono derivati a partire da un solido tridimensionale soggetto a
vincoli cinematici. Nel contesto della teoria al primo ordine, viene formulato
un nuovo elemento finito semplice, accurato e computazionalmente efficiente,
a partire da un approccio ibrido agli sforzi. La soluzione del problema bidimensionale è utilizzata per la ricostruzione delle tensioni lungo lo spessore del
laminato e si sviluppa una procedura in due fasi che utilizza il Recovery by
Compatibility in Patches, formulato recentemente e qui esteso al caso di pia-
2
Sommario
stre laminate, per migliorare l’accuratezza della soluzione bidimensionale. La
modellazione di elementi compositi anisotropi richiede di determinare un significativo numero di parametri elastici, che possono essere stimati tramite tecniche
numerico-sperimentali. In questo contesto, vengono analizzati due diversi estimatori bayesiani, evidenziando analogie e differenze tra i due. Inoltre, viene
proposta una procedura alternativa per migliorare la velocità di convergenza.
Nel terzo capitolo sono introdotte le leghe a memoria di forma, descrivendone
proprietà e principali applicazioni. Diversi modelli costitutivi mono e triassiali
di letteratura, basati su analogie con la plasticità, sono messi a confronto in
maniera critica, evidenziandone potenzialità e limiti, legati principalmente alla
definizione del diagramma di fase e alla scelta delle variabili interne. Successivamente, sono riportati alcuni risultati sperimentali sulla caratterizzazione di fili
a memoria di forma. Tali risultati evidenziano fenomeni non rappresentati dalla
corrente modellazione costitutiva; si propongono alcune idee per lo sviluppo di
un modello alternativo.
Il quarto capitolo riguarda le piastre composite attive con fili a memoria di
forma. Gli approcci di modellazione che possono essere impiegati per descrivere
il comportamento di tali strutture sono numerosi e presentano differenti vantaggi e svantaggi, sia dal punto di vista della complessità che della versatilità.
Un modello semplice, utile per descrivere le configurazioni di controllo della
forma e della rigidezza in un contesto comune, è proposto e implementato. Successivamente, il modello è validato sperimentalmente, facendo riferimento alla
configurazione di controllo della forma, la quale risulta essere la più sensibile
alla scelta dei parametri del modello. Il lavoro sperimentale è suddiviso in due
parti. Nella prima parte, un elemento composito attivo è realizzato incollando
alcuni fili a memoria di forma predeformati su una striscia di laminato in fibra
di carbonio. Si tratta di una struttura relativamente semplice da realizzare, ma
utile a dimostrare sperimentalmente il concetto illustrato nella prima parte del
capitolo. Nella seconda parte, si affrontano le problematiche relative alla realizzazione di un vero e proprio composito fibro-rinforzato con fili a memoria di
forma integrati, considerando diverse scelte in termini di materiali e tecnologie
impiegate. Anche se rimangono da affrontare diversi aspetti tecnologici, i risultati ottenuti dalla sperimentazione permettono di dimostrare il meccanismo di
funzionamento dei compositi con fili a memoria di forma, risultando inoltre in
ottimo accordo con le previsioni del modello proposto.
Introduction
This dissertation concerns active fibre-reinforced composites with embedded
shape memory alloy wires. The structural application of active materials allows to develop adaptive structures which actively respond to changes in the
environment, such as morphing structures, self-healing structures and power
harvesting devices. In particular, shape memory alloy actuators integrated
within a composite actively control the structural shape or stiffness, thus influencing the composite static and dynamic properties. Envisaged applications
include, among others, the prevention of thermal buckling of the outer skin of
air vehicles, shape changes in panels for improved aerodynamic characteristics
and the deployment of large space structures.
The study and design of active composites is a complex and multidisciplinary
topic, requiring in-depth understanding of both the coupled behaviour of active materials and the interaction between the different composite constituents.
Both fibre-reinforced composites and shape memory alloys are extremely active research topics, whose modelling and experimental characterisation still
present a number of open problems. Thus, while this dissertation focuses on
active composites, some of the research results presented here can be usefully
applied to traditional fibre-reinforced composites or other shape memory alloy
applications.
The dissertation is composed of four chapters.
In the first chapter, active fibre-reinforced composites are introduced by giving an overview of the most common choices available for the reinforcement,
matrix and production process, together with a brief introduction and classification of active materials.
The second chapter presents a number of original contributions regarding
4
Introduction
the modelling of fibre-reinforced composites. Different two-dimensional laminate theories are derived from a parent three-dimensional theory, introducing
a procedure for the a posteriori reconstruction of transverse stresses along the
laminate thickness. Accurate through the thickness stresses are crucial for the
composite modelling as they are responsible for some common failure mechanisms. A new finite element based on the First-order Shear Deformation Theory
and a hybrid stress approach is proposed for the numerical solution of the twodimensional laminate problem. The element is simple and computationally efficient. The transverse stresses through the laminate thickness are reconstructed
starting from a general finite element solution. A two stages procedure is devised, based on Recovery by Compatibility in Patches and three-dimensional
equilibrium. Finally, the determination of the elastic parameters of laminated
structures via numerical-experimental Bayesian techniques is investigated. Two
different estimators are analysed and compared, leading to the definition of an
alternative procedure to improve convergence of the estimation process.
The third chapter focuses on shape memory alloys, describing their properties and applications. A number of constitutive models proposed in the literature, both one-dimensional and three-dimensional, are critically discussed and
compared, underlining their potential and limitations, which are mainly related
to the definition of the phase diagram and the choice of internal variables. Some
new experimental results on shape memory alloy material characterisation are
also presented. These experimental observations display some features of the
shape memory alloy behaviour which are generally not included in the current
models, thus some ideas are proposed for the development of a new constitutive
model.
The fourth chapter, finally, focuses on active composite plates with embedded shape memory alloy wires. A number of different approaches can be used
to predict the behaviour of such structures, each model presenting different advantages and drawbacks related to complexity and versatility. A simple model
able to describe both shape and stiffness control configurations within the same
context is proposed and implemented. The model is then validated considering
the shape control configuration, which is the most sensitive to model parameters. The experimental work is divided in two parts. In the first part, an active
composite is built by gluing prestrained shape memory alloy wires on a carbon
fibre laminate strip. This structure is relatively simple to build, however it
5
is useful in order to experimentally demonstrate the feasibility of the concept
proposed in the first part of the chapter. In the second part, the making of a
fibre-reinforced composite with embedded shape memory alloy wires is investigated, considering different possible choices of materials and manufacturing
processes. Although a number of technological issues still need to be faced, the
experimental results allow to demonstrate the mechanism of shape control via
embedded shape memory alloy wires, while showing a good agreement with the
proposed model predictions.
6
Introduction
Capitolo 1 – Sommario
7
Compositi fibro-rinforzati attivi
I materiali compositi sono formati da due o più costituenti diversi, ciascuno dei
quali contribuisce a determinarne il comportamento complessivo. Più comunemente, il termine compositi si riferisce a materiali costituiti da una matrice la cui
funzione è di unire e proteggere le fibre, che costituiscono il principale elemento
resistente del composito. Anche limitandosi ai soli compositi fibro-rinforzati si
riscontra un’enorme varietà di materiali diversi.
Tra i primi ad interessarsi di compositi fu negli anni ’60 l’industria aerospaziale, in cerca di soluzioni per la realizzazione di componenti dal peso ridotto.
Negli ultimi decenni, i compositi hanno cominciato ad essere utilizzati in altri
settori, quali l’industria automobilistica, sportiva e delle costruzioni.
Il principale vantaggio nell’utilizzo dei materiali compositi è l’ottimo rapporto peso/resistenza, oltre alla possibilità di progettare il materiale a seconda
delle necessità strutturali. Definendo la posizione delle fibre e il numero di strati
è possibile progettare le proprietà del materiale in funzione della direzione di
caricamento e dei requisiti locali di resistenza e rigidezza. Questa flessibilità
implica però una notevole complessità associata alla progettazione di strutture
composite rispetto alle strutture tradizionali, dato che il materiale e il processo
di produzione devono essere definiti insieme alla struttura stessa.
Questi problemi, in aggiunta al costo ancora elevato dei materiali compositi, hanno finora ritardato l’utilizzo di tali materiali in applicazioni strutturali.
Recentemente la tendenza si sta invertendo e cominciano a nascere nuove idee
sulle possibili combinazioni di materiali diversi per ottenere un’infinita varietà
di proprietà strutturali. Lo sviluppo di materiali attivi, quali i materiali piezoelettrici e le leghe a memoria di forma, ha aperto la strada alla realizzazione
di compositi attivi. I materiali attivi mostrano un accoppiamento tra il campo
meccanico e altri campi della fisica (termico, elettrico, magnetico, ecc.), che può
essere sfruttato per realizzare sensori o attuatori. Inserendo un materiale attivo
in un composito, questo svilupperà caratteristiche attive. La principale applicazione dei compositi attivi consiste nel controllo delle proprietà strutturali, che
possono adattarsi a cambiamenti delle condizioni esterne.
Lo sviluppo dei compositi attivi presenta ulteriori difficoltà rispetto ai compositi fibro-rinforzati tradizionali. La modellazione di compositi attivi richiede
la descrizione del campo meccanico e di quello accoppiato, coinvolgendo mecca-
8
Capitolo 1 – Sommario
nismi di accoppiamento che in alcuni casi non sono ancora pienamente compresi.
Per materiali attivi di recente concezione, inoltre, non sono ancora disponibili
informazioni sul comportamento a lungo termine. Tutte queste difficoltà rendono i compositi attivi un tema di ricerca molto complesso e promettente, che
coinvolge numerose discipline (chimica, scienza dei materiali, ingegneria strutturale, etc.) per ottenere comportamenti strutturali che sarebbero impensabili
con materiali tradizionali.
In questo capitolo, una breve descrizione di alcune tra le più comuni possibilità a disposizione per la definizione del rinforzo, della matrice e del processo
produttivo per i compositi fibro-rinforzati tradizionali, insieme ad una panoramica su alcuni materiali attivi, permette di illustrare le enormi potenzialità dei
compositi attivi.
Chapter 1
Active fibre-reinforced
composites
Composites are materials made up of two or more constituents, each bestowing
different characteristics on the overall material behaviour. A number of very
different materials belong to this broad category, from traditional structural
materials such as concrete and masonry all the way to newly devised polymers
with embedded nanoparticles such as carbon nanotubes.
More commonly, the term composite refers to materials constituted by a matrix, whose function is to bind together and protect the fibres, which constitute
the main load-bearing elements of the composite. Even within the narrower
definition of fibre-reinforced composites, the variety of available materials is
enormous. In particular, composites can be classified according to the type of
matrix, reinforcement and technological process (Edwards, 1998).
Fibre-reinforced composites are becoming increasingly popular in a wide
range of applications. Historically, among the first to take an interest in composites was the aerospace industry in the 1960s (Soutis, 2005). The main reason
was the significant decrease in weight of composites as compared to traditional
materials. In the last decades, however, composites have started to penetrate
other industries, such as automotive, sporting goods and construction, where
they are mainly used as reinforcement to repair and enhance the performance
of existing structures (Bastianini et al., 2005; Stratford et al., 2004).
10
Active fibre-reinforced composites
Among the advantages of composite materials are the high strength to weight
ratio and the possibility to tailor the material to the structural needs. Indeed,
by selecting the fibre placement and number of layers it is possible to design
the material properties according to the direction of loading and the locally
required strength and stiffness properties. This, however, implies a significant
complexity in the design of composite structures as compared to traditional ones,
since the material and the production process need to be designed together with
the structure itself. Moreover, the material behaviour of composites in both
elastic and inelastic range is rather complex to predict, as it involves mesoscale
interaction of the different constituents.
The mentioned issues, together with the still high cost of composite materials
as compared to traditional solutions, have delayed the massive use of composites
in structural applications. In recent years, however, the interest in these kinds
of materials has been ever increasing. A significant example of the advances
in composite use is the new aircraft Boeing 787 Dreamliner, scheduled to enter
into service in 2008: it is the first aircraft whose primary structure is made
for as much as 50% of composites — including the fuselage and the wing. In
this case, the use of composites greatly simplifies the structural elements: by
manufacturing a one-piece fuselage section, the designers were able to eliminate
1,500 aluminium sheets and 40,000 - 50,000 fasteners (source: Boeing website1 ).
As fibre-reinforced composites become established, more ideas arise on how
best to combine different materials to achieve a virtually infinite variety of
structural properties. The development of active materials, such as piezoelectric
materials and shape memory alloys, has paved the way to the creation of active
composites. Active materials (also referred to as smart or intelligent materials)
exhibit coupling between the mechanical field and some other physical field
(thermal, electric, magnetic etc.), thus a stimulus in one of the fields provokes a
response in the coupled field. This coupling can be exploited for sensing and/or
actuation. In sensing, a stimulus in the mechanical field provokes a response in
the coupled field which can be measured, while in actuation the coupled field is
used to provoke a mechanical response. By embedding an active material in a
composite, the composite itself can exhibit active properties.
Active composites can be used for many purposes, the main idea being to
control and adapt the structural properties according to a change in external
1 www.boeing.com
1.1 Reinforcement
11
conditions (adaptive composites). Embedded sensing allows to monitor the
structural conditions, while actuators can change the dynamic properties, stiffness or shape according to need. Even more, embedded systems can be designed
with self-diagnosis and self-healing features.
The development of active composites poses even more challenges than traditional fibre-reinforced composites. Modelling active materials requires to take
into account both the mechanical and the coupled field, thus increasing the size
and complexity of the problem and, in many cases, introducing sources of non
linearity. Moreover, in some cases the mechanisms underlying the coupled behaviour are still not fully understood, thus further investigation on the active
material itself is required prior to considering the composite. Finally, for many
active materials of recent conception not much information is available regarding the long term behaviour, such as fatigue life and stability of the response
with increasing numbers of cycles. All the stated difficulties make active composites a very challenging yet promising topic for researchers, which involves
many different disciplines (chemistry, material science, structural engineering
etc.) working together to achieve structural performances which would have
been unthinkable with traditional materials.
An overview of the most common choices available for the reinforcement, matrix and production process for traditional fibre-reinforced composites (Mazumdar, 2002), together with a brief introduction and classification of active materials, allow to get a glimpse of the enormous potential of active fibre-reinforced
composites.
1.1
Reinforcement
Fibres are the main load-bearing element of the composite, thus it is mostly
on the choice of the reinforcement that the composite mechanical properties
depend. Choosing a reinforcement requires to define the type of fibres, their
size and arrangement.
Fibres can be made of a wide variety of materials, both inorganic (e.g. metal,
glass, carbon fibres) and organic (e.g. aramidic, natural fibres). Because of the
small size of the cross-section (5 to 25 µm for the most common fibres) and the
reduced amount of defects, fibres are generally stiffer and stronger than material
in bulk form and tend to break in a fragile manner. In Figure 1.1, typical stress
12
Active fibre-reinforced composites
2.5
Carbon HM Carbon HS
2
Kevlar 49
Kevlar 29
stress (GPa)
S - glass
E - glass
1.5
1
0.5
0
0
0.005
0.01
0.015
0.02
0.025
0.03
strain
Figure 1.1: Typical stress-strain curves for various fibres
3
Specific strength (GPa cm /g)
2.5
Kevlar 29 Kevlar 49
S - glass
2
Carbon HS
1.5
E - glass
Carbon HM
1
0.5
Steel
Aluminum
0
0
50
100
150
3
200
Specific modulus (GPa cm /g)
Figure 1.2: Specific properties of fibres and bulk materials
1.1 Reinforcement
13
strain curves in uniaxial tension are shown for various types of fibres, whereas
in Figure 1.2 the specific properties of various fibres are compared with those
of bulk steel and aluminium.
Glass fibres are the cheapest among the reinforcements, thus they are the
most common in applications. Their elastic modulus is comparatively low
(slightly better for the S-glass type than the E-glass), thus they are used for
both structural and non structural applications. Carbon fibres come in different varieties, classified as high strength and high modulus fibres. Their elastic modulus and specific elastic modulus are the highest, thus they are often
favoured in low-weight structural applications. However, their cost is significantly higher than that of glass fibres. Aramidic fibres are synthetic organic
fibres, also known with the commercial name of Kevlar. When drawn, aramidic
fibres acquire anisotropic properties, thus their behaviour in compression is significantly worse than in tension. Because of their microstructure, aramidic fibres
break in a more ductile way as compared to glass or carbon fibres; for this reason they are used in applications where energy dissipation is required, such as
motorcycle helmets and bulletproof vests.
A variety of different reinforcements can be obtained from the same type
of fibres, from thin two dimensional mats or fabrics to complex-shaped threedimensional preforms. The properties of the resulting composite depend mainly
on the fibre length (long versus short fibres) and orientation (random versus
oriented). In general, the best material properties are obtained with long continuous fibres which are oriented in the loading direction. This situation allows
to maximise the use of the material, but yields extremely anisotropic composites and may not be the best configuration, depending upon the application
and manufacturing process. Nearly isotropic properties can be obtained with
randomly oriented fibres, which can be held together with a resinous binder to
form a mat. Depending upon the length of the fibres, this is called chopped
fibres mat (short fibres, Figure 1.3(a)) or continuous mat (long fibres, Figure
1.3(b)).
Long continuous oriented fibres are arranged in bundles called rovings, which
can be used to make woven or non woven fabric. Woven fabric is made of fibres
woven together to form a single layer and it is classified according to the weave
style. Some common weave styles are shown in Figure 1.3(d). The weave style
defines the amount of fibres in each direction, thus it influences the degree of
14
Active fibre-reinforced composites
(a) Chopped fibres mat
plain
weave
(b) Continuous mat
satin
weave
(c) Non crimp fabric
twill
weave
(d) Some weave styles for woven fabric
Figure 1.3: Types of reinforcement
in-plane anisotropy of the composite. In plain weave, for example, the same
amount of fibres are laid along two orthogonal directions, thus the composite
properties along these two directions are the same. Non woven fabric, also called
non crimp fabric, is made of fibres aligned parallel to each other and stitched
together with polyester thread (Figure 1.3(c)). This allows the fibres to remain
straight, thus increasing the strength of the composite with respect to woven
fabric. Moreover, in non crimp fabric the fibres are aligned all along the same
direction, resulting in the most anisotropic composite behaviour.
1.2 Matrix
1.2
15
Matrix
The matrix has the role of binding and protecting the fibres in a composite.
Many different materials can be used as matrix, such as metals, cement and
polymers. Polymer matrices are by far the most common in composites applications and they are the only ones discussed here.
Polymers are subdivided into thermoplastic and thermoset resins. Because of
differences in the microstructure, thermoplastic resins can repeatedly melt and
harden when heated and cooled, whereas thermoset resins cannot be reformed
once cured. Thermoplastics tend to have lower mechanical characteristics and
higher viscosity than thermosets, thus they are often used for non structural
applications and will not be dealt with in details here.
The selection of a polymer matrix is determined by a variety of factors, involving both the service life of the composite (maximum service temperature,
fire and corrosion resistance, mechanical properties, etc.) and the manufacturing process (duration, temperature and pressure required in the cure cycle,
viscosity, etc.). As regards the composite manufacturing, the resin can be given
in liquid, semi-solid and solid form depending on the chosen process. To acquire
its service characteristics, it undergoes a curing process, that is a chemical reaction which results in the formation of a three-dimensional cross-linked network.
Depending upon the resin chemical composition and the properties which are
to be achieved in the final composite, curing can be carried out in different conditions of pressure and temperature, with or without the presence of additives
and catalysts.
Epoxy resins are the most widely used for structural applications because of
their good mechanical properties. They are extremely versatile, as their properties and cure cycle can be tailored to meet various application needs. Their
maximum service temperature goes from 90◦ − 120◦ C all the way up to 200◦ C2 .
Bismaleimide and polyimide resins provide higher service temperatures than
epoxies (up to more than 350◦ C), whereas lower cost resins include polyesters
and phenolic resins, the last ones providing also flame resistance and low smoking properties.
2 An indication of the maximum service temperature is given by the polymer’s glass transition temperature Tg , which is a selection parameter in resin producers’ data sheets.
Active fibre-reinforced composites
Hand lay-up
Pultrusion
Filament
winding
Spray-up
LOW
Quality of composite
HIGH
MEDIUM
16
SLOW
MEDIUM
Production rate
FAST
Figure 1.4: Production rate versus quality of the composite for some manufacturing processes
1.3
Manufacturing process
Many different composite manufacturing techniques are available, which differ
in the quality of the final product, cost, production rate, achievable size and
shape. Indeed, all these factors need to be taken into consideration when the
process is selected. In particular, good quality of the resulting composite and
high rate of production are usually contrasting needs, as the processes that
yield the best results in terms of fibre quantity and placement are usually more
time consuming. An exception is pultrusion, which is however suitable only for
cylindrical shaped composites. The achievable quality and production rate of
some manufacturing processes are reported in Figure 1.4.
Composite manufacturing consists of four phases: impregnation, lay-up, consolidation and solidification. In impregnation, fibres are thoroughly covered with
matrix to ensure a good fibre-matrix interface. Naturally, in this phase matrix
viscosity is a key factor. Depending upon the process, dry fibres or fibres already impregnated by the materials supplier (pre-pregs) are used. The lay-up
1.3 Manufacturing process
17
consists in positioning different layers of reinforcement one on top of the other
to make a laminated composite. Each layer (lamina) can be made with a different reinforcement and the fibre orientation can vary from one layer to the
other. Consolidation is a crucial phase which influences the quality of the final
composite: a pressure is applied to remove voids and create the best possible
contact between the laminae. Depending on the process, pressure can be applied
by hand, with a vacuum pump, in autoclave or press. Finally, solidification fixes
the composite shape: it is achieved by resin curing and can last from less than
a minute to several hours.
The placement and orientation of the layers are conventionally defined by
the laminate’s stacking sequence, or lamination scheme. The laminae, or plies,
are listed from the bottom to the top layer and defined by the fibre orientation with respect to an axis associated to the laminate (see Figure 1.5). If
not otherwise specified, all laminae have the same thickness. An example of
stacking sequence is (0/45/90/ − 45/0). When ply stacking sequence, material and geometry are symmetric about the laminate midplane, the laminate
is symmetric, and can be denoted by only the first half of the stacking se¯
quence: (0/45/90)s = (0/45/90/90/45/0), or (0/45/90)
s = (0/45/90/45/0)
(the bar marks a single central layer). An antisymmetric laminate is a laminate whose stacking sequence is antisymmetric about the midplane, for example (0/45/30/ − 30/ − 45/0). Cross-ply laminates are made up of layers with 0◦
and 90◦ orientation, angle ply laminates include also other stacking angles.
In the following, some composite manufacturing processes are described.
1.3.1
Hand lay-up
Hand lay-up is a very labour intensive technique, whose main advantage is to
allow the production of good quality parts which can be large and complexshaped. It consists in placing the layers on a lower mould by hand and then
applying pressure and temperature by means of a vacuum bag or in autoclave to
allow resin curing. Two main techniques are included in hand lay-up: pre-preg
lay-up and wet lay-up. In the first case, fabrics which are already impregnated
with the resin are used, in the second alternate layers of dry fibres and liquid
resin are applied using a roller. The use of pre-pregs allows to control more
effectively the amount of resin and fibres present in the final composite, thus
18
Active fibre-reinforced composites
b
b
a
a
Figure 1.5: Laminae fibre orientation
obtaining a material with better properties. On the other hand, pre-pregs are
more expensive than dry fibres and resin, require storage below room temperature and have a limited shelf life (usually six months to one year). In both
cases, the final result strongly depends upon the skills of the laminator.
1.3.2
Spray-up
As hand lay-up, spray-up is an open mould process which allows to produce
large and complex-shaped parts. It is a much faster technique, which however
gives worse results in terms of composite properties. The fibres and matrix are
sprayed on the mould by using a chopper gun, which chops and wets the fibres
prior to depositing them on the mould. Once the material has been sprayed,
rollers are used to remove entrapped air and the cure is usually done at room
pressure and temperature.
1.3.3
Moulding
Moulding processes allow a high production rate as compared to the techniques
described up to now. On the down side, their use is usually limited to non
1.3 Manufacturing process
19
structural parts because of the low properties obtained with moulding compounds. An exception is Resin Transfer Moulding (RTM), which is able to
produce structural quality parts with a medium production rate. Also, most
moulding techniques employ a lower and upper mould, as opposed to the single side mould used in hand lay-up and spray-up. Although the mould cost
increases, closed-mould processes are advantageous because of the low volatile
emissions and good surface finish on both sides.
Compression moulding and injection moulding make use of moulding compounds as raw materials. These are sheets (Sheet Moulding Compound, SMC)
or bulks (Bulk Moulding Compound, BMC) of chopped glass fibres, resin and
filler. These usually contain a low percentage of fibres (about 25%) and thus
do not yield structural quality material. In compression moulding, a certain
quantity of moulding compound (charge) is positioned in the bottom part of
the mould, then the top part is closed with a certain velocity. The mould is
usually pre-heated to facilitate the resin flow, after a few minutes of cure under
heat and pressure the mould is opened and the composite is removed. In injection moulding, the moulding compound is heated and injected through a nozzle
into the closed mould. Pressure and temperature are maintained during cure,
then after a few minutes the mould is opened and the part removed.
Resin Transfer Moulding differs from the described procedures in that only
the resin is injected. Dry fibre preforms are positioned between the top and
bottom moulds, then low viscosity resin is injected inside the mould. This
allows to control the amount and positioning of fibres and to obtain structural
quality parts with a comparatively fast production rate. Naturally, the resin
viscosity is a crucial issue in this process. In Vacuum Assisted RTM (VARTM),
the resin infiltration and fibre wetting is helped by the presence of void. For very
large parts, such as boat hulls, the top mould can be substituted by a vacuum
bag. In this case, thickness of the resulting composite is less controllable but
the cost of the mould is cut to half. This technique is also known with the name
of resin infusion and is a good and cost-effective alternative to wet lay-up when
it comes to the manufacturing of large and complex parts.
20
1.3.4
Active fibre-reinforced composites
Pultrusion
Pultrusion is a low cost, high production rate, structural quality technique to
produce cylindrical beam-like composites. Fibre rovings are impregnated by
passing through a resin bath, then they are pulled through a heated die where
they cure and achieve the desired shape. Curing occurs with high temperature
and low pressure. The process is continuous, therefore composite beams of any
length can be obtained. The disadvantages of this highly automated process are
the limited shapes that can be achieved and the impossibility to choose the fibre
orientation (unidirectional fibres along the beam axis or, at most, bidirectional
fabrics can be used).
1.3.5
Filament winding
Filament winding is also a highly automated process, which yields good performance composites. Fibre rovings are passed through a resin bath and then
wound over a rotating mandrel at a desired angle. Once the fibre winding is
over, the composite is cured at room or high temperature, then the mandrel is
extracted. The choice of fibre orientation is improved with respect to pultrusion,
even though not every possible orientation can be achieved. As for pultrusion,
the main disadvantage of filament winding is the limitation in the shapes that
can be achieved (tubular, closed parts).
1.4
Active materials
Due to their coupled behaviour, active materials can be used for sensing and/or
actuation purposes. The coupling can be related to intrinsic material characteristics, usually at the microstructural level, or to the ‘intelligent’ combination
of different traditional materials. A variety of materials display active characteristics, thus different classifications are introduced to help identify the most
appropriate for the sought application.
Some classification criteria are discussed here. One possibility, adopted in
the following, is to distinguish active materials according to the coupled field
(electrical, thermal, magnetic, chemical etc.). Other classifications include the
type of material (metal, polymer, ceramic etc.) or the response obtained in
terms of force and/or strain. The last classification is particularly useful when
1.4 Active materials
21
2
Shape
memory
alloys
blocking stress [MPa]
10
1
10
Active
Ceramics
0
10
Electro
Active
Polymers
-1
10
-2
10
10
-2
10
-1
10
0
1
10
% strain
10
2
10
3
Figure 1.6: Classification of some active materials according to the stress-strain
response
choosing a sensor or an actuator for a specific application. Indeed, the order of
magnitude of the stress and strain which can be achieved from different active
materials is extremely variable, as it can be seen in Figure 1.6. In particular,
it is clear from the graph how shape memory alloys are significantly superior
to active ceramics and electroactive polymers in both output stress and strain,
thus making a good candidate for the creation of active composites.
An overview of some active materials and the principle lying at the basis of
their coupled behaviour is given in the following.
1.4.1
Electrical coupling
The most common materials exhibiting electromechanical coupling are piezoelectric materials. The piezoelectric effect consists in an accumulation of electric charges as a consequence of applied stress (direct piezoelectric effect) or
in a deformation generated by an applied electric field (inverse piezoelectric effect). The effect is related to the presence within the material of charged dipoles
22
Active fibre-reinforced composites
whose orientation changes according to the applied electric field. Some crystals
naturally display the piezoelectric effect, whose magnitude can be increased
by the polarization process. Most piezoelectric materials used in applications
are ceramics, but also some polymers are manufactured to display piezoelectric
properties. Piezoceramic wafers can develop large forces, however the associated deformations are rather small (of the order of nanometers), therefore more
complex configurations, such as stacks or bimorphs, are generally created to
enhance the available deformation.
Among the polymers displaying electromechanical coupling, one should include dielectric elastomers. These are devices obtained by placing a layer of
very compliant and nearly incompressible dielectric material between two thin
compliant electrodes. By applying a current to the electrodes, these tend to
get nearer to each other, thus squeezing the dielectric layer which elongates in
the perpendicular direction due to the Poisson effect. Some issues related to
the fabrication and use of dielectric elastomers are the choice of the dielecric
material and the electrodes, which need to be extremely compliant in order not
to break during the actuation, and the use of very high voltages, of the order of
some kV.
1.4.2
Thermal coupling
All materials exhibit thermomechanical coupling, related mostly to the thermal
expansion behaviour. However, in shape memory alloys this coupling is much
stronger than in traditional materials and it is related to complex changes in
the material microstructure. Shape memory alloys are the subject of Chapter 3
and they are not further discussed here.
1.4.3
Magnetic coupling
Materials exhibiting magnetomechanical coupling include magnetorheological
fluids and magnetostrictive materials. Ferromagnetic shape memory alloys also
belong to the latter group. Magnetorheological fluids are liquids whose viscosity
greatly changes with an applied magnetic field, to the point of becoming viscous
solids. Magnetostrictive materials can be deformed via an applied magnetic
field, thanks to the orientation of the magnetisation axis. In ferromagnetic
1.4 Active materials
23
shape memory alloys, different orientation of the martensitic microstructure
with an applied magnetic field are responsible for the macroscopic deformation.
1.4.4
Chemical coupling
A new class of active materials whose coupled behaviour is related to chemical
reactions are ionic polymers. The actuation mechanism is related to the change
in volume of the polymer as it acquires or loses ions in oxidation-reduction
reactions. Due to the type of material and the microstructural changes involved,
very low forces but extremely large deformations can be achieved as compared
to other active materials. The chemical reaction is usually triggered by electrical
current, while the polymer is immersed in an electrolytic solution which provides
the necessary ions. For this reason, ionic polymers are usually classified as wet,
as opposed to piezoelectric polymers or dielectric elastomers, which can work
in a dry environment. Recent research work, however, is aimed at devising new
ways to provide ions for the chemical reaction, which would allow ionic polymers
to operate in air.
24
Active fibre-reinforced composites
Capitolo 2 – Sommario
25
Modellazione dei compositi fibro-rinforzati
Negli ultimi anni l’utilizzo dei materiali compositi in applicazioni ad elevata
tecnologia è incrementato in maniera significativa. Lo sviluppo di nuove applicazioni, incluse quelle basate su materiali attivi integrati nella struttura, richiede
la definizione di strumenti di modellazione semplici ed accurati in grado di descrivere il comportamento strutturale di elementi compositi. Per questo, gli
strumenti analitici e numerici per la modellazione di tali materiali costituiscono
un tema di ricerca molto attivo.
Sono state proposte numerose teorie strutturali per i laminati, molte delle
quali descrivono il composito come un elemento di piastra o guscio bidimensionale. Le teorie bidimensionali sono in genere semplici e richiedono un numero
limitato di parametri, per questo costituiscono una base ideale per lo sviluppo di strategie numeriche quale il metodo degli elementi finiti. D’altra parte,
la riduzione di una struttura tridimensionale multistrato ad un singolo strato
equivalente porta alla perdita di informazioni, in particolare relative alle tensioni lungo lo spessore, che potrebbero rivelarsi cruciali per l’accurata valutazione
del comportamento del composito e dei suoi meccanismi di danneggiamento. La
formulazione di un modello bidimensionale a partire dalla teoria tridimensionale rimuove in parte tale inconveniente, permettendo di ricostruire a posteriori
le tensioni lungo lo spessore. Nella sezione 2.1 vengono derivati alcuni modelli
bidimensionali secondo questa logica e viene discussa la procedura analitica per
la ricostruzione delle quantità tridimensionali.
Una volta scelto un modello analitico, è necessario definire procedure numeriche accurate e computazionalmente efficienti. Nella sezione 2.2 si propone
un nuovo elemento finito basato sulla teoria al primo ordine, definita nella sezione 2.1. Una formulazione ibrida agli sforzi ed un’accurata selezione delle
approssimazioni per sforzi e spostamenti permettono di costruire un elemento
semplice, stabile e computazionalmente efficiente, preparando la strada per la
ricostruzione delle quantità tridimensionali.
La procedura analitica di ricostruzione, descritta nella sezione 2.1, non è
automaticamente applicabile a partire da una soluzione bidimensionale approssimata. Nascono infatti alcuni problemi riguardanti la convergenza delle quantità ricostruite ed il soddisfacimento delle condizioni al contorno del problema
tridimensionale. Nella sezione 2.3 si definisce una strategia generale per la rico-
26
Capitolo 2 – Sommario
struzione delle tensioni lungo lo spessore a partire da soluzioni bidimensionali
agli elementi finiti.
La modellazione dei materiali compositi richiede la conoscenza delle proprietà elastiche del laminato, difficili da determinare con i metodi sperimentali
tradizionali. Le tecniche numerico-sperimentali basate sulla misura delle frequenze naturali del composito rappresentano una valida alternativa ai metodi
statici tradizionali. Nella sezione 2.4 sono messe a confronto alcune tecniche
operanti in un contesto bayesiano, concentrandosi in particolare sull’influenza
dei valori iniziali sul risultato dell’operazione di stima.
Chapter 2
Modelling of
fibre-reinforced composites
Recently, the use of fibre-reinforced composites in high technology applications
has dramatically increased. The development of new composite applications,
including those based on embedded active materials, has engendered the need
for simple and accurate modelling tools to describe their structural behaviour.
In particular, both analytical and numerical tools for composite modelling are
the subject of ongoing research.
Many laminate theories have been proposed, a number of them describing
the composite as a two-dimensional plate or shell structure (see, for example,
Carrera, 2002; Carvelli and Savoia, 1997; Ghugal and Shimpi, 2002; Reddy,
1997). Two-dimensional theories are generally simple and require a limited
number of parameters, thus they are an ideal basis for the development of numerical solution strategies like the finite element method. On the other hand,
reducing a three-dimensional multilayered composite structure to an equivalent single layer leads to a loss of information which might be crucial for the
accurate evaluation of the composite behaviour and failure modes. For this
reason, it is convenient to formulate the two-dimensional model as descending
from a parent three-dimensional theory. This allows to reconstruct some of the
three-dimensional quantities by post processing the two-dimensional solution.
In Section 2.1, the outlined derivation is carried out for various two-dimensional
28
Modelling of fibre-reinforced composites
theories and the post processing allowing to reconstruct the three-dimensional
quantities is discussed.
Once the analytical model has been established, accurate and computationally efficient numerical solution procedures need to be defined. Focusing on the
finite element method, in Section 2.2 a new finite element is proposed, based on
the First-order Shear Deformation Theory established in Section 2.1. A hybrid
stress formulation and a careful selection of the displacement and stress approximations leads to a simple, stable and computationally efficient element and
paves the way for the numerical post processing of the two-dimensional finite
element solution.
Reconstruction of three-dimensional quantities starting from approximate
two-dimensional solutions, such as those obtained via finite elements, is not
straightforward. Indeed, problems arise concerning the convergence of the reconstructed quantities and the satisfaction of three-dimensional boundary conditions. A general strategy for the reconstruction of three-dimensional quantities
starting from a two-dimensional finite element solution is defined in Section 2.3.
Composites modelling requires the knowledge of the laminate material properties, which are not easily determined via traditional testing methods. For
this reason, Section 2.4 deals with the analysis and comparison of numericalexperimental techniques for the estimation of the laminate material properties
from natural frequency data.
2.1
Analytical modelling of laminated plates
The modelling of complex and multi-phase materials such as composites can
be approached from a variety of points of view. In particular, it is possible to
distinguish between micromechanical and macromechanical approaches. In the
first case, the different constituents (fibres and matrix) are modelled separately
and their interaction is taken into consideration, while in the second some homogenisation technique is employed to reduce the overall composite behaviour
to that of a homogeneous equivalent continuum.
The two approaches are fundamentally different in complexity and capability
to model the behaviour of composite materials. Micromechanical approaches,
though complex, are able to reproduce phenomena, such as fibre slipping, matrix cracking and other mechanisms involving the different constituents, which
2.1 Analytical modelling of laminated plates
29
cannot be observed in the context of macromechanical models. These, on the
other hand, are simple and require the introduction of a limited number of variables, thus they are suitable to model large scale structures with a reasonable
computational effort.
A bridge between the two strategies is constituted by the so called multiscale
techniques. These strategies allow to jump back and forth between the micro and
macromechanical approaches, passing information between the two worlds and
observing each phenomenon at the appropriate scale. For the use of multiscale
techniques in the analysis of laminated composites, see Ladevèze et al. (2006).
In the following, only macroscale models are taken into consideration.
Macromechanical approaches, as already pointed out, model each lamina as
a homogeneous material with non isotropic material properties. Low symmetry
material models include monoclinic materials, having a single plane of material
symmetry, and orthotropic materials, which present three orthogonal planes
of material symmetry. Because of the presence of fibres aligned along one or
two orthogonal directions, unidirectional and woven fabric laminates are usually
modelled as orthotropic materials.
Once the lamina has been established as a homogeneous anisotropic continuum, there still remain various choices on the modelling of the laminated
composite. One can distinguish between Equivalent Single Layer (ESL) and
layerwise theories. ESL theories are discussed in detail in the following section,
a brief account on the idea behind layerwise theories is given here.
As the name suggests, layerwise theories model each lamina, or layer, as
a plate (or shell) structural element, then impose interlaminar continuity conditions on the displacements and transverse stresses, thus joining the laminae
together. In particular, displacements are assumed to be C 0 -continuous in the
thickness, with discontinuous derivatives between the laminae, allowing for the
presence of discontinuous strains. On the other hand, transverse stresses are set
to be continuous between the laminae: this is possible because of the change
in constitutive properties, due to the change in material or in the lamina fibre
orientation. Partial layerwise theories account for continuous transverse shear
stresses across laminae, while full layerwise theories include also the continuity
of transverse normal stress. Depending on the assumptions, each layerwise theory yields a different approximation of a full three-dimensional theory, always
satisfying interlaminar continuity conditions and thus allowing to properly ac-
30
Modelling of fibre-reinforced composites
count for transverse shear and transverse normal effects. These are necessary
to accurately model the behaviour of thick laminates and to predict phenomena
occurring at the lamina level, such as delamination.
2.1.1
Equivalent Single Layer theories
Equivalent Single Layer theories model laminated plates as an equivalent single
lamina with complex constitutive properties. They are fully two-dimensional
models, which are derived by generalising the theories already developed for
single layer plates. As such, many ESL theories are present, each originated
from a different single layer plate theory and so making different choices as
regards the admissible displacement field.
Common to all ESL theories is the loss of interlaminar continuity of transverse stresses. Indeed, in these models the displacement field typical of a single
layer plate is postulated for a laminate, which is non homogeneous in the thickness direction. This discrepancy with respect to the three-dimensional description is a drawback of ESL theories with respect to layerwise models, however
it can be partially overcome by considering the ESL theory as descending from
a parent three-dimensional theory. Indeed, if a structural theory is postulated
directly, all information which is not contained in the assumed description is
lost. On the other hand, if the theory is established as descending from a parent three-dimensional theory some of this information can be reconstructed in
the post processing of the structural theory results.
Plate theories can be derived by considering a three-dimensional body with
special shape and partitionability (plate-like body) and setting some restrictions on the admissible deformations. Via the theory of constrained continua,
the three-dimensional compatibility, constitutive and equilibrium equations are
written based on the defined kinematic ansatz, then the equations for the twodimensional flat body (plate) can be derived in different ways (DiCarlo et al.,
2001; Teresi and Tiero, 1997). Later, once the two-dimensional problem is
solved, post processing considerations allow to reconstruct some of the quantities that have been lost in the three-dimensional to two-dimensional mapping.
In particular, crucial to the soundness of the procedure is the concept of
reactive stresses (Lembo and Podio-Guidugli, 2007). If the ansatz on the admissible displacements is regarded as internal constraints to the plate-like body,
2.1 Analytical modelling of laminated plates
d
n
b
p
z
(+)
O
x
zk
zk -1
31
W
¶W
th
k lamina
p
( )
Figure 2.1: Plate-like body
the three-dimensional stress field σ decomposes into an active and a reactive
part:
σ = σA + σR .
(2.1)
The active part σ A is related to the deformation via constitutive equations; the
reactive part σ R is given the role of maintaining the constraints, while doing no
work for each admissible deformation. Thus, reactive stresses can be regarded
as reactions to the constraints. The null-working condition can be stipulated
pointwise or partwise, with resulting integral conditions corresponding to the
peculiar partitionability in plate-like subdomains. It can be easily realised that,
when the three-dimensional plate-like body is mapped into the two-dimensional
plate model, only active stresses come into play. Thus, σ A is constitutively
determined by the solution of the plate problem, while σ R can be selected a
posteriori by enforcing the three-dimensional equilibrium equations, in order to
improve the approximation the three-dimensional stress field σ.
In the following, the outlined derivation is briefly carried out for some of
the most common plate theories, introducing a common notation. They include
the Classical Laminated Plate Theory (CLPT), which is the laminated plate
equivalent of Kirchhoff-Love single layer plate theory; the First-order Shear
Deformation Theory (FSDT), based on Reissner-Mindlin assumptions; a general
higher-order theory and the third-order theory proposed by Reddy (1984).
A plate-like body is a three-dimensional body whose reference shape is a
right cylinder of modest height h (Figure 2.1). Let (x1 , x2 , z) = (x, z) denote the
32
Modelling of fibre-reinforced composites
Material coordinate system (lamina)
b
z
c
x2
a
x1
Problem coordinate system (laminate)
Figure 2.2: Material and problem coordinate systems
coordinates of a point with respect to a Cartesian system with the z-axis parallel
to the generators of the cylinder. Notice that the problem coordinate system
(x1 , x2 , z) is in general different from the material coordinate system (a, b, c)
associated to each lamina. The latter is related to the material symmetry, as
depicted in Figure 2.2. The cylinder cross section at z = 0 is denoted by Ω,
with boundary ∂Ω, while the top and bottom surfaces at z = ±h/2 are denoted
by Ω(+) and Ω(−) , respectively. The typical material fibre is denoted by F and
its undeformed configuration is defined
by (x
´ ≤ z ≤ h/2). Body
´ =
³ const., −h/2
³
(−) (−)
(+) (+)
are prescribed inside
, px , pz
forces (bx , bz ) and surface loads px , pz
the body and on the top and bottom surfaces, respectively.
Classical Laminated Plate Theory
In the Classical Laminated Plate Theory, the following kinematic ansatz is made
on the admissible displacement field d:
∇2z d
= 0,
(2.2)
∇z d · k
= 0,
(2.3)
= −∇z dx ,
(2.4)
∇x d z
with k being the unit vector normal to Ω and ∇z and ∇x the derivatives with
respect to z and (x1 , x2 ) respectively. Condition (2.2) states that material fibres
F remain straight after the deformation, while (2.3) precludes fibre extension.
Finally, condition (2.4) imposes the orthogonality of the fibres to the deformed
middle surface of the plate.
2.1 Analytical modelling of laminated plates
33
Due to the constraints, d assumes the form:
·
d=
u (x) − z∇x w (x)
w (x)
¸
,
(2.5)
where u (x) and w (x) are the in-plane and transverse displacements of a point
belonging to the Ω. Note that the three-dimensional displacement field of the
plate-like body is fully defined once u and w are known. These quantities, called
the generalised displacements, are function only of the position of the fibre F,
thus the fibre is the minimum unit for the two-dimensional model. The same is
true in the other plate theories presented in the following.
The three-dimensional strain tensor ε can be derived from the compatibility
equations:
·
¸
1
e
γ
ε= 1 T 2
, ε = ∇(s) d,
(2.6)
γ
ε
z
2
where ∇(s) = sym grad is the compatibility operator, e is the in-plane strain
tensor, γ the transverse shear strain vector and εz the transverse normal strain.
These quantities can be expressed by introducing the generalised strains, which
are once again function only of the fibre position x. Introducing the expression
for the displacement field, Equation (2.5), into Equation (2.6) we obtain
e
=
(s)
∇(s)
x u − z∇x (∇x w) = µ + zχ,
(2.7)
γ
=
∇x w − ∇x w = 0,
(2.8)
εz
=
∇z w = 0,
(2.9)
where µ (x) and χ (x) are the membranal strains and the curvatures, respectively. The physical meaning of the generalised strains is clarified by Figure
2.3, which shows the deformation of a small element under the effect of each
generalised strain component.
Note that the transverse normal strain εz is null as required by condition
(2.3). Also, due to condition (2.4), transverse shear strains γ are null in the
CLPT. This is a major drawback of this theory, since shearing effects are negligible only for thin plates. The problem is emphasised when modelling laminated
composite plates, whose anisotropy renders shear effects even more significant
than in single layer isotropic plates.
34
Modelling of fibre-reinforced composites
Figure 2.3: Generalised strains for the CLPT
2.1 Analytical modelling of laminated plates
35
Analogously to the strain tensor, the three-dimensional stress tensor σ can
be separated into the in-plane stresses s, the transverse shear stresses τ and the
out of plane normal stress σz :
·
¸
s
τ
σ=
.
(2.10)
τ T σz
The three-dimensional equilibrium equations are not influenced by the kinematic
ansatz, thus they are common to the derivation of all theories. They do not
enter the derivation of the two-dimensional theory but they are crucial in the
definition of the reactive stress field. As such, they are not reported here but
recalled later in this section, where the post processing of the two-dimensional
solution is discussed.
The three-dimensional constitutive relations, relating the active stress to the
strain of the constrained plate-like body, are defined for each lamina, which is
assumed to be formed by a linearly elastic material having a general monoclinic
symmetry, with symmetry plane parallel to Ω. The active transverse stresses
are null:
τ A = 0, σzA = 0
(2.11)
and the constitutive equations for the k-th lamina are written as:
sA = C̃(k)
m e,
(2.12)
(k)
where C̃m collects the elastic coefficients of the constrained material, that are,
in principle, different from those of the unconstrained material. Using standard arguments of the theory of constrained continua, the constrained elasticity
tensor can be put in the common form:
(k)
C̃(k)
m = Cm ,
(k)
(2.13)
where Cm are the reduced in-plane elastic tensors of the (unconstrained) material of the k-th lamina. The use of the reduced in-plane elastic constants
is justified by deriving the constitutive equations via mixed variational methods (Teresi and Tiero, 1997); this derivation allows to reconcile the contrasting
assumptions of null strain and null stress in the thickness direction, which is
a typical inconsistency of plate theories. Note that only in-plane constitutive
equations are reported, as in CLPT transverse strains are null, thus τ = τ R and
36
Modelling of fibre-reinforced composites
σz = σzR and transverse stresses need to be reconstructed via three-dimensional
equilibrium in the post processing.
Once the three-dimensional equations for the constrained plate-like body
have been established, the three-dimensional body is mapped into the twodimensional plate by considering the fibre F as the minimum unit.
The fibre kinematics is described by the generalised displacements and strains,
as defined in Equations (2.5) and (2.7).
The two-dimensional equilibrium equations can be obtained by applying
the virtual work principle for all admissible displacements consistent with the
kinematic assumptions. Then integration by parts and localisation yields the
plate balance equations at both internal points
∇x · N + qx = 0, ∇x · (∇x · M) + ∇x · c + qz = 0,
(2.14)
and boundary points of the flat region Ω, together with the definition of the
two-dimensional stress measures and the two-dimensional load descriptors:
Z
Z
N=
sdz, M =
F
zsdz,
(2.15)
F
Z
qx =
Z
³
´
´
h ³ (+)
(−)
(−)
bx dz + p(+)
+
p
,
c
=
zb
dz
+
p
−
p
,
x
x
x
x
x
2
F
F
(2.16)
Z
qz =
F
+ p(−)
bz dz + p(+)
z .
z
(2.17)
The plate constitutive equations can be determined by substituting the constitutive relations for the constrained material in the form of Equation (2.12)
into the definition of the stress resultants:
N
= Cm µ + Cmb χ,
(2.18)
M
= Cmb µ + Cb χ,
(2.19)
2.1 Analytical modelling of laminated plates
37
where the laminate stiffness matrices are given by
Cm
=
n.layers
X
(zk − zk−1 ) C(k)
m ,
(2.20)
k
Cmb
Cb
=
=
1
2
1
3
n.layers
X
¡ 2
¢ (k)
2
zk − zk−1
Cm ,
(2.21)
¡ 3
¢ (k)
3
zk − zk−1
Cm .
(2.22)
k
n.layers
X
k
(k)
Notice that the constitutive matrices Cm for the k-th lamina need to be written
in the problem coordinates, thus they are different for laminae made of the same
material but with different orientation of the reinforcement. This feature causes
the typical coupling of the bending and membrane behaviour via the term Cmb ,
which is zero only if the laminate stacking sequence is symmetric.
As underlined earlier, once the plate problem is solved the active part of
the stress is constitutively determined. Then, three-dimensional equilibrium
equations can be used to evaluate the reactive part, as discussed later in this
section.
First-order Shear Deformation Theory
The First-order shear deformation theory partially overcomes the drawbacks of
the Classical Laminated Plate Theory by removing the condition of orthogonality of the fibre with respect to the plate middle surface. Thus, the kinematic
ansatz is the following:
∇2z d =
0,
(2.23)
∇z d · k =
0.
(2.24)
Since it includes transverse shear deformation, FSDT is more suitable to model
moderately thick plates, but yields layerwise constant transverse shear stresses
which violate interlaminar equilibrium and may not be acceptable when accurate
information on such quantities is required. As in the CLPT, the transverse
stresses approximation can be improved by introducing the reactive stress field,
as it will be discussed further in this section. Unlike in the Classical Theory,
38
Modelling of fibre-reinforced composites
however, the reconstructed transverse shear stresses can be used to improve
iteratively the solution of the two-dimensional problem.
For these reasons, the FSDT constitutes an ideal basis for the development of
numerical strategies for the structural analysis of laminated composite plates.
Within this context, a new finite element based on a hybrid stress approach
is formulated and tested in Section 2.2, while a numerical procedure for the
reconstruction of the transverse stresses is presented in Section 2.3.
The FSDT three-dimensional displacement field is the following:
·
¸
u (x) + zθ (x)
d=
.
(2.25)
w (x)
Observe that, with respect to the CLPT, the new generalised displacements θ
are introduced: they are the rotations of the fibre with respect to the undeformed
configuration.
Three-dimensional strains are expressed as follows:
e
=
(s)
∇(s)
x u + z∇x θ = µ + zχ,
(2.26)
γ
=
∇x w + θ,
(2.27)
εz
= 0.
(2.28)
Note that, as anticipated, the transverse shear strains are not zero, but they
are constant along the fibre. Thus, the generalised strains are µ, χ and γ.
Moreover, the curvature χ, though having the same role in the definition of
the in-plane strain tensor e, assumes a different expression with respect to the
CLPT. Figure 2.4 shows the deformation of a small element under the effect of
each generalised strain component of the FSDT.
The constitutive relations for each lamina include, besides the in-plane part
reported in Equation (2.12), also the transverse shear part:
τ A = κ ¯ C(k)
s γ,
(2.29)
where symbol ¯ denotes the product component by component and κ contains
the so called shear correction factors:
·
¸
κ11 κ12
κ=
.
(2.30)
κ12 κ22
2.1 Analytical modelling of laminated plates
Figure 2.4: Generalised strains for the FSDT
39
40
Modelling of fibre-reinforced composites
The matrix κ naturally arises from the derivation of the constitutive equations
via mixed variational methods (Teresi and Tiero, 1997), accounting for the difference between the transverse shear stresses as seen by the three-dimensional
and two-dimensional models. Here, κ is assumed to be constant along the plate
thickness. It should be remarked that the shear correction factors are not known
a priori and the usual values κ11 = κ22 = 5/6 and κ12 = 0 are correct only for
homogeneous plates. The shear correction factors are the key of the iterative
strategy which can be used in the FSDT to improve the two-dimensional solution. For a laminated plate, they depend upon the stacking sequence and
the solution of the problem itself. Once a tentative two-dimensional solution
is available, reconstruction of three-dimensional transverse shear stresses allows
to update κ and the new shear correction factors can be used to solve the
two-dimensional problem once again (Auricchio and Sacco, 1999a; Noor et al.,
1990). A procedure to evaluate the shear correction factors based on a mixed
variational method is discussed later in this section.
Three-dimensional to two-dimensional mapping occurs as in the CLPT. The
kinematics is described by the generalised displacements u, w and θ and the
generalised strains µ, χ and γ as defined for the FSDT.
The field two-dimensional equilibrium equations take the following form:
∇x · N + qx = 0, ∇x · M − S + c = 0, ∇x · S + qz = 0,
(2.31)
where symbols are as defined in Equations (2.15) to (2.17) and the transverse
shear stress resultant S is defined as
Z
S=
τ dz.
(2.32)
F
The in-plane and bending constitutive equations for the FSDT plate are the
same as those introduced in CLPT (Equations (2.18) and (2.19)), while the
transverse shear constitutive equation is given here:
S
= Cs γ,
(2.33)
where
Cs
= κ¯
n.layers
X
k
(zk − zk−1 ) C(k)
s .
(2.34)
2.1 Analytical modelling of laminated plates
41
For further convenience, the inverse relations are also introduced:
µ =
Fm N + Fmb M,
(2.35)
χ =
Fmb N + Fb M,
(2.36)
γ
Fs S.
(2.37)
=
As already anticipated, post processing of the two-dimensional solution and
evaluation of the reactive stresses assumes a particular significance in the FSDT,
as it allows to iteratively update the shear correction factors, thus improving
the two-dimensional plate solution. The post processing is discussed later in
this section.
Higher-order theories and Reddy’s third-order theory
A way to improve the transverse shear stresses obtained in the FSDT model
is to loosen the constraint on the in-plane part of the displacement. Thus, the
kinematic ansatz of the FSDT is replaced with the following:
∇nz d =
∇z d · k
0, n > 2,
(2.38)
= 0.
(2.39)
In this case, the fibres do not remain straight but deform according to a polynomial of order (n − 1). For this reason, such theories are known as higher-order
theories.
A well-known higher-order theory is the third-order theory proposed by
Reddy (1984), often referred to with the acronym HSDT (Higher-order Shear
Deformation Theory). It is derived from a general third-order theory (n = 4
in condition (2.38)) by introducing a further condition on the transverse shear
stresses, which are taken to be null at the top and bottom surfaces of the plate.
The development of a general higher-order theory and of the HSDT in particular
is shown in the following.
The displacement field d for a general higher-order theory is the following:
"
d=
u (x) +
Pn−1
m (m)
(x)
m=1 z θ
w (x)
#
,
(2.40)
42
Modelling of fibre-reinforced composites
while for Reddy’s HSDT it becomes
·
d=
u (x) + zθ (x) + cz 3 [θ (x) + ∇x w (x)]
w (x)
¸
.
(2.41)
Here, θ (m) are higher-order kinematic parameters, with θ (1) coinciding with the
¡
¢
rotations θ as defined in the FSDT, c = −4/ 3h2 , where h is the laminate
thickness. For a general higher-order theory, the loosened kinematic ansatz
increases the range of admissible displacements, thus more generalised displacement parameters are required to describe it. On the other hand, the conditions
on the transverse shear stresses imposed by Reddy’s HSDT allows to eliminate
some of these parameters, leaving only those already present in the FSDT. This
is obviously an advantage from the point of view of model complexity, however
the extra constraints have significant consequences on the independence of the
generalised strains, as discussed in the following.
For higher-order theories, higher-order curvatures χ(m) and generalised transverse shear strains γ (m) are introduced:
e
=
∇(s)
x u+
n−1
X
(m)
z m ∇(s)
=µ+
x θ
m=1
γ
=
∇x w +
n−1
X
n−1
X
z m χ(m) ,
(2.42)
m=1
mz
m−1 (m)
m=1
θ
=
n−1
X
z m−1 γ (m) .
(2.43)
m=1
Note that γ (1) and χ(1) coincide with the classical transverse shear strains and
curvatures, γ and χ, as defined in the FSDT. For Reddy’s HSDT, on the other
hand, we have:
h
i
(s)
3
(s)
e = ∇(s)
∇(s)
x u + z∇x θ + cz
x θ + ∇x (∇x w) =
= µ + zχ(1) + z 3 χ(3) ,
γ
2
= ∇x w + θ + 3cz (θ + ∇x w) = γ
(2.44)
(1)
2
+z γ
(3)
.
(2.45)
With respect to the FSDT, the third-order curvatures χ(3) and generalised
transverse shear strains γ (3) are introduced. However, taking a closer look
to the expressions above one realises that γ (1) and γ (3) are linearly dependent,
since they differ only because of the multiplicative constant 3c. Thus, transverse
2.1 Analytical modelling of laminated plates
43
shear strains are not made up of a linear and a higher order part, but the fibre
F is constrained to deform according to a third-order polynomial orthogonal to
the top and bottom surfaces (because of the condition on null transverse shear
stresses on the two surfaces, which reflects on the dual strain components). Also,
the higher-order curvatures χ(3) are linearly dependent on the derivatives of
γ (1) (as happens also in general higher-order theories). Thus, those introduced
are not new and independent parameters (as can be expected, since no new
generalised displacement was introduced with respect to the FSDT), but they
are deeply intertwined.
For simplicity, the number of generalised strains can be reduced by bringing
the constant c out of the definition of the generalised strains. The expressions
become:
e = µ + zχ + cz 3 χ(h) ,
¡
¢
γ = 1 + 3cz 2 γ,
(2.46)
(2.47)
where µ, χ and γ are the same as those defined in the FSDT, and higherorder curvatures χ(h) are introduced. Figure 2.5 shows the deformation of a
small element due to each generalised strain component for Reddy’s HSDT.
Note that, for what said before, in the elements depicting constant higher order
curvatures, linear transverse shear strains are also present.
Differently from the FSDT, the higher-order theories yield transverse shear
stresses which are not layerwise constant, thus it is generally thought that such
theories do not require shear correction factors. However, as pointed out by
Huang (1994), this is not the case for multilayered laminates. Indeed, as already pointed out, interlaminar continuity of transverse shear stresses is not
satisfied by any ESL theory, thus shear correction factors involving the difference between constitutively determined and post processed shear stresses can
always be defined. These considerations, though included for completeness, are
beyond the scope of this work, therefore higher-order theories are here presented
without the use of shear correction factors. Thus, the k-th lamina equations become:
A
sA = C(k)
= C(k)
(2.48)
m e, τ
s γ.
The two-dimensional kinematics is described by the generalised displacements and strains previously introduced. As already mentioned, for Reddy’s
44
Modelling of fibre-reinforced composites
Figure 2.5: Generalised strains for the HSDT
2.1 Analytical modelling of laminated plates
45
HSDT they are the same ones as for the FSDT, even though the resulting
three-dimensional displacement field is different. For higher-order theories, more
parameters are introduced.
Two-dimensional field equilibrium equations are as follows for a general
higher-order theory:
∇x · N + qx = 0, ∇x · S(1) + qz = 0,
(2.49)
∇x · M(m) − S(m) + c(m) = 0 with 1 ≤ m < n,
(2.50)
where higher-order generalised stresses and load descriptors are introduced as
the dual quantities of higher order generalised strains:
Z
Z
M(m) =
z m sdz, S(m) =
z m−1 τ dz,
(2.51)
F
F
µ ¶m h
i
h
m
=
z m bx dz +
p(+)
+ (−1) p(−)
.
x
x
2
F
Z
c(m)
(2.52)
For Reddy’s HSDT:
∇x · N + qx = 0, ∇x · M + ∇x · M(h) − S + c + c(h) = 0,
³
´
−∇x · ∇x · M(h) + ∇x · S + ∇x · c(h) + qz = 0,
where
Z
Z
M(h) =
cz 3 sdz, S =
Z
c(h) =
F
µ
cz 3 bx dz + c
F
h
2
F
¶3
(2.53)
(2.54)
¡
¢
1 + 3cz 2 τ dz,
(2.55)
´
(−)
−
p
.
p(+)
x
x
(2.56)
³
In the same way, two-dimensional constitutive equations contain higherorder terms, whose membrane and bending parts are still coupled:
N
= Cm µ +
n−1
X
(m)
Cmb χ(m) ,
(2.57)
m=1
M(p)
(p)
= Cmb µ +
n−1
X
(p,m)
Cb
χ(m) ,
(2.58)
m=1
S(p)
=
n−1
X
m=1
C(p,m)
γ (m) ,
s
(2.59)
46
Modelling of fibre-reinforced composites
where the laminate stiffness matrices are given by
n.layers
X Z zk
(m)
Cmb =
z m C(k)
m ,
(p,m)
Cb
=
n.layers
X Z zk
zk−1
k
C(p,m)
s
=
(2.60)
zk−1
k
n.layers
X Z zk
zk−1
k
z m+p C(k)
m ,
(2.61)
z m+p−2 C(k)
s .
(2.62)
For Reddy’s HSDT:
N
(h)
= Cm µ + Cmb χ + Cmb χ(h) ,
M
=
(h)
=
S
=
M
(h)
Cmb µ + Cb χ + Cbb χ(h) ,
(h)
(h)
(h)
Cmb µ + Cbb χ + Cb χ(h) ,
C(h)
s γ,
(2.63)
(2.64)
(2.65)
(2.66)
where
(h)
Cmb
=
n.layers
X Z zk
k
(h)
Cbb
=
n.layers
X Z zk
k
(h)
Cb
=
=
zk−1
n.layers
X Z zk
k
C(h)
s
zk−1
zk−1
n.layers
X Z zk
k
zk−1
cz 3 C(k)
m ,
(2.67)
cz 4 C(k)
m ,
(2.68)
c2 z 6 C(k)
m ,
(2.69)
¡
¢2
1 + 3cz 2 C(k)
s .
(2.70)
As in the previous cases, post processing allows to reconstruct reactive
stresses from three-dimensional equilibrium. In particular, for a n-th order theory, n-th order in-plane stresses yield (n + 1)-th order transverse shear stresses.
Transverse refinement
The higher-order theories described in the previous paragraph introduce a polynomial refinement of the in-plane displacement in order to enrich the FSDT
2.1 Analytical modelling of laminated plates
47
kinematic hypothesis. A different possibility is to improve the transverse displacement by relaxing the kinematic ansatz on the fibre extension, that is to
say
∇z d · k 6= 0.
(2.71)
The in-plane part of the displacement field d can be chosen according to the
FSDT or any of the higher-order theories previously described, while the transverse part becomes
n
X
dz = w (x) +
z m w(m) (x) .
(2.72)
m=1
The full derivation of the two-dimensional theory proceeds as usual and it is
not reported here. Instead, a few comments are given on the consequences of
introducing a transverse refinement.
Differently from in-plane refinement, indeed, transverse refinement leads to
non zero transverse normal strain:
εz =
n
X
mz m−1 w(m) =
m=1
n
X
z m−1 ε(m)
z .
(2.73)
m=1
As a consequence, constitutive equations for the transverse normal stress and
strain components need to be introduced. Moreover, work-conjugate stress resultants involving the transverse normal stress σz arise, defined as
Z
(m)
Sz =
z m−1 σz dz.
(2.74)
F
From the previous equation, it is clear that this type of refinement allows the
transverse normal stress σz to play an active role in the two-dimensional plate
model. Thus, differently from all the models discussed previously, we have
σzA 6= 0. Still, the two-dimensional stress representation could be improved in
the post processing by introducing the reactive stresses.
Reconstruction of the transverse stresses
The decomposition of the three-dimensional stress field σ into an active and a
reactive part can be used to improve the approximation of the stress field, once
the solution of the two-dimensional problem has been obtained. Indeed, while
48
Modelling of fibre-reinforced composites
the active stresses σ A are constitutively determined by the two-dimensional solution, the reactive stresses σ R can be selected by enforcing the three-dimensional
equilibrium equations. This reconstruction procedure can be carried out for
every ESL theory, however it gains a particular significance in the FSDT theory, as the reconstructed stresses can be used to iteratively evaluate the shear
correction factors.
The local three-dimensional equilibrium conditions are given by
divσ + b = 0,
(2.75)
or by the following extended form:
∇x · s + ∇z τ + bx = 0,
(2.76)
∇x · τ + ∇z σz + bz = 0.
(2.77)
By assuming the in-plane stresses s as completely constitutively determined
(s = sA , sR = 0), Equation (2.76) can be used to determine the transverse shear
stresses:
Z z
−
τ (z) = −p(−)
x
−h
2
(∇x · s + bx ) dz,
then Equation (2.77) allows to evaluate the transverse normal stress:
Z z
σz (z) = −p(−)
−
(∇x · τ + bz ) dz.
z
(2.78)
(2.79)
−h
2
Notice that the above solutions have been given directly in terms of total stresses
(sum of active and reactive parts), as is usual for the technical literature.
As it was already anticipated, the outlined reconstruction procedure can be
applied to any ESL theory. In CLPT, transverse deformations are null due to
the kinematic hypothesis introduced, thus the transverse stresses consist of the
sole reactive part:
τ CLPT = τ R , σzCLPT = σzR .
(2.80)
In this case, the reconstruction procedure can be used to obtain a better approximation of the three-dimensional stress field but it does not influence the
two-dimensional plate solution. In FSDT, on the other hand, the transverse
shear term appears in the two-dimensional equations, thus
τ FSDT = τ A + τ R , σzFSDT = σzR .
(2.81)
2.1 Analytical modelling of laminated plates
49
The reconstructed transverse shear stress profiles can be used to iteratively evaluate the shear correction factors (2.30), allowing to improve the two-dimensional
plate solution.
In order to ensure consistency with both the three-dimensional and twodimensional models, the reconstructed transverse stresses should satisfy the
boundary conditions on the top and bottom surfaces of the plate-like body:
µ
¶
µ ¶
h
h
τ −
= −p(−)
,
τ
= p(+)
(2.82)
x
x ,
2
2
¶
µ
µ ¶
h
h
= −p(−)
= p(+)
σz −
,
σ
(2.83)
z
z
z .
2
2
Moreover, in FSDT, the reconstructed transverse shear stresses should satisfy
the static equivalence condition, Equation (2.32). It is proven in the following
that, if the stress resultants N, M and S satisfy the plate equilibrium equations,
Equation (2.31), then the transverse stresses as determined by Equations (2.78)
and (2.79) meet all the aforementioned conditions.
Proof. The first of conditions (2.82) is trivially met. To verify the second, one
can simply evaluate τ at the top surface:
µ ¶
h
τ
=
2
=
Z
−p(−)
−
x
h
2
−h
2
−p(−)
− ∇x ·
x
(∇x · s + bx ) dz =
Z
Z
sdz −
F
bx dz.
(2.84)
F
Using Equations (2.15) and (2.16) yields
µ ¶
h
τ
= −∇x · N − qx + p(+)
x .
2
(2.85)
Thus, if membrane forces satisfy pointwise the plate equilibrium conditions
(2.31), the second of conditions (2.82) is met.
To prove the static equivalence condition (2.32), we can multiply the threedimensional equilibrium condition (2.76) by z and integrate it along the thickness:
Z
Z
Z
z∇x · sdz +
z∇z τ dz +
zbx dz = 0.
(2.86)
F
F
F
50
Modelling of fibre-reinforced composites
Integrating by parts the second term and recalling Equations (2.15) and (2.16)
lead to
· µ ¶
µ
¶¸
Z
´
h
h
h ³ (+)
h
τ dz +
∇x · M −
τ
+τ −
+c−
px − p(−)
= 0. (2.87)
x
2
2
2
2
F
Therefore, since τ satisfies the equilibrium conditions on the top and bottom
surfaces, if the internal moments M satisfy the plate equilibrium conditions
(2.31), then it follows
Z
S−
τ dz = 0.
(2.88)
F
As regards the boundary conditions for σz , while the first of conditions (2.83)
is again trivial, the second can be verified by evaluating σz at the top laminate
surface:
µ ¶
Z h2
h
(−)
(∇x · τ + bz ) dz =
σz
= −pz −
2
−h
2
Z
Z
= −p(−)
−
∇
·
τ
dz
−
bz dz.
(2.89)
x
z
F
Using Equations (2.32) and (2.17) yields
µ ¶
h
σz
= −∇x · S − qz + p(+)
z .
2
F
(2.90)
Thus, if the third of Equations (2.31) is pointwise satisfied, the boundary condition for σz on the top laminate surface is met.
Once the transverse shear stress profiles are known, shear correction factors can be determined by enforcing the stationarity of the Hellinger-Reissner
functional with respect to N, M and S (Teresi and Tiero, 1997). Focusing the
attention on the transverse shear term per unit area, we have:
Z ³
´
F(k)
τ
−
γ
δτ dz,
(2.91)
s
F
where τ is to be intended as the total three-dimensional transverse shear stress.
This stress term can be represented in the general form
τ = gh (z) ¯ S (x) + gp (z, x) ,
(2.92)
2.1 Analytical modelling of laminated plates
51
where gh collects piecewise shape functions defining the shear stress profiles
in the thickness and vanishing on the upper and lower faces and gp collects
null average functions on the thickness, accounting for the actual equilibrium
conditions on the two faces.
Substituting Equation (2.92) into Equation (2.91) and making it stationary
with respect to S yields
Fs S − γ + γ0 = 0,
(2.93)
which defines the transverse shear flexibility response of the laminate, where
Fs
=
γ0
=
Z
C−1
=
(gh ⊗ gh ) ¯ F(k)
s
s dz,
F
Z ³
´
F(k)
g
¯ gh dz,
p
s
(2.94)
(2.95)
F
being ⊗ the dyadic product. Notice that, as observed by Teresi and Tiero
(1997), the resulting configuration of the plate is not natural, since γ0 is in
general different from zero, although the parent plate-like body has been supposed in a natural configuration. However, in practice, this term is generally not
considered, since gp in Equation (2.92) is set to zero according to the classical
FSDT stress hypothesis, stating that transverse shear stress components vanish
on the upper and lower faces. This is simply a different manner to mimic the
three-dimensional behaviour, which does not imply any variational crime since
the above procedure is of mixed type and Equation (2.92) is not required to
necessarily satisfy all the equilibrium conditions. A discussion on the influence
of non homogeneous conditions can be found in the paper by Carrera (2007).
Considering Equations (2.29) and (2.94) permits to compute the shear correction factors, once τ has been reconstructed starting from the plate solution.
Unfortunately, as already noted, the solution itself depends on the shear factors,
so that their computation results in a non linear problem. A successful strategy
to tackle this problem is to select some tentative values for the shear factors,
for example those for homogeneous plates, and use them to start an iterative
predictor-corrector procedure, as suggested by Auricchio and Sacco (1999a) and
Noor et al. (1990).
52
2.2
Modelling of fibre-reinforced composites
A hybrid-stress finite element for laminated
plates
The Equivalent Single Layer theories described in the previous section are the
most common tool for the analysis of laminated composite plates. In particular, the FSDT is often favoured as a simple yet accurate description of the
three-dimensional plate-like body behaviour, especially when used in conjunction with the transverse stresses reconstruction and iterative evaluation of the
shear correction factors. For these reasons, the FSDT is an ideal basis for the
development of numerical solution strategies, such as the finite element method.
Several laminated plate finite elements were developed in the last decades
(see among others Alfano et al. (2001); Auricchio et al. (2001); Auricchio and
Sacco (1999a, 2003); Auricchio et al. (1999, 2006); Cazzani et al. (2005); Cen
et al. (2002); Kim and Cho (2007)) based on FSDT. The basic features of a
competitive element are: simplicity, computational efficiency and accuracy. In
particular, the element should not suffer from shear locking or other pathologies,
such as spurious kinematic modes or parasitic shear stresses. Moreover, it should
be accompanied by an effective procedure to reconstruct the three-dimensional
transverse stresses through the laminate thickness. Based on these desirable features and starting from the element earlier formulated by Auricchio and Taylor
(1994) for single layer isotropic plates, Auricchio and Sacco (1999a) developed a
four node mixed-enhanced laminated plate finite element, recently extended to
treat monoclinic laminae (Auricchio et al., 2006). Alfano et al. (2001), on the
other hand, generalised the well known MITC finite element (Bathe, 1996) to
laminated composite plates. Both the approaches were proposed in conjunction
with a procedure to update the shear correction factors and to reconstruct the
transverse stress profiles, based on three-dimensional equilibrium. In the following, a new finite element for laminated plates is proposed based on a hybrid
stress approach, while a numerical strategy for the post processing of the finite
element solution is discussed in Section 2.3.
A new quadrilateral four node finite element for laminated plates with general monoclinic laminae is developed in this section, based on the FSDT as
derived in Section 2.1.1 (Daghia et al., 2008). A hybrid stress formulation is
adopted by generalising the one presented by de Miranda and Ubertini (2006)
for single layer isotropic plates. In particular, the formulation is rewritten to
2.2 A hybrid-stress finite element for laminated plates
53
include the membrane-bending coupling typical of laminated plates. The new
element is designed to be simple, free of locking and readily implementable into
existing finite element codes. To this purpose, the transverse displacement approximation is enhanced via the so called linked interpolation, which includes
nodal rotation parameters through higher-order interpolation functions. Moreover, an approximation which is a priori equilibrated within each element is
assumed for both membrane and bending-shear stress resultants. It is ruled
by the minimum number of parameters for element stability, which are eliminated at the element level. The resulting element has four nodes, five degrees
of freedom per node and involves only compatible displacement functions. It is
locking-free, passes all the patch tests and exhibits little sensitivity to geometric
distortions. It should be noted that the equilibrated stress assumption, besides
being a key point of the element formulation, acquires a special importance
in view of the reconstruction of the transverse stress profiles, as discussed in
Section 2.3.
Note that, in the following, tensorial notation is abandoned and matrix notation is used instead, as is common for finite element derivation.
2.2.1
Hybrid stress formulation
The mixed variational formulation of the FSDT stems from the stationarity
condition for the Hellinger-Reissner functional:
Z
£ T
¤
1
ΠHR = −
N (Fm N + Fmb M) + MT (Fmb N + Fb M) + ST Fs S dΩ +
2 Ω
Z
£ T
¤
+
N Dm u + MT Dm θ + ST (Ds w + θ) dΩ − Πext ,
(2.96)
Ω
where Πext is the work done by external loads, Dm and Ds are differential
operators. Assuming that N, M and S satisfy a priori the plate equilibrium
equations, the mixed functional ΠHR reduces to the following hybrid stress
functional:
Z
£ T
¤
1
ΠHY = −
N (Fm N + Fmb M) + MT (Fmb N + Fb M) + ST Fs S dΩ +
2 Ω
Z
¡ T T
¢
T
+
(2.97)
u nm N + θ T nT
m M + wns S d (∂Ω) − Π̂ext ,
∂Ω
54
Modelling of fibre-reinforced composites
4
h2
3
4
h2
x2
h1
3
x1
h1
2
1
1
2
Figure 2.6: Coordinate systems for the plate element
where Π̂ext is the work done by external boundary loads, nm and ns contain
the components of the outward normal to the boundary ∂Ω:

nm
n1
= 0
n2

·
¸
0
n1
.
n2  , ns =
n2
n1
(2.98)
The hybrid stress functional ΠHY , involving both stress resultants and generalised displacements as independent variables, is the variational support for
the subsequent development of the finite element scheme. Notice that in the discretised version of functional (2.97): (i ) stress resultants are required to satisfy
the plate equilibrium equations within each element Ωe and can be discontinuous across element boundaries ∂Ωe , (ii ) generalised displacements are active
only on the element boundaries ∂Ωe and are required to be continuous across
them.
2.2.2
Finite element assumptions
Geometry description
The element geometry is represented based on a four node scheme. The transformation between the local Cartesian coordinates (x1 , x2 ) and the natural coordinates (η1 , η2 ) (see Figure 2.6) is given by
xi = a1i η1 + a2i η1 η2 + a3i η2 ,
(2.99)
2.2 A hybrid-stress finite element for laminated plates
where



a1i
−1
 a2i  = 1  1
4
−1
a3i
1 1
−1 1
−1 1

 xi1
−1 
xi2
−1  
 xi3
1
xi4
55


,

(2.100)
being xij the local coordinates of the jth corner.
For later convenience, a skew coordinate system (η̄1 , η̄2 ) is introduced as
follows:
j2
j1
(2.101)
η̄1 = η1 + η1 η2 , η̄2 = η2 + η1 η2 ,
j0
j0
where
j0 = a11 a32 − a12 a31 , j1 = a11 a22 − a12 a21 , j2 = a21 a32 − a22 a31 .
(2.102)
Assumed generalised displacements
In order to avoid locking, the displacement interpolation for w needs to be
enriched with respect to that used for θ. For this reason, the interpolation for u
and θ is simply bilinear, ruled by the standard nodal degrees of freedom qu and
qθ , while the interpolation for w, bilinear in the transverse nodal displacements
qw , is enriched with higher order functions ruled by the nodal rotations qθ
(linked interpolation):
u = Uu qu , θ = Uθ qθ , w = Uw qw + Lqθ .
(2.103)
This allows to improve the interpolation for w without introducing additional
kinematic degrees of freedom.
Matrices U• collect the standard four node shape functions:
Ui =
1
(1 + η1i η1 ) (1 + η2i η2 ) ,
4
(2.104)
being η1i and η2i the natural coordinates at node i.
The higher order part is defined as follows:
Lqθ = −
4
X
i=1
¡ 2
¢
1
L̂i li θni
− θni
,
(2.105)
56
Modelling of fibre-reinforced composites
1
2
where li is the length of side i, θni
and θni
are the components of the rotation
in the direction normal to side i at its two ends, respectively, and L̂i are edge
bubble functions defined as
1
[1 + (1 − δi4 ) η1 ] [1 − (1 − δi2 ) η1 ] [1 + (1 − δi1 ) η2 ] [1 − (1 − δi3 ) η2 ] ,
16
(2.106)
where δij is the Kronecker delta.
L̂i =
Note that the above interpolation leads to constant transverse shear strains
along each side of the element. However, this is generally not sufficient to remove
locking due to interior inconsistency and shear strains need to be accommodated
in the element interior using, for example, internal bubble functions (Auricchio
and Sacco, 1999a; Auricchio et al., 2006). In the present formulation these
additional degrees of freedom are not necessary since the displacements actually
play their role only on the element boundary.
Assumed stress resultants
The stress resultants are approximated independently on each element and such
to satisfy the plate equilibrium equations pointwise within the element. Using
the skew coordinates η̄1 , η̄2 , the resultant moments M and shear forces S are
assumed as suggested by de Miranda and Ubertini (2006) for the 9βQ4 element:

1 0 0
a211 η̄2

·
¸
a212 η̄2
 0 1 0
M

=  0 0 1 a11 a12 η̄2

S
 0 0 0
0
0 0 0
0
a11 η̄1 a31 η̄2
0
0
1
0
a231 η̄1
a232 η̄1
a31 a32 η̄1
0
0
0
a12 η̄1 a32 η̄2
0
0
1
a211 η̄1 η̄2
a212 η̄1 η̄2
a11 a12 η̄1 η̄2
a11 η̄2
a12 η̄2
a231 η̄1 η̄2
a232 η̄1 η̄2
a31 a32 η̄1 η̄2
a31 η̄1
a32 η̄1

·
¸

 βb
,

 βs

(2.107)
2.2 A hybrid-stress finite element for laminated plates
57
and the membrane forces N as suggested by Yuan et al. (1993):

1 0 0
a211 η̄2
N= 0 1 0
a212 η̄2
0 0 1 a11 a12 η̄2

a231 η̄1
a232 η̄1  βm ,
a31 a32 η̄1
(2.108)
where parameters βm , βb and βs are local to each element. The above approximations are proved to satisfy pointwise the plate equilibrium equations in the
homogeneous form. According to de Miranda and Ubertini (2006), the presence
of field loads can be taken into account by a particular solution (Np , Mp , Sp ,)
of the plate equilibrium equations. Therefore, the final approximation can be
written in compact form as
N = Pm βm + Np , M = Pb βb + Pbs βs + Mp , S = Ps βs + Sp .
(2.109)
T
In the case of general loading (qT
x = [q1 , q2 ], c = [c1 , c2 ], qz ), the particular
solution (Np , Mp , Sp ) can be chosen as
 Z x1

q1 dx1 
 −
 Z0 x2


,
Np = 
(2.110)

−
q
dx
2
2


0
0
 Z


Mp = 



 Z x1

(S1p − c1 ) dx1 
q
dx
z
1
0
Z x2


 , Sp = − 1 
 Z0 x2
.
(S2p − c2 ) dx2 
2

qz dx2
0
0
0
x1
(2.111)
This expression has been adopted for the numerical tests.
It should be emphasised that the chosen approximation is optimal in the
sense that it involves the minimum number of β parameters while providing
element stability, i.e. without introducing any zero energy mode.
2.2.3
Finite element equations
Introducing the above assumptions in the hybrid stress functional (2.97) referred
to the single element and making it stationary yield the following discrete ele-
58
Modelling of fibre-reinforced composites
ment equations:
where

Hm
m
 ¡ mb ¢T
H =  Hsm
¡ mb ¢T
Hbm
"
H
GT
Hmb
sm
Hbs + Hss
¡ b ¢T
Hbs
G
0
#"
β
q
#
"
=
g
h
#

 u
Hmb
Gm
bm

Hbbs  , G =  0
0
Hbb
,
(2.112)
0
Gw
s
0

0
,
Gθbs + Gwθ
s
θ
Gb
(2.113)







hu
qu
gm
βm







, h = hw  .
, q = qw , g =
β=
gs
βs
hθ
gb
qθ
βb

(2.114)
Here, g are the terms due to the prescribed loads and h are the element nodal
generalised forces. Observe that the equations are strongly coupled due to the
coupling in the stress resultants approximation, the linking on the displacement
interpolation and the membrane-bending coupled behaviour of laminated plates.
The matrices involved are defined as follows:
Z
Z
T
mb
Hm
=
P
F
P
dΩ,
H
=
PT
(2.115)
m
m
m
m
sm
m Fmb Pbs dΩ,
Ω
Ω
Z
Z
Hmb
PT
Hss =
PT
(2.116)
bm =
m Fmb Pb dΩ,
s Fs Ps dΩ,
Ω
Ω
Z
Z
Hbs =
PT
Hbbs =
PT
(2.117)
bs Fb Pbs dΩ,
bs Fb Pb dΩ,
Ω
Ω
Z
Z
u T
T
Hbb =
PT
F
P
dΩ,
(G
)
=
UT
(2.118)
b
b
b
m
u nm Pm d (∂Ω) ,
Ω
∂Ω
Z
Z
¡ θ ¢T
T
T
T
(Gw
UT
Gbs =
UT
(2.119)
s) =
w ns Ps d (∂Ω) ,
θ nm Pbs d (∂Ω) ,
∂Ω
∂Ω
Z
Z
¡ wθ ¢T
¡ θ ¢T
T
Gs
=
LT n T
Gb =
UT
(2.120)
s Ps d (∂Ω) ,
θ nm Pb d (∂Ω) .
∂Ω
∂Ω
The inner parameters β can be condensed out at the element level and the
elemental equations take the standard form involving only nodal displacements:

 

 

hu
Kuu Kuw Kuθ
qu
fu
 hw  =  KT
(2.121)
Kww Kwθ   qw  −  fw  .
uw
T
T
hθ
Kuθ Kwθ Kθθ
qθ
fθ
2.2 A hybrid-stress finite element for laminated plates
59
x2
u1=w=q1=0
L/2
D
u2=w=q2=0
C
A
B
L/2
x1
Figure 2.7: Simply supported square plate, regular and distorted mesh patterns
As it can be immediately realised, this relation can be easily implemented into
existing finite element codes. Notice that, with respect to the mixed-enhanced
finite element proposed by Auricchio and Sacco (1999a) and Auricchio et al.
(2006) and based on linked interpolation for the transverse displacement, the
present element does not require internal bubble functions and enhanced incompatible modes, whose number depends on the material response (5+8×3
for monoclinic layers). The greater simplicity reflects also on a higher computational efficiency since only 14 inner parameters should be eliminated at the
element level.
2.2.4
Numerical testing of the finite element performance
In this section, some numerical tests are carried out in order to investigate the
performance of the proposed laminate finite element (HQ4). The same test case
is later used in Section 2.3 to test the transverse shear stress reconstruction
procedure.
A simply supported square plate of side L (Figure 2.7) is considered under
both uniform and sinusoidal load, with maximum intensity q. The side to
60
Modelling of fibre-reinforced composites
0
-1
-1.5
-2
1
-2.5
-1
0/90 - sinusoidal load
-0.5
Log(error %)
Log(error %)
-0.5
-3
0
HQ4, regular 0/90/0 - sinusoidal load
HQ4, distorted
-1
-1.5
-2
-2.5
-0.5
0
Log(L/N)
0.5
-3
1
-1
-0.5
0
Log(L/N)
0.5
Figure 2.8: Convergence in energy norm for sinusoidal load and both the stacking sequences
thickness ratio of the laminate is L/h = 10. Two different stacking sequences
are analysed: a symmetric (0/90/0) laminate and an antisymmetric (0/90) one.
Due to the in-plane double symmetry, only one quarter of the plate is studied.
The regular and distorted mesh patterns used in the analyses are shown in
Figure 2.7.
The lamina mechanical properties in the material coordinates are the following:
Ea = 25 Eb , νab = 0.25,
Gac = Gab = 0.5 Eb , Gbc = 0.2 Eb .
(2.122)
The shear correction factors are computed by assuming cylindrical bending
(Laitinen et al., 1995) and are considered as constant (that is they are not
updated using the recovered shear stress profiles): κ11 = 235445/404004, κ22 =
289/360, κ12 = 0 for the (0/90/0) laminate, κ11 = 297680/362481, κ22 = κ11
and κ12 = 0 for the (0/90) laminate.
The reference solution are computed according to Reddy (1997).
Global and local convergence properties are discussed for both the load cases
and both the stacking sequences, using regular and distorted meshes. In the
uniform load case, the results are compared with those predicted by the MITC
laminated plate element, implemented in the commercial software ADINA.
Figure 2.8 shows the convergence in energy norm in the case of sinusoidal load
2.2 A hybrid-stress finite element for laminated plates
-1
0/90/0 - sinusoidal load
-2
Log(error %)
Log(error %)
-3
-4
-5
-1
HQ4, regular
0/90/0 - uniform load
ADINA, regular
HQ4, distorted
ADINA, distorted
-2
-3
-4
2
-1
2
-0.5
0
Log(L/N)
-5
0.5
-1
-0.5
0
Log(L/N)
0/90 - uniform load
0/90 - sinusoidal load
-2
Log(error %)
-2
Log(error %)
0.5
-1
-1
-3
2
-4
-5
61
-1
-0.5
0
Log(L/N)
-3
-4
-5
0.5
2
-1
-0.5
0
Log(L/N)
0.5
Figure 2.9: Convergence of the transverse displacement at point A for different
loads and stacking sequences
-1
0/90 - sinusoidal load
-2
Log(error %)
-2
Log(error %)
-1
HQ4, regular
0/90 - uniform load
ADINA, regular
HQ4, distorted
ADINA, distorted
-3
-4
-3
-4
2
2
-5
-1
-0.5
0
Log(L/N)
0.5
-5
-1
-0.5
0
Log(L/N)
0.5
Figure 2.10: Convergence of the in-plane displacement at point B for the (0/90)
stacking sequence and both uniform and sinusoidal load
62
Modelling of fibre-reinforced composites
0
-2
-3
-4
-6
-1
-3
-4
-5
-0.5
0
Log(L/N)
-6
0.5
0
-1
0.5
M2
-1
Log(error %)
-2
-3
-4
-2
-3
-4
2
2
-5
-5
-6
-0.5
0
Log(L/N)
0
M1
-1
Log(error %)
-2
2
2
-5
S2
-1
Log(error %)
Log(error %)
-1
0
S1
HQ4, regular
ADINA, regular
HQ4, distorted
ADINA, distorted
-1
-0.5
0
Log(L/N)
0
-6
0.5
-1
-0.5
0
Log(L/N)
0.5
M12
Log(error %)
-1
-2
-3
-4
2
-5
-6
-1
-0.5
0
Log(L/N)
0.5
Figure 2.11: Convergence of the stress resultants at point C for the (0/90/0)
stacking sequence and uniform load
2.2 A hybrid-stress finite element for laminated plates
0
2
HQ4, regular
ADINA, regular
HQ4, distorted
ADINA, distorted
N1 and N2
-1
Log(error %)
Log(error %)
1
0
-1
-2
-3
63
S1 and S2
-2
-3
-4
-5
2
2
-4
-1
-0.5
0
Log(L/N)
-6
0.5
0
-1
-0.5
0
Log(L/N)
0
M1 and M2
M12
-1
-2
Log(error %)
Log(error %)
-1
-3
-4
-5
-6
0.5
-0.5
0
Log(L/N)
-3
-4
2
-5
2
-1
-2
0.5
-6
-1
-0.5
0
Log(L/N)
0.5
Figure 2.12: Convergence of the stress resultants at point C for the (0/90)
stacking sequence and uniform load
Table 2.1: Transverse displacement w̄ at point A for simply supported (0/90/0)
laminate subjected to uniform load
Ref. sol. (Reddy, 1997)
mesh
HQ4, regular
HQ4, distorted
ADINA, regular
ADINA, distorted
3×3
1.17482
1.17610
1.17282
1.15553
1.16754
9×9
1.16831
1.16853
1.16808
1.16620
27×27
1.16763
1.16765
1.16760
1.16739
64
Modelling of fibre-reinforced composites
Table 2.2: Transverse displacement w̄ at point A and in-plane displacement ū1
at point B for simply supported (0/90) laminate subjected to uniform load
Ref. sol. (Reddy, 1997)
mesh
HQ4, regular
HQ4, distorted
ADINA, regular
ADINA, distorted
3×3
1.97191
1.96109
1.92668
1.88549
w̄
1.95058
9×9
1.95293
1.95227
1.94807
1.94390
27×27
1.95084
1.95077
1.95030
1.94983
3×3
0.13285
0.13147
0.12836
0.12683
ū1
0.12940
9×9
0.12978
0.12968
0.12929
0.12915
27×27
0.12944
0.12943
0.12939
0.12937
Table 2.3: Stress resultants at point C for simply supported (0/90/0) laminate
subjected to uniform load
Ref. sol. (Reddy, 1997)
mesh
HQ4, regular
HQ4, distorted
ADINA, regular
ADINA, distorted
Ref. sol. (Reddy, 1997)
mesh
HQ4, regular
HQ4, distorted
ADINA, regular
ADINA, distorted
3×3
-18.7782
-18.7426
-18.3930
-19.0768
S̄1
-18.9515
9×9
-18.9311
-18.9063
-18.8925
-18.8650
3×3
6.92084
6.96898
6.59260
6.44039
M̄1
7.01608
9×9
7.00641
7.01050
6.96909
6.92332
Ref. sol. (Reddy, 1997)
HQ4, regular
HQ4, distorted
ADINA, regular
ADINA, distorted
3×3
-0.31562
-0.28380
-0.31968
-0.24488
27×27
-18.9492
-18.9486
-18.9450
-18.9346
27×27
7.01501
7.01388
7.01086
7.00380
3×3
-1.23522
-0.99111
-1.16665
-1.68125
S̄2
-1.12587
9×9
-1.13633
-1.10086
-1.12802
-1.21503
27×27
-1.12704
-1.12529
-1.12610
-1.13727
3×3
1.16411
1.13975
1.11246
1.15971
M̄2
1.14959
9×9
1.15150
1.14901
1.14565
1.15100
27×27
1.14981
1.14953
1.14916
1.14976
M̄12
-0.31692
9×9
-0.31661
-0.31456
-0.31742
-0.30914
27×27
-0.31689
-0.31671
-0.31698
-0.31610
2.2 A hybrid-stress finite element for laminated plates
65
Table 2.4: Stress resultants at point C for simply supported (0/90) laminate
subjected to uniform load
Ref. sol. (Reddy, 1997)
mesh
HQ4, regular
HQ4, distorted
ADINA, regular
ADINA, distorted
Ref. sol. (Reddy, 1997)
mesh
HQ4, regular
HQ4, distorted
ADINA, regular
ADINA, distorted
3×3
-.03235
.01035
-.07579
-.17646
N̄1 = N̄2
-0.01910
9×9
-.02001
-.01967
-.02437
-.05685
3×3
3.65353
3.56463
3.49680
3.38096
M̄1 = M̄2
3.67599
9×9
3.67387
3.66430
3.65598
3.63580
27×27
-.01920
-.01774
-.01968
-.02435
27×27
3.67575
3.67427
3.67377
3.67138
3×3
-9.73822
-9.13132
-9.59597
-10.20210
S̄1 = S̄2
-9.71082
9×9
-9.70950
-9.62003
-9.69853
-9.67910
27×27
-9.71064
-9.70547
-9.70948
-9.70421
3×3
-0.65847
-0.58042
-0.64660
-0.54975
M̄12
-0.66318
9×9
-0.66253
-0.65794
-0.66156
-0.64927
27×27
-0.66311
-0.66264
-0.66300
-0.66174
for the two different stacking sequences, with N denoting the number of elements
per side. As it can be observed an optimal behaviour is experienced with a
very low sensitivity to geometry distortions. Pointwise convergence results are
reported in Figures 2.9 and 2.10, showing the transverse displacement w in the
plate centre (point A in Figure 2.7) and the in-plane displacement u1 at point B,
respectively. Note that the in-plane displacements are different from zero in the
(0/90) antisymmetric laminate because of the membrane-bending coupling. In
addition, Figures 2.11 and 2.12 show the convergence of the generalised stresses
at point C under uniform load for the symmetric and the antisymmetric case,
respectively. Finally, Tables 2.1-2.4 report some results predicted by the present
element and ADINA. The quantities are rendered dimensionless as follows:
w̄ = w
100
100Eb
4,
qh (L/h)
ū1 = u1
100
100Eb
4,
qh (L/h)
100
.
qh (L/h)
(2.123)
(2.124)
qh (L/h)
(L/h)
As it can be noted, a superconvergent behaviour is observed for the stress
resultants at point C, using either HQ4 or ADINA. Indeed, the convergence rate
obtained elsewhere is one, as expected (see for example Figures 2.15 and 2.16
in Section 2.3 for convergence at point D).
N̄ = N
2,
M̄ = M
qh2
2,
S̄ = S
66
Modelling of fibre-reinforced composites
All these results show that the performance of the HQ4 element is comparable with that of the MITC element from both the point of view of accuracy
and convergence rate. However, the proposed element tends to be less sensitive
to mesh distortions and more accurate in terms of stress resultants.
2.3
Reconstruction of the transverse stresses
Once the solution of the plate problem is available, the transverse stresses can
be reconstructed through Equations (2.78) and (2.79) introducing the constitutively determined in-plane stresses (Daghia et al., 2008). A satisfactory approximation of the transverse shear stresses is particularly important in order to
iteratively update the shear correction factors, as outlined in Section 2.1.1. The
passage from the analytical strategy to the numerical implementation, however,
requires a few careful considerations.
First of all, the need for pointwise equilibrated stress resultants as pointed
out in Section 2.1.1 is not necessarily met by approximate numerical solutions.
Indeed, if pointwise two-dimensional equilibrium is lacking, automatic satisfaction of the boundary conditions and static equivalence of the reconstructed
transverse stresses is lost. For this reason, the two-dimensional stresses obtained
via a hybrid stress approach, such as the one proposed in Section 2.2, are ideal
candidates for a post processing based on local three-dimensional equilibrium.
On the other hand, the reconstruction procedures associated to other finite elements, such as those by Alfano et al. (2001), Auricchio and Sacco (1999a) and
Auricchio et al. (2006), need to be corrected to properly account for the boundary equilibrium conditions and the static equivalence. The same equilibrium
defect affects also the procedures proposed by Noor et al. (1994) and Park and
Kim (2003), both based on a superconvergent recovery of the in-plane stresses
in the different layers.
The second issue to be taken into consideration concerns the convergence
of the numerically reconstructed transverse stresses to the theoretical (FSDT)
plate solution. Indeed, accuracy of the reconstructed shear stresses depends on
the accuracy of the first derivatives of the finite element stress resultants, as
they are used to determine the in-plane active stress. This is a crucial point
of all the procedures based on local equilibrium, especially for low order plate
finite elements like the one presented in Section 2.2 or those proposed by Alfano
2.3 Reconstruction of the transverse stresses
67
et al. (2001), Auricchio and Sacco (1999a), Auricchio et al. (2006) and Cen
et al. (2002), and generally leads to non convergent results. Here, a two stages
reconstruction strategy is proposed to overcome this difficulty. First, a patchbased superconvergent procedure is applied to recover the stress resultants by
locally post processing the finite element solution, then the recovered values
are used to reconstruct the transverse shear stresses, following the idea early
proposed by Noor et al. (1994). In this paper, the recovery is carried out by
the procedure called Recovery by Compatibility in Patches (RCP) (Benedetti
et al., 2006; Ubertini, 2004), which has been recently extended to homogeneous
plate structures (Castellazzi et al., 2006) and is here established for laminates.
Besides being simple, stable, robust and very efficient from the computational
point of view, it has been proved to be superconvergent in certain circumstances
and, using patches centred on an element instead of a node, yields recovered
stress resultants which satisfy the plate equilibrium equations. This is a key
feature in this context since, as already mentioned, the reconstruction procedure
can be applied without any correction due to non equilibrated terms. Once
accurate transverse shear stresses are reconstructed, the third local equilibrium
condition can be used to compute the transverse normal stress profile. Indeed,
the accuracy of this second reconstruction step depends on the accuracy of
the first derivatives of the transverse shear stresses. Some possible strategies to
further improve the quality of the resultant transverse normal stress are outlined
in the following.
It should be emphasised that the whole procedure based on RCP recovery
and shear stresses reconstruction is quite general and can be successfully applied
to different laminate finite elements, without any modification or adjustment.
The outlined strategy differs from the one proposed by Noor et al. (1994) for
the assumed recovery technique and patch configuration, which guarantees local equilibrium. On the other hand, it is similar to the one proposed by Cen
et al. (2002), based on a hybrid enhanced elementwise post processing procedure. However, the present recovery works on patches instead of simple elements
and this is the key feature which actually guarantees robustness and superconvergence. A comparative analysis of various models and techniques to evaluate
out of plane stresses a priori or a posteriori was carried out by Carrera (2000),
who also proposed a similar reconstruction procedure based on a partial mixed
formulation, called weak form of Hooke’s law and operating on single elements
68
Modelling of fibre-reinforced composites
instead of patches. Other strategies to evaluate laminate transverse stresses
proposed in the literature include the simplified reconstruction procedure proposed by Rolfes and Rohwer (1997), based on assuming two cylindrical bending
modes and neglecting the influence of membrane forces, and, more recently,
the enhanced first-order theory allowing for through the thickness recovery of
displacements and stresses developed by Kim and Cho (2007).
2.3.1
Recovery by Compatibility in Patches
Recovery by Compatibility in Patches (RCP) is a stress recovery procedure
based on the minimisation of the complementary energy associated to a patch
of elements, considered as a separate system, among an assumed set of equilibrated fields. Hence, it substantially attempts to enhance equilibrium while
relaxing compatibility. Here, it is applied in the version based on the element
patch configuration (Benedetti et al., 2006), that is with patches defined by an
element (central element) and the union of elements surrounding it, as shown
in Figure 2.13. The same patch configuration was early adopted by Mohite and
Upadhyay (2002) as the support for a strain recovery procedure. The new solution computed over the patch is taken directly as the recovered solution for the
central element, with no need of any additional averaging process. In this way,
the recovered stress resultants on each element are simply obtained by forming
the associated patch and applying the RCP procedure over it. Notice that the
recovered solution is discontinuous across the elements but satisfies pointwise
the plate equilibrium equations within each element, as required to be used for
the subsequent reconstruction procedure of transverse stresses.
To apply the recovery procedure, a new approximation for stress resultants
over the patch is introduced:




Nr
Np
 Mr  = Pr α +  Mp  ,
Sr
Sp
(2.125)
where the superscript r denotes the recovered solution, Pr is a matrix of selfequilibrated modes, α is a vector of unknown parameters and the last term is
a particular solution of the plate equilibrium equations, which is taken as equal
to the one selected for the plate finite element formulation (see Section 2.2.2).
2.3 Reconstruction of the transverse stresses
69
Figure 2.13: Element patch for the RCP
In particular, an approximation ruled by 14 α-parameters is selected based
on linear polynomial expansions in terms of local coordinates (x1 , x2 ) centred
on the patch:



 r  

N

 Mr  = 



Sr



1
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
1
0
0
0
0
0
x2
0
0
0
0
0
0
0
0
x1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
x1
0
0
1
0
0
0
0
x2
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
x1
0
0
0
0
0
0
0
x2
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
0
x1
0
1
0
0
0
0
0
x2
1
0







 α.





(2.126)
Then, the RCP minimisation yields the following compatibility condition
over each patch Ωp :
Z
£ rT ¡ r
¢
¡
¢
¡
¢¤
δN
µ − µh + δMrT χr − χh + δSrT γ r − γ h dΩp = 0
Ωp
∀ (δNr , δMr , δSr ) ,
(2.127)
where (µr , χr , γ r ) are the strain components obtained from the recovered stress
¡
¢
resultants via the plate constitutive equations and µh , χh , γ h are the strain
components resulting from the finite element solution. They could be computed
from either the assumed finite element displacements, via the plate compatibility
70
Modelling of fibre-reinforced composites
equations, or the assumed finite element stress resultants, via the plate constitutive equations. These two possibilities offered by mixed finite elements have
been explored by Castellazzi et al. (2006). Here, the first alternative based on
the finite element kinematics is adopted.
Substituting Equation (2.125) into Equation (2.127) leads to a system of 14
linear algebraic equations, whose solution permits to determine the α-parameters
and, hence, the recovered stress resultants over the central element of the patch:
Hα = g,
(2.128)
where
H
=
j−1
g
=

nep Z
X
PrT 
Ωj
sym
Fmb
Fb

0
0  Pr dΩ,
Fs
(2.129)

Dm Uu qu − Fm Np − Fmb Mp
 dΩ,
PrT 
Dm Uθ qθ − Fmb Np − Fb Mp
Ωj
Ds Uw qw + (Ds L + Uθ ) qθ − Fs Sp

nep Z
X
j=1
Fm
(2.130)
where

Dm
∂/∂x1
=
0
∂/∂x2

0
∂/∂x2  ,
∂/∂x1
·
Ds =
∂/∂x1
∂/∂x2
¸
,
(2.131)
nep is the number of elements in the patch and Ωj the domain of the general
element of the patch.
It should be remarked that the outlined procedure is very simple, easily
implementable into existing codes, unconditionally stable and computationally
efficient. In particular, differently from the procedure proposed by Alfano et al.
(2001) for MITC elements, which is an L2 projection over the entire domain,
it is a minimum condition over small patches of elements with an extremely
low computational cost. Moreover, the RCP procedure has been proved to be
truly superconvergent on certain regular mesh patterns but, anyway, extremely
robust on general mesh patterns (Benedetti et al., 2006), with an almost optimal
rate of convergence. In the present case, the rate of convergence expected for
the recovered stress resultants is about two in energy norm (one order higher
than the finite element stress resultants).
2.3 Reconstruction of the transverse stresses
2.3.2
71
Reconstruction of the transverse shear stresses
The higher convergence rate of the recovered stress resultants is expected to
ensure convergence of their first derivatives in the element interior and, as a
consequence, convergence of the subsequent reconstruction procedure of transverse shear stresses:
Z z n
(−)
τ (z) = −px +
D∗m C(k)
m
−h
2
r
[(Fm N + Fmb Mr ) + z (Fmb Nr + Fb Mr )] − bx } dz,
(2.132)
where the in-plane stresses are computed using the recovered stress resultants
and the symbol ∗ denotes the adjoint operator.
2.3.3
Reconstruction of the transverse normal stress
Once accurate transverse shear stress profiles have been obtained, they can be
used in Equation (2.79) to evaluate the transverse normal stress. As already
mentioned, the whole procedure is statically admissible, so that the resultant
transverse normal stress automatically satisfies both the boundary equilibrium
conditions. It should be remarked, however, that the reconstruction accuracy
depends on the first derivatives of the transverse shear stresses with respect to
the in-plane coordinates, as they enter Equation (2.79), and consequently on the
second derivatives of the stress resultants. For this reason, the RCP recovery
should be carried out using at least a quadratic stress approximation. With
this consideration, the quality of the predicted profiles is generally satisfactory
at the centre of the elements, but tends to deteriorate for thick laminates while
moving towards the element boundaries. Indeed, even with a quadratic stress
approximation, the proposed recovery procedure, as well as most of the available superconvergent recovery procedures, is not designed to guarantee a priori
convergence of second derivatives everywhere within the element. For this reason, accurate transverse normal stresses are not a priori guaranteed everywhere
within the element by using the recovered shear stresses computed according to
the previous section. However, the resultant profiles are accurate enough for a
preliminary evaluation of thick laminate response. Note that, differently from
the transverse shear stresses which are needed also to evaluate the shear correction factors, the transverse normal stress does not influence at all the plate
72
Modelling of fibre-reinforced composites
solution.
Two different strategies can be devised to further enhance accuracy of the
transverse normal stress and guarantee its convergence in any case (see the paper
by Vallet et al. (2007) for some details on Hessian recovery techniques). The first
strategy is to apply a second superconvergent recovery procedure, inspired by
the RCP procedure, to improve transverse shear stresses before reconstructing
the transverse normal stress using three-dimensional equilibrium. Differently
from the second regularisation proposed by Alfano et al. (2001), the procedure
should be patch-based and should preserve equilibrium. Similarly to the first
RCP recovery previously described, compatibility is relaxed by enforcing the
following condition on each patch:
Z
£ r̄T ¡ r̄
¢
¡
¢¤
¡
¢
δs
e − er + δτ r̄T γ r̄ − γ r dV = 0 ∀ δsr̄ , δτ r̄ ,
(2.133)
Ωp ×F
where the superscript r denotes the quantities related to the first RCP recovery,
while the superscript r̄ denotes the newly recovered quantities. The assumptions for sr̄ and τ r̄ should satisfy the three-dimensional equilibrium and the
interlaminar continuity conditions, in particular the in-plane stresses should be
piecewise linear along the thickness direction while the transverse shear stresses
should contain at least piecewise parabolic terms. The second strategy is to
properly modify the outlined recovery procedure of the stress resultants by using larger node centred patches and separately interpolate the stress resultants
and their first and second derivatives, so to ensure the required convergence.
Both the strategies are promising but they are still under investigation. Thus,
only the numerical results on transverse shear stresses are discussed in the next
section.
2.3.4
Numerical testing of the reconstruction procedure
In this section, a numerical investigation of the procedure proposed to reconstruct transverse shear stresses from recovery and three-dimensional equilibrium
is carried out. The test case considered is the same as in Section 2.2.4. The twodimensional problem finite element solution is obtained using the HQ4 element
presented in the previous section.
The first set of graphs highlights the key role of the RCP procedure in this
context. Figure 2.14 shows the convergence in energy norm of the shear stress
2.3 Reconstruction of the transverse stresses
0
-0.5
-1
-1.5
-2
0/90 - sinusoidal load
-0.5
Log(error %)
Log(error %)
0
0/90/0 - sinusoidal load
regular
regular, recovery
distorted
distorted, recovery
73
-1
-1.5
-2
1
-2.5
-2.5
1
-3
-1
-0.5
0
Log(L/N)
0.5
-3
-1
-0.5
0
Log(L/N)
0.5
Figure 2.14: Convergence in energy norm of the shear stress profiles for sinusoidal load and both the stacking sequences, using HQ4 with or without RCP
recovery
profiles reconstructed using directly the finite element in-plane stresses or those
recovered by RCP:
Z
T
error =
(τ − τref ) F(k)
(2.134)
s (τ − τref ) dV.
Ω×F
The sinusoidal load case is considered with both the stacking sequences. The reference solution for the shear stress profiles is that obtained from the continuum
FSDT, using the same shear correction factors.
The plots of Figure 2.14 reveal that the reconstruction procedure used directly with the finite element results does not converge globally, although it
may converge at some special points. Indeed, this is always the case if the procedure is applied to two-dimensional solutions obtained using four node finite
elements. On the contrary, using the RCP recovery the reconstructed shear
stress profiles globally converge at an optimal rate. This anomalous behaviour
is due to the low convergence rate of the finite element stress resultants, that
is generally one as shown by Figures 2.15 and 2.16, exception done for some
special points, such as the centre of the element, where occasionally it may be
two, as shown by Figures 2.11 and 2.12 in Section 2.2. Figures 2.15 and 2.16,
indeed, depict the convergence of the moments at point D, before and after
RCP recovery. It should be noted that point D coincides with a node in all the
74
Modelling of fibre-reinforced composites
0
0
M1
regular
regular, recovery
-1
Log(error %)
Log(error %)
-1
M2
-2
1
-3
-2
1
2
-3
2
-4
-1
-0.5
0
Log(L/N)
-4
0.5
-1
-0.5
0
Log(L/N)
0.5
0
M12
Log(error %)
-1
1
-2
-3
2
-4
-1
-0.5
0
Log(L/N)
0.5
Figure 2.15: Convergence of the moments at point D by HQ4 with or without
RCP recovery, for the (0/90/0) stacking sequence and sinusoidal load
0
regular
regular, recovery
0
M1 and M2
-2
-1
Log(error %)
Log(error %)
-1
1
-3
-4
M12
1
-2
-3
2
2
-1
-0.5
0
Log(L/N)
0.5
-4
-1
-0.5
0
Log(L/N)
0.5
Figure 2.16: Convergence of the moments at point D by HQ4 with or without
RCP recovery, for the (0/90) stacking sequence and sinusoidal load
2.3 Reconstruction of the transverse stresses
75
sequence of refined meshes (see Figure 2.7). Since both finite element and recovered stress resultants are discontinuous across elements, the values reported
have been computed for the element having D as the top right-hand node. It
should be remarked that point D is a severe point for recovery procedures, since
the error tends to concentrate on the element boundaries. Nevertheless, RCP
recovery significantly improves the solution.
To sum up, the above results demonstrate that RCP procedure can be successfully applied also to laminated plates and confirm the superconvergence
properties. In fact the recovered stress resultants, besides being generally more
accurate, exhibit a convergence rate of almost two everywhere and not only at
the element centre. As a consequence, the RCP-based reconstructed shear stress
profiles globally converge, as shown by Figure 2.14.
To complete the experimental evaluation, Figures 2.17-2.20 show the dimensionless transverse shear stress profiles at point C for the two different stacking
sequences and the sinusoidal load. The shear stress is rendered dimensionless
as follows:
τ̄zs = τzs
100
.
q (L/h)
(2.135)
In the first two figures, the profiles reconstructed by making use of RCP recovery
are shown for three mesh refinements. As it can be observed, the profiles quickly
converge to the reference solution. In Figures 2.19 and 2.20, on the other hand,
the profiles predicted with or without recovery are illustrated for 9×9 meshes of
regular and distorted elements. Observing for example the results in terms of τ̄z2
for the (0/90/0) laminate with the distorted mesh (black dots in Figure 2.19),
it is evident that the shape of the profile is very different from the reference
solution, although the shear force (represented by the area beneath the graph)
is similar. This confirms that, while the value of the shear force is close to the
exact one, the quality of the shear stress profiles may be very poor unless the
RCP recovery is employed. In this case, the shear stress profiles are accurately
captured even on coarse meshes.
76
Modelling of fibre-reinforced composites
0.5
0.5
0.25
0.25
Ref. sol.
Mesh 3x3
Mesh 9x9
Mesh 27x27
z/h 0
z/h 0
-0.25
-0.5
-20
-0.25
-15
-10
-5
0
-0.5
-8
-6
-4
-2
0
tz2
tz1
Figure 2.17: Shear stress profiles at point C using RCP recovery on various
regular meshes, for the stacking sequence (0/90/0) and sinusoidal load
0.5
0.25
0.5
Ref. sol.
Mesh 3x3
Mesh 9x9
Mesh 27x27
0.25
z/h 0
z/h 0
-0.25
-0.25
-0.5
-20
-15
-10
tz1
-5
0
-0.5
-20
-15
-10
-5
0
tz2
Figure 2.18: Shear stress profiles at point C using RCP recovery on various
regular meshes, for the stacking sequence (0/90) and sinusoidal load
2.3 Reconstruction of the transverse stresses
0.5
0.5
0.25
0.25
Ref. sol.
Regular, without recovery
Regular, with recovery
Distorted, without recovery
Distorted, with recovery
z/h 0
z/h 0
-0.25
-0.5
-20
77
-0.25
-15
-10
-5
0
-0.5
-8
-6
tz1
-4
-2
0
tz2
Figure 2.19: Shear stress profiles at point C with and without RCP recovery on
9 × 9 meshes for the stacking sequence (0/90/0) and sinusoidal load
0.5
0.25
0.5
Ref. sol.
Regular, without recovery
Regular, with recovery
Distorted, without recovery
Distorted, with recovery
0.25
z/h 0
z/h 0
-0.25
-0.25
-0.5
-20
-15
-10
tz1
-5
0
-0.5
-20
-15
-10
-5
0
tz2
Figure 2.20: Shear stress profiles at point C with and without RCP recovery on
9 × 9 meshes for the stacking sequence (0/90) and sinusoidal load
78
2.4
Modelling of fibre-reinforced composites
Bayesian estimation of laminates’ elastic constants
Equivalent Single Layer and layerwise laminate theories model laminae as orthotropic continua, thus they require the definition of the lamina constitutive
properties in the material reference system. Even if the properties in the transverse direction are neglected, the independent material constants are significantly more than in the isotropic case, and generally more difficult to determine.
Indeed, traditional static tests on laminates with a general stacking sequence
are not able to identify the separate material properties because of the coupled
behaviour described in the previous sessions. The typical solution is to test ad
hoc specimens, whose laminae contain fibres oriented all in the same direction.
Thus, results obtained for a (0)n laminate are generalised to a single lamina,
regardless of the scale effects which might influence this passage. Moreover,
even on a fully unidirectional laminate some material properties, such as the
in-plane and transverse shear moduli, are not easily determined.
An alternative approach to determine material properties for a laminate
with a general stacking sequence is represented by the so called numericalexperimental techniques. These are estimation methods which attempt to minimise in some sense the difference between measured data and the response predicted by an assumed model. Structural applications include identification of
the characteristics of damaged and undamaged structures via dynamic and static experiments (see, for example, Di Paola and Bilello, 2004). Non destructive
techniques based on the use of vibration data to identify structural characteristics date back to the 1970s. Statistical analysis of vibrating structural systems
was early carried out by Collins and Thomson (1969) and Hasselman and Hart
(1972). A probabilistic method to identify the elastic constants of simple structures starting from natural frequency data was proposed by Collins et al. (1974).
Since the 1990s, the same approach has been applied to laminated composite
plates (see for example Viola et al. (2003) and the references therein).
Here, only natural frequency measurements are supposed to be available for
the identification process, since they can be evaluated experimentally through
a simple modal test and, in most approaches present in the literature, they are
generally found to be sufficient to estimate satisfactorily the elastic constants.
Anyway, the estimation process could be improved by adding more information
2.4 Bayesian estimation of laminates’ elastic constants
79
on the structural response, such as that coming from mode shapes.
Many approaches have been proposed which differ in the choice of the theoretical and numerical model and the procedure used to estimate the elastic
constants. In particular, these procedures aim at minimising a non linear objective function, which involves the difference between measured and numerical
frequencies, and differ in the objective function and the algorithm used to solve
the non linear minimisation problem.
The simplest objective functions are ordinary least squares or weighted least
squares (Frederiksen, 1997). These functions take into account only the information coming from the experiment, and require no a priori knowledge on the
possible parameter values. Although general, this approach needs robust and advanced search algorithms to obtain reliable results. Optimum design techniques
were employed by Hwang and Chang (2000), experimental design techniques
together with the response surface method were employed by Bledzki et al.
(1999) and Rikards et al. (2001) and a feasible directions non linear interior
point algorithm was employed by Araujo et al. (2002, 2000).
Alternatively, more complex objective functions can be adopted by taking
into account the a priori information on the elastic constants (Bayesian framework). This is the approach used, among others, by Lai and Ip (1996) for thin
plates and later extended to thick plates by Bartoli et al. (2003). A different
choice to take into account the a priori knowledge on initial values can be found
in the work by Hongxing et al. (2000) and Sol et al. (1997). A comparison
between different methods for the identification of in-plane elastic constants of
steel orthotropic plates was carried out by Lauwagie et al. (2003). In all these
cases, the minimum of the objective function can be easily computed through a
simple iterative algorithm, provided that the initial guesses of the elastic constants are not too far from the target values. Indeed, this is often the case for
laminated composites, since a reasonable set of initial estimates can be found
through the rule of mixtures, given the geometric properties of the laminae (fibre
and matrix volume fraction) and an initial guess on the components’ material
properties (fibre and matrix elastic constants).
In this section, two well known estimators which make use of a priori information are considered for the identification of elastic constants of thick plates
based on measured natural frequencies (Daghia et al., 2007a). The first estimator stems from Bayes theorem and can be strictly termed as Bayesian. The
80
Modelling of fibre-reinforced composites
second estimator is the minimum variance estimator. In the following, they are
shortly referred to as B and MVE. The two estimators are compared by focusing
on the role played by the a priori information. Although the two formulations
are very different, a detailed analysis reveals a certain resemblance between
the two, which helps to understand their behaviour and suggests an improved
solution procedure (referred to as MVE-mod in the following). To complete
the estimation process, four nodes finite elements based on Reddy’s HSDT are
used to model the plate. A higher-order theory is used since it is expected
to accurately capture the first natural modes, which are usually measured for
estimation purposes.
2.4.1
Bayesian estimators comparison
In a physical problem, the relationship between the problem data (input) and
the outcome of the experiment (output) can be generally represented by
Y = η(β) + ε,
(2.136)
where Y is the vector containing the output values, η is a function representing the model used to describe the phenomenon, β is the vector of the input
parameters and ε is the error, which includes both modelling and measurement
errors. In structural engineering, β are the geometrical and mechanical properties of the structure and Y represents the structural response, for example in
terms of natural frequencies.
While Equation (2.136) describes the direct problem, that is predicting the
experiment outcome given the problem data, a number of different objective
functions can be defined in order to estimate the input parameters from a set
of measured experimental data. In particular, focusing on Bayesian estimators,
the data required by the estimation procedure includes a set of initial parameter
estimates µ.
Bayesian estimator (B)
The Bayesian estimator (Beck and Arnold, 1977) is obtained by applying Bayes
theorem of conditional probabilities, that states:
f (β|Y) =
f (Y|β)f (β)
,
f (Y)
(2.137)
2.4 Bayesian estimation of laminates’ elastic constants
81
where f (a) is the probability density function of the variable a and f (a|b) stands
for the probability of the event a given that the event b has occurred. In the B
estimator, the aim is to maximise the function (2.137). The probability densities
f (Y|β) and f (β) are given by the following functions:
s
Ã
!
−1
(Y − η)T VY (Y − η)
|VY |−1
exp −
,
(2π)m
2
s
Ã
!
−1
(β − µ)T Vβ (β − µ)
|Vβ |−1
exp −
,
(2π)p
2
f (Y|β) =
f (β) =
(2.138)
(2.139)
where VY and Vβ are the covariance matrices of the experimental error and
initial estimates, respectively, m is the number of experimental observations and
p is the number of parameters. Equation (2.138) represents the information
deduced from the experiment, while Equation (2.139) represents the a priori
information on the parameters to be estimated. The matrices VY and Vβ have
to be specified by the analyst.
Using Equations (2.138) and (2.139), the natural logarithm of equation
(2.137) takes the form:
1
ln(f (β|Y)) = − [(m + p) ln(2π) + ln |VY | + ln |Vβ | + SB ] − ln (f (Y)) ,
2
(2.140)
where the only term dependent on the parameters is
T
T
SB = (Y − η) VY−1 (Y − η) + (β − µ) Vβ−1 (β − µ) ,
(2.141)
thus the B estimator is obtained by minimising SB . Calculating the derivatives
with respect to β and setting them equal to zero, one obtains:
h
i
5β SB = 2 −ST (β)VY−1 (Y − η(β)) − Vβ−1 (µ − β) = 0,
(2.142)
where
S(β) = 5β η(β)
(2.143)
is the sensitivity matrix. Notice that in Equation (2.142) the parameters β
appear both explicitly and implicitly.
82
Modelling of fibre-reinforced composites
In order to minimise SB , an iterative procedure is established. Let b be a
vector of initial parameter values. The term η(β) is approximated by a truncated Taylor series expansion in the neighbourhood of b:
η(β) ≈ η(b) + S(b)(β − b),
S (β) ≈ S (b) .
(2.144)
Hence, Equation (2.142) takes the form
ST (b)VY−1 [Y − η(b) − S(b)(β − b)] + Vβ−1 (µ − β) ≈ 0
(2.145)
and a recurrence relation can be established:
¡
¢T
£
¡
¢
¡
¢¤
¡
¢
Sk VY−1 Y − η k − Sk bk+1 − µ − Sk µ − bk − Vβ−1 bk+1 − µ =0,
(2.146)
where the superscript k denotes the quantity calculated at the k-th iteration,
bk are the prior set of estimates and bk+1 are the parameter estimates at the
end of the current iteration.
Solving for bk+1 yields:
bk+1 = µ +
h¡
Sk
¢T
VY−1Sk + Vβ−1
i−1¡
Sk
¢T
£
¡
¢¤
VY−1 Y − η k − Sk µ − bk .
(2.147)
Minimum variance estimator (MVE)
The estimator presented in the work of Collins et al. (1974) is obtained by
minimising the variance associated with the estimator.
The function η(β) is expanded in Taylor series as in (2.144). The relation
(2.136) between the experimental measurements and the parameters becomes:
Y
≈
η(b) + S(b)(β − b) + ε.
(2.148)
Then, introducing the following quantities
β̄ = (β − b),
Ȳ = Y − η(b),
Equation (2.148) can be written as
Ȳ = Sβ̄ + ε.
(2.149)
2.4 Bayesian estimation of laminates’ elastic constants
83
The variables β̄ are normally distributed with zero mean and covariance
matrix Vβ̄ = Vβ . The experimental errors ε are assumed to be normally
distributed, with zero mean and covariance VY , and uncorrelated with β̄. Thus,
the mean of Ȳ is null, the covariance of Ȳ and the covariance of Ȳ and β̄ are
given by
VȲ β̄ = SVβ .
(2.150)
VȲ = SVβ ST + VY ,
The following inverse relation holds:
β̄ ∗ = GȲ,
(2.151)
where β̄ ∗ is the vector of the estimates of β̄.
The aim is to find the estimator G that minimises the variance associated
with the estimator itself:
E[(β̄ ∗ − β̄)(β̄ ∗ − β̄)T ] = GVȲ GT − VȲT β̄ GT − GVȲ β̄ + VY .
(2.152)
This yields:
G = VȲT β̄ VȲ−1
(2.153)
and, with some algebra, the estimator can be put in the form
β = b + Vβ ST (SVβ ST + VY )−1 (Y − η(b)).
(2.154)
By setting β = bk+1 and b = bk , the following iteration scheme can be
established:
£
¤−1
bk+1 = bk + Vβ (Sk )T Sk Vβ (Sk )T + VY
(Y − η k ).
(2.155)
At the end of each iteration it is possible to evaluate the new covariance matrix
of the parameters Vβk+1 by
Vβk+1
= Vβ − VȲT β̄ VȲ−1 VȲ β̄ =
£
¤−1 k
= Vβ − Vβ (Sk )T Sk Vβ (Sk )T + VY
S Vβ .
(2.156)
Remarks on the equivalence of the estimators
The two iteration schemes (2.147) and (2.155) coalesce if posing µ = bk . In
other words, the MVE estimator can be interpreted as a B estimator with
initial parameters not kept fixed, but updated at each iteration. A proof of
this statement can be obtained through the inversion lemma of matrices.
84
Modelling of fibre-reinforced composites
Proof. The following matrix identity holds:
(IP + AB)−1 A = A(IN + BA)−1 ,
(2.157)
where AP ×N and BN ×P are general matrices, IP and IN are the identity matrices of dimensions P and N . By posing A = Vβ ST VY−1 and B = S, Equation
(2.157) becomes:
(Vβ−1 + ST VY−1 S)−1 ST VY−1
=
Vβ ST (VY + SVβ ST )−1 ,
(2.158)
which shows the equivalence between Equations (2.147) and (2.155) under the
hypothesis µ = bk .
To sum up, the two estimators coincide in the linear case, but the iterative
corrections needed for non linear problems may lead to very different results.
In the B estimator, the a priori information enters the objective function and
somehow conditions the final estimates, besides driving the estimation process.
In the MVE estimator, on the other hand, the a priori information provides
starting values to the solution algorithm and its influence gradually decreases
as the iterations proceed, so that only the experimental information is retained
at the end.
Convergence criterion
The iteration schemes presented in (2.147) and (2.155) require the definition of
a convergence criterion. Here, a criterion based on the parameter estimates is
adopted: the iterations stop when the updated parameters bk+1 fall within a
certain interval from the previous set of estimates bk . This can be written as
¯
¯
¯ bk+1 − bk ¯
¯ i
i ¯
∀i,
(2.159)
¯
¯ ≤ TOL
¯
¯
bki
where TOL is the prescribed tolerance.
Estimator components for the dynamic identification of laminated
composite plates
So far, the estimators have been presented in general terms, while in the following the meaning of the quantities introduced is specified for the dynamic
identification of laminated plates.
2.4 Bayesian estimation of laminates’ elastic constants
85
The experimental data is composed of a set of undamped natural frequencies fexp of the free edge plate. The output of both model and experiment is
represented in terms of the plate eigenvalues, obtained as the solution of the
classical eigenvalue problem
(K − λM) a = 0,
(2.160)
which are related to the undamped natural frequencies by the following relation:
2
λ = (2πf ) .
(2.161)
A simple displacement based finite element formulated for Reddy’s HSDT was
used to determine the stiffness and mass matrices K and M. Anyway, the same
procedure can obviously be applied using any model, provided it can accurately
capture the natural frequencies which are involved in the estimation process.
The deviation associated to the experimental eigenvalues can be evaluated
as
µ
¶
∂λexp
δλexp =
δf = 8π 2 fexp δf ,
(2.162)
∂fexp
where δf is the deviation of fexp .
The parameters to be estimated are the reduced stiffness coefficients of the
(k)
laminae in the material coordinates, C̄ij , from which the laminae elastic constants can be obtained. In particular, in order to model both the in-plane and
through the thickness plate behaviour, the constants to be estimated are Ea , Eb ,
νab , Gab , Gac and Gbc , where the material coordinate system (a, b, c) is defined
in Figure 2.2. The deviation associated to the stiffness parameters is obtained
from the deviation associated to the elastic constants.
The quantities which appear in the estimator formulae (2.147) and (2.155)
are
YT
=
T
=
η
b =
£
£
h
λexp,1
λ1
(k)
C̄ij
λexp,2
λ2
i
···
,
VY
=
diag(δλexp,i ),
Vβ
=
diag(δ C̄ij ),
(k)
···
λm
¤
λexp,m
,
¤
,
86
Modelling of fibre-reinforced composites
where m is the number of experimental frequencies considered and λi are the
numerical eigenvalues corresponding to the current value of the stiffness coefficients b.
The typical component of the sensitivity matrix S is:
Sij =
∂λi
,
∂bj
(2.163)
where bj denotes the j-th component of b. It can be evaluated, as early proposed
by Collins and Thomson (1969) and Fox and Kapoor (1968), by
1
∂λi
∂K
= T
ai .
aT
∂bj
ai Mai i ∂bj
(2.164)
Notice that the sensitivity coefficients (2.163) can be normalised as follows (Ayorinde and Yu, 2005):
∂λi bj
sij =
.
(2.165)
∂bj λi
In this way the sensitivity of each mode to different parameters can be evaluated
and compared.
Analysis and comparison
In this section the behaviour of the B and MVE estimators is analysed and
compared through some numerical tests, devised ad hoc to highlight similarities and differences. Two different single-layer orthotropic rectangular plates,
whose geometric characteristics are reported in Table 2.5, are modelled through
Reddy’s third order theory for a given set of elastic constants (target). The two
plates are discretised by uniform meshes of 8×4 and 8×8 elements, respectively.
Supposing to operate in ideal conditions, the experimental frequencies are taken
as equal to the numerical ones for the estimation process. This eliminates all
the sources of error (such as modelling and measurement errors, see Equation
(2.136)) and the estimators are expected to yield the target elastic constants,
starting from a set of guessed initial values. On the other hand, only the first
14 natural frequencies are considered in the estimation (see Table 2.6) since
it is well known that in real experimental conditions it is difficult to measure
frequencies associated to higher modes. Moreover, capturing higher modes may
require more accurate plate models.
2.4 Bayesian estimation of laminates’ elastic constants
87
Table 2.5: Characteristics of Plate 1 and Plate 2
Plate 1
Plate 2
length L
200 mm
100 mm
width W
100 mm
100 mm
mean
thickness h
6.6 mm
10 mm
density ρ
1000 kg/m3
1500 kg/m3
stacking
sequence
[0]
[0]
Table 2.6: Input frequencies of Plate 1 and Plate 2
Plate 1
Plate 2
mode
no.
fexp
mode
no.
fexp
mode
no.
fexp
mode
no.
fexp
1
2
3
4
5
6
7
611.23
2002.7
2171.2
2338.5
2478.2
3810.3
5165.3
8
9
10
11
12
13
14
5406.2
6121.1
6389.0
6520.0
7357.6
8343.2
8809.4
1
2
3
4
5
6
7
2555.5
3591.6
6069.2
9146.4
10677
11305
11512
8
9
10
11
12
13
14
13985
15744
16334
16440
18067
18156
18371
Various numerical tests have been carried out for different initial guesses
and parameter covariance. The final estimates and number of iterations performed by the estimation procedures are reported in Tables 2.7-2.12. To check
convergence, the tolerance has been set to TOL = 10−3 . The deviation of the
experimental data was assumed to be δf = 0.01 · fexp . In order to establish a
common ground for comparison, the deviation associated to each elastic constant is taken proportional to the constant itself by a factor d. The symbol indicates that the procedure does not converge. Since the two plates considered
are single-layer, the superscript (k) referring to the k-th lamina is dropped in
the following.
Since both the algorithms are based on the sensitivity matrix, a sensitivity analysis is useful to understand the behaviour of the two estimators. The
absolute values of the normalised sensitivity coefficients (2.165) are plotted in
Figures 2.21 and 2.22 for Plate 1 and Plate 2, respectively.
Both diagrams show that the first natural frequencies are more sensitive to
C̄11 , C̄22 and C̄66 , that is the stiffness coefficients associated to the two elastic
88
Modelling of fibre-reinforced composites
Table 2.7: Test 1 on Plate 1: final estimates and iterations
dev.
d
Initial
B
B
B
MVE
MVE
MVE
Target
1/3
1/6
1/10
1/3
1/6
1/10
Ea
(GPa)
100
145.0
142.0
150.1
149.8
150
Eb
(GPa)
7
10.41
10.47
9.996
10.01
10
νab
0.33
0.2087
0.2149
0.2539
0.2389
0.25
Gab
(GPa)
4
3.008
3.019
3.000
3.000
3
Gac
(GPa)
4
3.554
3.924
2.995
3.015
3
Gbc
(GPa)
1
1.404
1.277
2.004
1.989
2
n.
iter.
15
7
87
85
Table 2.8: Test 2 on Plate 1: final estimates and iterations
dev.
d
Initial
B
B
B
MVE
MVE
MVE
Target
1/3
1/6
1/10
1/3
1/6
1/10
Ea
(GPa)
180
152.4
156.7
159.2
150.0
150.0
150.3
150
Eb
(GPa)
12
9.870
9.663
9.630
10.00
10.00
9.987
10
νab
0.20
0.2787
0.2726
0.2639
0.2493
0.2466
0.2626
0.25
Gab
(GPa)
2
2.997
2.981
2.962
3.000
3.000
3.000
3
Gac
(GPa)
2
2.791
2.397
2.272
3.001
3.005
2.977
3
Gbc
(GPa)
4
2.261
3.442
3.801
1.999
1.998
2.013
2
n.
iter.
10
9
7
34
41
18
Table 2.9: Test 3 on Plate 1: final estimates and iterations
dev.
d
Initial
B
B
B
MVE
MVE
MVE
Target
1/3
1/6
1/10
1/3
1/6
1/10
Ea
(GPa)
130
148.9
147.7
146.5
150.0
149.9
149.8
150
Eb
(GPa)
11
10.12
10.21
10.28
9.999
10.00
10.01
10
νab
0.20
0.2122
0.2099
0.2104
0.2509
0.2458
0.2382
0.25
Gab
(GPa)
3.5
3.002
3.004
3.007
3.000
3.000
3.000
3
Gac
(GPa)
3.5
3.113
3.234
3.366
2.999
3.006
3.015
3
Gbc
(GPa)
1.5
1.832
1.695
1.602
2.001
1.997
1.991
2
n.
iter.
6
6
5
20
19
56
2.4 Bayesian estimation of laminates’ elastic constants
89
Table 2.10: Test 1 on Plate 2: final estimates and iterations
dev.
d
Initial
B
B
B
MVE
MVE
MVE
Target
1/3
1/6
1/10
1/3
1/6
1/10
Ea
(GPa)
280
205.7
198.1
195.64
210.2
201.8
207.6
207
Eb
(GPa)
28
19.98
19.68
19.62
19.97
19.82
19.99
20
νab
0.19
0.2626
0.2656
0.2660
0.2393
0.5290
0.3296
0.25
Gab
(GPa)
15
9.914
9.8723
9.845
9.932
9.910
9.925
10
Gac
(GPa)
15
10.21
11.88
12.67
9.497
10.24
9.751
10
Gbc
(GPa)
8
4.090
4.298
4.408
4.159
4.086
4.075
4
n.
iter.
5
5
7
4
139
197
Table 2.11: Test 2 on Plate 2: final estimates and iterations
dev.
d
Initial
B
B
B
MVE
MVE
MVE
Target
1/3
1/6
1/10
1/3
1/6
1/10
Ea
(GPa)
160
275.6
296.3
197.2
229.5
207
Eb
(GPa)
26
20.38
23.82
20.48
19.93
20
νab
0.31
0.5312
0.3084
0.4001
0.2837
0.25
Gab
(GPa)
7
11.13
11.09
9.861
9.968
10
Gac
(GPa)
7
2.988
2.362
4.627
3.869
10
Gbc
(GPa)
6
9.804
9.393
7.426
13.11
4
n.
iter.
11
12
8
82
Table 2.12: Test 3 on Plate 2: final estimates and iterations
dev.
d
Initial
B
B
B
MVE
MVE
MVE
Target
1/3
1/6
1/10
1/3
1/6
1/10
Ea
(GPa)
160
210.8
215.7
216.1
206.9
206.8
210.2
207
Eb
(GPa)
14
19.95
19.82
19.41
20.00
20.00
19.97
20
νab
0.31
0.2260
0.2211
0.2235
0.2364
0.2235
0.2216
0.25
Gab
(GPa)
7
10.01
9.927
9.862
10.00
10.00
9.933
10
Gac
(GPa)
7
9.418
8.651
8.316
10.05
10.08
9.517
10
Gbc
(GPa)
6
4.077
4.368
4.896
4.002
4.005
4.164
4
n.
iter.
6
6
5
66
13
18
90
Modelling of fibre-reinforced composites
C11
C12
C22
C44
C55
C66
normalized sensitivity
1
0.8
0.6
0.4
0.2
0
1
2
3
4
5
6
7
8
9
10 11 12 13 14
mode
Figure 2.21: Forward sensitivity analysis for Plate 1
moduli Ea and Eb and the in-plane shear modulus Gab . This means that the
experimental data contain more information on C̄11 , C̄22 and C̄66 rather than
on C̄44 , C̄55 and C̄12 . As a consequence, the transverse shear moduli Gbc and
Gac and Poisson’s ratio νab are expected to be more difficult to estimate with
respect to the in-plane elastic constants.
The sensitivities are also influenced by the plate aspect ratio. Plate 1
(L/W = 2) is more sensitive to C̄12 , which accounts for directional effects.
On the other hand, Plate 2 (L/W = 1) is less sensitive to C̄11 , since the modulus Eb is much lower than Ea and, consequently, most low frequency modes
involve the b direction. Therefore the geometry of Plate 1 is preferable as it allows a more effective overall identification of the elastic constants. This should
be taken into consideration when setting up the experiment. More details on
optimal experiment design can be found in the work by Frederiksen (1998).
Some of the quantities which enter the estimation process should be defined
by the analyst. These are the initial parameter estimates µ and the covariance
matrices Vβ and VY . Obviously, convergence of the iterative procedures (2.147)
and (2.155) strongly depends upon these choices. The covariance matrices influ-
2.4 Bayesian estimation of laminates’ elastic constants
C11
C12
C22
C44
C55
C66
1
normalized sensitivity
91
0.8
0.6
0.4
0.2
0
1
2
3
4
5
6
7
8
9 10 11 12 13 14
mode
Figure 2.22: Forward sensitivity analysis for Plate 2
ence the amount of parameter correction at each iteration. High confidence on
the initial estimates (that is, small entries in Vβ ) results in small corrections,
while low confidence on the initial estimates may lead to large corrections, especially in the first step. On the other hand, the smaller the covariance VY
associated with the experimental values, the larger is the parameter correction.
Thus, for convergence reasons, Vβ should not be too large and VY not too
small.
Another important choice is the combined definition of the initial estimates
for Ea and Eb . In particular, for initial value sets with Ea overestimated and
Eb underestimated, or viceversa, the estimation procedures tend to fail. An
explanation can be given by considering the sensitivity coefficients in Equation
(2.165). Both si1 and si3 are positive for every mode. As a consequence, initial values for C̄22 higher than the real ones result in model frequencies higher
than the real ones. Analogously, initial values for C̄11 lower than the real ones
result in model frequencies lower than the real ones. These two opposite effects
are both strong, because the sensitivity coefficients are high, and may engender an oscillatory behaviour of the stepwise corrected parameters with loss of
92
Modelling of fibre-reinforced composites
convergence.
As already pointed out, the difference between the two estimators, from the
analytical point of view, lies on the role of the initial estimates. In the B estimator, the initial values are fixed, while in the MVE they serve as starting values
of the iterative scheme, which are updated at each iteration. As a consequence,
the dependence of the final estimates upon the initial guesses is very different
in the two estimators (see Tables 2.7-2.12).
If the algorithm converges to the global minimum, the MVE estimates depend only upon the information available in the experimental data: high sensitivity to the parameters implies that the target values are obtained. In this
case, any difference between final estimates and target values is due to the stop
criterion. On the contrary, different initial values or different deviations yield
significantly different estimates using the B algorithm. The bias is stronger as
the deviation becomes smaller and the parameters are forced to remain close to
the initial guesses. A clear example of this behaviour can be seen by observing
Tables 2.8 and 2.9: the Ea estimates are included in the interval between the
initial and target values and move closer to the initial guesses as the deviation
d becomes smaller. On the other hand, the B estimator is much more efficient than the MVE estimator, as clearly revealed by the number of iterations
performed.
2.4.2
A modified procedure
Here, a two stages solution strategy is presented aiming at improving convergence of the MVE estimator. The modified MVE estimator is denoted by MVEmod. The idea stems from the observation that there are two sets of parameters
with very different influence on the natural frequencies. In fact, as noticed in
the previous section, the eigenvalues are very sensitive to C̄11 , C̄22 and C̄66 and
much less to C̄12 , C̄44 and C̄55 (see Figures 2.21 and 2.22). For this reason, the
correction of the last three stiffness coefficients is much slower. This suggests
to split the MVE estimation procedure into two stages as follows.
• In the first stage, the estimation procedure is carried out using small
values for the parameter deviations, in order to avoid the risk of non
convergence. The tolerance is set to a value larger than the prescribed one:
TOL = 10−2 . This stage is designed for the parameters most sensitive to
2.4 Bayesian estimation of laminates’ elastic constants
93
Table 2.13: Test 1 on Plate 1: MVE-mod final estimates and iterations
dev.
d
Initial
stage 1
stage 2
stage 1
stage 2
Target
1/6
1/3
1/10
1/3
Ea
(GPa)
100
151.2
150.0
149.2
150.0
150
Eb
(GPa)
7
9.868
10.00
10.13
10.00
10
νab
0.33
0.3129
0.2507
0.2080
0.2492
0.25
Gab
(GPa)
4
2.998
3.000
3.002
3.000
3
Gac
(GPa)
4
2.892
2.999
3.083
3.001
3
Gbc
(GPa)
1
2.196
2.001
1.815
1.999
2
n.
iter.
41
16
9
17
Table 2.14: Test 2 on Plate 1: MVE-mod final estimates and iterations
dev.
d
Initial
stage 1
stage 2
stage 1
stage 2
stage 1
stage 2
Target
1/3
1/2
1/6
1/3
1/10
1/3
Ea
(GPa)
180
149.9
150.0
149.7
150.0
150.4
150.0
150
Eb
(GPa)
12
10.01
10.00
10.02
10.00
9.978
10.00
10
νab
0.20
0.2416
0.2498
0.2234
0.2492
0.2635
0.2506
0.25
Gab
(GPa)
2
3.000
3.000
3.001
3.000
2.999
3.000
3
Gac
(GPa)
2
3.011
3.000
3.024
3.001
2.958
2.999
3
Gbc
(GPa)
4
1.994
2.000
1.987
1.999
2.030
2.000
2
n.
iter.
24
7
9
15
14
11
the frequency data (typically Ea , Eb and Gab ), which quickly converge.
• In the second stage, the initial values are taken as the final estimates of the
first stage, but the deviations associated to the parameters are increased.
This is expected to accelerate convergence of the remaining parameters,
without interfering with the estimation of the parameters which have already reached convergence during the first stage. The tolerance is set to
the prescribed value: TOL = 10−3 .
To show the advantages of the MVE-mod procedure, the same numerical
tests discussed in the previous section are considered. The results are shown in
Tables 2.13-2.18. As it can be seen from the last column, the final estimates
are very similar to the ones predicted by the classical MVE procedure, but
convergence is faster.
94
Modelling of fibre-reinforced composites
Table 2.15: Test 3 on Plate 1: MVE-mod final estimates and iterations
dev.
d
Initial
stage 1
stage 2
stage 1
stage 2
stage 1
stage 2
Target
1/3
1/2
1/6
1/3
1/10
1/3
Ea
(GPa)
130
150.1
150.0
150.0
150.0
149.5
150.0
150
Eb
(GPa)
11
9.990
10.00
10.00
10.00
10.08
10.00
10
νab
0.20
0.2598
0.2503
0.2407
0.2491
0.2214
0.2491
0.25
Gab
(GPa)
3.5
3.000
3.000
3.000
3.000
3.001
3.000
3
Gac
(GPa)
3.5
2.987
3.000
3.004
3.001
3.054
3.001
3
Gbc
(GPa)
1.5
2.008
2.000
1.996
1.999
1.893
1.999
2
n.
iter.
9
7
5
10
7
17
Table 2.16: Test 1 on Plate 2: MVE-mod final estimates and iterations
dev.
d
Initial
stage 1
stage2
stage 1
stage 2
stage 1
stage 2
Target
1/3
1/2
1/6
1/3
1/10
1/3
Ea
(GPa)
280
210.1
210.2
201.7
201.7
206.7
208.0
207
Eb
(GPa)
28
19.97
19.97
19.81
19.82
19.99
19.99
20
νab
0.19
0.2394
0.2392
0.3299
0.5467
0.2569
0.4095
0.25
Gab
(GPa)
15
9.932
9.932
9.905
9.911
9.920
9.927
10
Gac
(GPa)
15
9.500
9.497
10.87
10.19
10.06
9.523
10
Gbc
(GPa)
8
1.459
4.159
4.148
4.080
4.083
4.056
4
n.
iter.
3
1
24
30
8
61
Table 2.17: Test 2 on Plate 2: MVE-mod final estimates and iterations
dev.
d
Initial
stage 1
stage 2
Target
1/10
1/3
Ea
(GPa)
160
227.7
230.0
207
Eb
(GPa)
26
20.16
19.93
20
νab
0.31
0.3298
0.2795
0.25
Gab
(GPa)
7
9.994
9.968
10
Gac
(GPa)
7
3.897
3.869
10
Gbc
(GPa)
6
10.74
13.15
4
n.
iter.
24
12
2.4 Bayesian estimation of laminates’ elastic constants
95
Table 2.18: Test 3 on Plate 2: MVE-mod final estimates and iterations
dev.
d
Initial
stage 1
stage 2
stage 1
stage 2
stage 1
stage 2
Target
2.4.3
1/3
1/2
1/6
1/3
1/10
1/3
Ea
(GPa)
160
206.5
207.0
207.4
206.9
212.1
210.0
207
Eb
(GPa)
14
20.00
20.00
20.00
20.00
19.96
19.97
20
νab
0.31
0.2140
0.2452
0.2223
0.2393
0.2199
0.2227
0.25
Gab
(GPa)
7
9.998
10.00
10.00
10.00
9.944
9.931
10
Gac
(GPa)
7
10.15
10.02
9.981
10.04
9.254
9.547
10
Gbc
(GPa)
6
4.003
4.001
4.011
4.002
4.184
4.161
4
n.
iter.
5
44
8
44
8
4
Testing of the estimator procedures
Various case studies are solved using the B, MVE and MVE-mod procedures.
Two cross-ply laminates, having the same material properties and stacking sequence but different aspect ratio, are considered in the first case study. Initially,
the estimation is carried out using the natural frequencies obtained numerically,
then a real experiment is simulated by using pseudo-experimental input frequencies. In the second case study, a cross-ply and an angle-ply laminates are
taken into consideration to investigate the effect of the stacking sequence; in
this example, the input frequencies are again assumed as equal to the numerical
frequencies, so to disregard any sources of error. Finally, the third case study
deals with a single-layer carbon-epoxy plate studied by Frederiksen (1997). In
this case, the input frequencies are real experimental measurements, so that
both modelling and measurement errors are present.
In stage 2 of the MVE-mod procedure, the deviations are twice those in the
first stage.
Cross-ply symmetric laminate
Two cross-ply symmetric laminates with stacking sequence (0/90/90/0) and different geometry were considered. Plate 1c is square, while Plate 2c is rectangular
with sides ratio 1 : 2. The plates were discretised using 20×20 and 20×10 finite
elements, respectively, their characteristics are reported in Table 2.19. Figure
2.23 represents the natural modes associated to the considered frequencies for
the rectangular plate. The input frequencies are reported in Table 2.20 and the
96
Modelling of fibre-reinforced composites
Mode 1
Mode 2
Mode 5
Mode 6
Mode 9
Mode 10
Mode 13
Mode 3
Mode 7
Mode 11
Mode 4
Mode 8
Mode 12
1
Mode 14
Figure 2.23: Natural modes corresponding to the frequencies considered in the
analysis (Plate 2c)
results for Plates 1c and 2c are reported in Table 2.21.
In Plate 1c, only the elastic constants Ea and Gab were evaluated within less
than 10% of the target values, while the other parameters were not estimated
correctly. On the other hand, in Plate 2c much better results were obtained,
especially using the MVE and MVE-mod estimators. These results show once
again the importance of the choice of the experiment, as already pointed out in
the examples concerning single layer orthotropic plates. As regards the behaviour of the different algorithms, in Plate 2c the MVE and MVE-mod algorithms
yield the target values, while the B estimator still shows a bias which is higher
for the parameter for which information is less. Also, it can be seen that the
MVE-mod requires less iterations than the original MVE.
The elastic constants for Plate 2c are estimated using pseudo-experimental
2.4 Bayesian estimation of laminates’ elastic constants
97
Table 2.19: Characteristics of Plate 1c and Plate 2c
length L
100 mm
100 mm
Plate 1
Plate 2
width W
100 mm
50 mm
mean
thickness h
4 mm
4 mm
density ρ
1500 kg/m3
1500 kg/m3
stacking
sequence
[0/90/90/0]
[0/90/90/0]
Table 2.20: Input frequencies of Plates 1c and 2c
Plate 1
mode
1
2
3
4
5
6
7
Plate 2
fnum
546.46
1482.1
1832.9
3269.4
3416.4
3961.3
4105.3
mode
8
9
10
11
12
13
14
fnum
4245.4
5878.9
7428.7
7651.2
7698.1
7792.5
8242.7
mode
1
2
3
4
5
6
7
fnum
1074.4
3268.5
3820.8
5704.8
6037.4
7544.4
7688.2
fpseudo
1089.6
3234.0
3830.7
5613.3
6027.9
7542.9
7595.8
mode
8
9
10
11
12
13
14
fnum
8095.6
10717
11476
12677
12710
12994
14145
fpseudo
8192.0
10788
11624
12767
12814
13084
14089
Table 2.21: Final estimates and iterations for Plates 1c and 2c
Plate 1c
Initial
deviation
B
MVE
MVE-mod
1
2
Target
Plate 2c
Initial
deviation
B
MVE
MVE-mod
Target
1
2
Ea
(GPa)
150
18
124.4
125.5
125.6
125.5
125
Eb
(GPa)
8
3
4.459
4.153
4.167
4.152
5
Ea
(GPa)
150
18
123.6
125.0
125.0
125.0
125
Eb
(GPa)
8
3
4.079
5.000
4.993
5.000
5
νab
0.3
0.1
0.5441
0.7757
0.7479
0.7757
0.25
νab
0.3
0.1
0.5353
0.2535
0.2718
0.2508
0.25
Gab
(GPa)
5
0.5
2.512
2.489
2.489
2.489
2.5
Gac
(GPa)
5
0.5
4.263
4.195
4.176
4.200
2.5
Gbc
(GPa)
3
0.5
0.4114
0.3602
0.3662
0.3591
1
n.
iter.
Gab
(GPa)
5
0.5
2.507
2.500
2.500
2.500
2.5
Gac
(GPa)
5
0.5
3.564
2.500
2.500
2.500
2.5
Gbc
(GPa)
3
0.5
0.7374
0.9999
0.9995
1.000
1
n.
iter.
5
20
12
5
9
64
38
13
98
Modelling of fibre-reinforced composites
Table 2.22: Final estimates and iterations for Plate 2c - pseudoexperimental
frequencies
Initial
deviation
B
MVE
MVE-mod
Target
1
2
Ea
(GPa)
150
18
119.8
120.8
121.1
120.8
125
Eb
(GPa)
8
3
4.100
4.830
4.582
4.830
5
νab
0.3
0.1
0.5469
0.4371
0.6817
0.4395
0.25
Gab
(GPa)
5
0.5
2.581
2.572
2.612
2.572
2.5
Gac
(GPa)
5
0.5
3.739
2.668
2.673
2.668
2.5
Gbc
(GPa)
3
0.5
0.7843
1.041
0.9658
1.041
1
n.
iter.
7
48
17
11
frequencies to simulate a real experimental set of data. Pseudo-experimental
input data are generated by introducing some error on the natural frequencies
obtained numerically. Each numerical frequency fnum is substituted by a random value fpseudo extracted from a Gaussian probability distribution with mean
fnum and deviation δf . As already stated, the deviation δf for each frequency
is taken proportional to the frequency itself. This represents the fact that higher
natural frequencies are usually affected by more experimental error that lower
frequencies. The results are reported in Table 2.22.
Comparing the final estimates in Tables 2.21 and 2.22, it can be seen that
the error introduced in the input frequencies reflects on the results obtained by
all procedures, affecting particularly the parameters for which little information
is available, such as νab . Besides that, the MVE and MVE-mod estimators yield
analogous results, while the B estimator shows the bias already noticed in the
tests carried out using the numerically generated frequencies.
Cross-ply versus angle-ply non symmetric laminate
A cross-ply and an angle-ply laminate, having the same geometrical and mechanical characteristics, are considered in this case study. The specimens characteristics are reported in Table 2.23 and the input natural frequencies computed
by a uniform mesh of 20 × 20 elements are reported in Table 2.24. Table 2.25
collects the results for the three procedures. The more complicated stacking
sequence of the angle-ply laminate increases the difficulties in the identification
of the in-plane elastic constant. In the B estimator, this is reflected in final
2.4 Bayesian estimation of laminates’ elastic constants
99
Table 2.23: Characteristics of cross-ply and angle-ply plates
length L
100 mm
100 mm
width W
50 mm
50 mm
thickness h
3.3 mm
3.3 mm
density ρ
1500 kg/m3
1500 kg/m3
stacking sequence
[(0/90)8 0(90/0)8 ]
[(0/45/90/ − 45)4 0(45/90/ − 45/0)4 ]
Table 2.24: Input frequencies for cross-ply and angle-ply plates
Cross-ply
mode no.
1
2
3
4
5
6
7
fexp
1369.7
2335.6
3598.6
6255.1
7390.8
8372.0
8755.9
mode no.
8
9
10
11
12
13
14
Angle-ply
mode no.
1
2
3
4
5
6
7
fexp
10103
11779
12755
12872
16071
17280
18504
fexp
2074.4
2250.7
4956.8
5602.9
7456.6
8609.7
8613.2
mode no.
8
9
10
11
12
13
14
fexp
10657
11641
13459
15491
17182
19113
19121
Table 2.25: Final estimates and iterations for cross-ply and angle-ply plates
Cross-ply
Initial
deviation
B
MVE
MVE-mod
1
2
Target
Angle-ply
Initial
deviation
B
MVE
MVE-mod
Target
1
2
Ea
(GPa)
150
17.5
120.1
125.0
124.9
125.0
125
Eb
(GPa)
17
4
13.69
9.458
9.449
9.491
9.5
Ea
(GPa)
150
17.5
119.0
124.9
124.3
125.0
125
Eb
(GPa)
17
4
12.84
9.306
8.653
9.469
9.5
νab
0.25
0.05
0.2812
0.3058
0.3306
0.3013
0.30
νab
0.25
0.05
0.3231
0.3233
0.4175
0.3036
0.30
Gab
(GPa)
8
1.5
5.989
6.000
5.999
6.000
6
Gac
(GPa)
8
1.5
6.206
5.996
5.983
5.999
6
Gbc
(GPa)
5
1.5
3.206
3.000
2.985
3.000
3
n.
iter.
Gab
(GPa)
8
1.5
6.943
6.103
6.504
6.017
6
Gac
(GPa)
8
1.5
6.178
5.998
5.983
6.000
6
Gbc
(GPa)
5
1.5
3.179
2.996
2.986
2.999
3
n.
iter.
3
39
10
13
3
94
9
34
100
Modelling of fibre-reinforced composites
estimates that are more influenced by the initial values. In the MVE and MVEmod procedures, the number of iterations required increases, but the accuracy
of the final estimates is not significantly modified.
Single-layer carbon-epoxy plate
A single-layer unidirectional carbon-epoxy (T300 carbon fibre) composite plate
is considered. The elastic constants were estimated by Frederiksen (1997) using
a two step approach. Firstly, a thin specimen was considered for the estimation
of the four in-plane elastic constants. Subsequently, a second thicker specimen
was taken in order to estimate the two transverse shear moduli. In this work,
only the second specimen is taken into consideration in order to simultaneously
estimate all the six elastic constants. The known specimen characteristics are
reported in Table 2.26 and the measured natural frequencies are reported in
Table 2.27.
The initial set of estimates needed for the estimation process are obtained
through the rule of mixtures, with a fibre volume fraction vf of 60%. Producers’
technical sheets provide the following data:
carbon fibre T300
Ef = 230 − 240 GPa
νf = 0.27
epoxy resin
Em = 7 GPa
νm = 0.36
where the subscripts f and m stand for fibre and matrix elastic constants. The
values of E and ν allow to evaluate Gf and Gm , then the rule of mixtures can
be applied:
Ea
=
Eb
=
νab
=
Gab = Gac
=
Ef vf + Em vm = 144 GPa
Ef Em
= 16.5 GPa
Ef vm + Em vf
νf vf + νm vm = 0.30
Gf Gm
= 6 GPa
Gf vm + Gm vf
2.4 Bayesian estimation of laminates’ elastic constants
101
where vm is the matrix volume fraction. The initial value for Gbc is taken as
4 GPa. The plate is analysed by a 50 × 50 element mesh, which has been
verified to satisfactorily resolve the first 14 natural modes.
The results of the estimation process are collected in Table 2.28, together
with the assumed deviations. As it can be observed, they are in good agreement
with the values found by Frederiksen (1997).
Comparing the estimators, one notices once again that the B estimator exhibits some bias in the identification of the constants νab , Gac and Gbc , because
of the less sensitivity of the natural frequencies. On the contrary, this is not
observed in the MVE estimator, which yields quite good estimates.
Table 2.26: Characteristics of carbon-epoxy plate
length L
100.1 mm
width W
53.0 mm
mean
thickness h
3.36 mm
density ρ
1537 kg/m3
stacking
sequence
[0]
fibre volume
fraction
60%
Table 2.27: Measured frequencies for carbon-epoxy plate
mode no.
1
2
3
4
5
6
7
fexp
1121.4
2795.4
2971.5
3590.6
3633.1
5947.5
7655.1
mode no.
8
9
10
11
12
13
14
fexp
7671.4
8073.0
8431.8
10212
10281
14007
14048
Table 2.28: Estimates for carbon-epoxy plate
Initial
deviation
B
MVE
Frederiksen (1997)
Ea
(GPa)
144
15
113.5
113.3
113.0
Eb
(GPa)
16.5
3.5
8.47
8.52
8.50
νab
0.30
0.05
0.361
0.323
0.323
Gab
(GPa)
6
1.5
4.44
4.45
4.45
Gac
(GPa)
6
1.5
4.31
4.42
4.43
Gbc
(GPa)
4
1.5
3.39
2.91
2.97
102
Modelling of fibre-reinforced composites
Capitolo 3 – Sommario
103
Le leghe a memoria di forma
Le leghe a memoria di forma sono materiali in grado di recuperare la forma
originale in seguito a grandi deformazioni, secondo due diverse modalità:
• se il materiale è deformato a bassa temperatura, il recupero avviene in
seguito al riscaldamento sopra una temperatura critica. Questa proprietà
è nota con il nome di effetto a memoria di forma;
• se il materiale è deformato ad alta temperatura, il recupero avviene in
seguito alla semplice rimozione del carico applicato. Questa proprietà è
nota con il nome di superelasticità.
L’effetto a memoria di forma fu osservato per la prima volta negli anni ’30 su
leghe di oro e cadmio. L’interesse in questi materiali si diffuse soprattutto a partire dagli anni ’60, quando le proprietà di recupero della forma furono osservate
in una lega di nichel e titanio, ribattezzata Nitinol dal nome del laboratorio
dove la scoperta fu effettuata (Nichel Titanium Naval Ordnance Laboratory).
Nonostante la presenza di proprietà a memoria di forma in altre leghe, il Nitinol
è ancora oggi la lega a memoria di forma più diffusa commercialmente.
Le proprietà macroscopiche dei materiali a memoria di forma sono legate a cambiamenti microstrutturali che avvengono all’interno del materiale. La
struttura e l’organizzazione dei cristalli all’interno del materiale sono infatti
influenzate dalla temperatura e dallo sforzo applicato, dando origine ad un significativo accoppiamento termomeccanico. Le proprietà e le caratteristiche
microstrutturali dei materiali a memoria di forma sono discusse nella sezione
3.1.
I materiali a memoria di forma mostrano il comportamento accoppiato tipico dei materiali attivi. In particolare, l’effetto a memoria di forma associato al riscaldamento può essere utilizzato per realizzare attuatori, mentre sono
stati portati avanti alcuni tentativi di realizzare sensori basati sulle proprietà
associate alle diverse microstrutture. D’altra parte, il recupero superelastico
della forma non richiede alcun intervento esterno; per questo la superelasticità
è spesso sfruttata in applicazioni passive. Nella sezione 3.2 sono discusse alcune
applicazioni delle leghe a memoria di forma.
La modellazione dei materiali attivi, in particolare delle leghe a memoria di
forma, costituisce un problema ancora aperto. In letteratura sono stati proposti
104
Capitolo 3 – Sommario
numerosi modelli costitutivi, ognuno dei quali descrive solo alcuni degli aspetti
del complesso comportamento termomeccanico dei materiali a memoria di forma. Inoltre, non tutti gli aspetti del comportamento delle leghe a memoria di
forma sono, ad oggi, stati chiariti. Nella sezione 3.3 si propone una discussione
critica di diversi modelli costitutivi monoassiali e triassiali per le leghe a memoria di forma, mentre nella sezione 3.4 sono riportati e discussi alcuni risultati
sperimentali che mostrano comportamenti non descritti dai modelli costitutivi
correnti.
Chapter 3
Shape memory alloys
Shape memory alloys (SMA) are metal alloys able to recover their original shape
after large deformations (Otsuka and Wayman, 1999). Depending upon the alloy
material properties and the external conditions, shape recovery can occur in two
ways:
• if the material is deformed at low temperature, its original shape can be
recovered by heating it above a characteristic temperature. This property
is known as the shape memory effect (SME);
• if the material is deformed at high temperature, its original shape can be
recovered by simply removing the applied load. This property is known
as superelasticity or pseudoelasticity (SE).
The shape memory effect was first observed in gold-cadmium alloys (Chang,
1951; Ölander, 1932). Widespread interest on shape memory materials, however, had to wait until the 1960s, when shape memory properties were observed
in a nickel-titanium alloy (Buehler et al., 1963). From the name of the laboratory where these effect were observed, the Ni-Ti alloy became known as Nitinol
(Nickel Titanium Naval Ordnance Laboratory) and it is still the most common
commercially available shape memory alloy. Various other materials are known
to display shape memory properties, such as ternary nickel-titanium alloys (NiTi-Cu, Ni-Ti-Fe, etc.) and copper based alloys (Cu-Zn, Cu-Al-Ni, etc.).
106
Shape memory alloys
The macroscopic properties of shape memory alloys are related to changes in the material microstructure. Indeed, the crystal structure and arrangement within these materials are influenced by both temperature and applied
stress, resulting in a strong thermomechanical coupling. The shape memory
alloy properties and microstructure are discussed in Section 3.1.
Shape memory alloys exhibit the coupled behaviour typical of active materials. In particular, the shape recovery associated to heating typical of the shape
memory effect can be used for actuation, while some attempts have been carried
out to use the property changes associated to the different microstructures for
sensing. On the other hand, superelastic shape recovery involves no temperature change, therefore superelasticity is generally used in passive applications.
Some applications of shape memory alloys are discussed in Section 3.2.
The modelling of active materials, and in particular of shape memory alloys, is still an open problem. Indeed, a number of constitutive models have
been proposed in the literature, each describing different characteristics of the
complex shape memory alloy behaviour. Moreover, some features of the shape
memory alloy material behaviour are not fully understood to this day. A critical
discussion of various one and three-dimensional shape memory alloys constitutive models is carried out in Section 3.3, while in Section 3.4 some experimental
observation are presented and discussed regarding phenomena which are not
included in the current constitutive modelling.
3.1
Shape memory alloy properties
Shape memory alloys exhibit two peculiar properties, known as the shape memory effect and superelasticity. If the material is deformed at low temperature,
upon unloading it retains an apparently plastic deformation, which can be recovered by heating it above a characteristic temperature (thermal recovery, known
as the shape memory effect). At high temperature, on the other hand, large
deformations can be recovered by simply removing the applied load (mechanical
recovery, known as superelasticity).
The shape memory effect is depicted in Figure 3.1 in the stress-straintemperature space. The material starts from the reference configuration 1 at a
given temperature and zero stress. If the stress is increased, the material behaviour is at first linearly elastic (slope 1–2), then apparently plastic deformations
3.1 Shape memory alloy properties
107
4
s
3
2
e
1
5
7
6
T
stress
heat
cool
Figure 3.1: Shape memory effect
s
3
4
2
5
6
1
e
stress
remove stress
Figure 3.2: Superelasticity
108
Shape memory alloys
s
SE
SME
T
e
T
Figure 3.3: Shape memory effect and superelasticity
start to develop at nearly constant stress (plateau 2–3). Unlike the plastic deformation, however, the shape memory alloy deformation reaches saturation and
further increase of the load leads to a new linear elastic branch (slope 3–4).
Upon unloading, a residual deformation is present (point 5), which can be recovered by heating the material to a characteristic temperature under no load
(points 6–7). Finally, if the material is cooled down to the initial temperature,
the initial material state 1 is completely recovered.
Superelasticity is depicted in the stress-strain space in Figure 3.2. The loading branch is similar to the one described for the shape memory effect (points
1–4), whereas a stress plateau is present upon unloading (points 5–6), leading to
no residual deformation at zero stress. Due to the presence of a small hysteresis
loop between the loading and unloading plateau, this property is often referred
to as pseudoelasticity.
These apparently unrelated properties can be observed in the same material
if it is tested at different temperatures (Figure 3.3). Indeed, the characteristic
temperature that separates the shape memory alloy and superelasticity temperature ranges depends upon the alloy type and chemical composition, thus it is
possible to find alloys which show either of the effects at room temperature.
The thermal shape recovery associated to shape memory effect can be used
for actuation purposes. Indeed, if shape recovery occurs under an applied load,
mechanical work is developed. Figure 3.4 depicts a very simple experiment: a
3.1 Shape memory alloy properties
109
heat
cool
Figure 3.4: Shape recovery under applied load
R
Flexinol°
wire 0.15 mm in diameter was able to lift a mass of 200 g by 10
mm when electrically heated. On the other hand, if the shape memory alloy is
completely constrained and prevented from recovering its original shape, upon
heating it develops high recovery stresses which depend upon the temperature
and initial deformation. This phenomenon is usually referred to as constrained
recovery and it is very important in applications such as shape memory alloy hybrid composites (discussed in Chapter 4), as in that case the composite
constitutes a constraint to the shape memory alloy wires.
The described shape memory effect is also known as one-way shape memory
effect, as only the high-temperature shape is memorised by the material. Indeed,
no change in shape occurs upon cooling under no applied load (Figure 3.1, points
7–1). The two-way memory effect, on the other hand, is related to the ability of
the material to remember two different shapes in the cold and hot configuration.
This property allows to move between the two shapes by simply changing the
temperature, with no need of applied load. The two-way shape memory effect
is schematically represented in Figure 3.5(b). It should be noted that, while
the one-way shape memory effect is an intrinsic characteristic of shape memory
110
Shape memory alloys
stress
heat
heat
cool
cool
(a) One-way shape memory effect
(b) Two-way shape memory effect
Figure 3.5: One-way and two-way shape memory effect
alloys, two-way memory needs to be induced by particular treatments, such as
training. Training, that is repeatedly cycling the material in different ways, is
also useful to stabilise the shape memory alloys properties.
3.1.1
Martensitic phase transitions
The shape memory alloys macroscopic behaviour depends upon the presence
within the material of two solid phases with different degrees of symmetry,
which can transform one into the other due to a change in temperature and
stress conditions.
The parent phase, austenite, has a high symmetry crystal structure (typically
body centred cubic) and it is stable at high temperature and low levels of stress.
Martensite, on the other hand, has lower symmetry (typically rhombohedral or
monoclinic), thus it can appear in a number of cristallographically equivalent
variants, which differ in their orientation with respect to the material axes. The
different martensitic variants formed in a single grain of polycrystalline shape
memory alloy can be observed in Figure 3.6. The phase transition between the
two crystal structures is displacive, thus it requires only small movements of the
atoms within the crystal lattice and its evolution is time-independent1 .
Figure 3.7 depicts the different crystal structures and arrangements in the
appropriate zones of the uniaxial stress-temperature plane. The forward austenite to martensite phase transition is exothermic and can be triggered by a de1 Diffusive phase transitions, on the other hand, require breaking of the crystal lattice and
long distance atom movements, thus their evolution depends upon the time of reaction.
3.1 Shape memory alloy properties
111
Figure 3.6: Formation of different martensitic variants in a single grain of polycrystalline CuZnAl alloy during uniaxial tensile loading (Patoor et al., 2006)
detwinned
martensite
d
s
M
A
M
t
«
®
d
M
t
M «A
twinned
martensite
austenite
T
Figure 3.7: Crystal structures and arrangements in the stress-temperature plane
112
Shape memory alloys
G
A®M
DG
M®A
DG
Ms
T0
As
T
Figure 3.8: Free energy of austenite and martensite as a function of temperature
crease in temperature or by an applied load. If the transformation is purely
temperature-induced (Figure 3.7, M t ↔ A), the martensite variants rearrange
themselves to accommodate the macroscopic austenitic shape (twinned martensite). In the presence of an applied load (Figure 3.7, M d ↔ A), the growth of
some martensite variants is favoured and a macroscopic deformation is associated to the phase transition (detwinned martensite). An applied load can also
cause martensite reorientation, resulting in detwinning of twinned martensite
(Figure 3.7, M t → M d ) or change in the orientation of detwinned martensite
with a change in the direction of the applied load. It should be clear from this
description that the low symmetry martensite crystals can be arranged in different ways, depending upon the material stress and temperature history. The
microscopic martensite crystal arrangements may correspond to a macroscopic
change with respect to the austenitic shape. In the reverse phase transition, on
the other hand, all martensite variants transform into the only possible austenite
crystal structure, thus recovering the original memorised shape.
The forward and reverse phase transitions do not occur in the same stress
and temperature conditions. Indeed, a hysteresis is present, leaving a range
of stresses and temperatures in which the two phases can coexist. Figure 3.8
3.1 Shape memory alloy properties
113
shows a schematic representation of the Gibbs free energy of the two phases as
a function of temperature. Here, T0 denotes the equilibrium temperature, that
is the temperature at which the energy of both phases is the same. The energy
change in a system related to a martensitic transformation, ∆G, is composed of
a chemical term, related to the change in crystal structure, and a non chemical
term, related to the elastic and surface energies. For phase transition to occur,
the driving force for the transformation needs to overcome the non chemical
energy change. For this reason, forward transformation starts at a temperature
Ms , below T0 , and its driving force is denoted with ∆G A→M in Figure 3.8, while
reverse transformation starts at a temperature As , above T0 , and its driving
force is denoted with ∆G M →A in Figure 3.8.
For materials displaying the shape memory effect, the energy barrier to be
overcome for phase transition is quite small. This kind of martensitic transformations are usually defined as thermoelastic, as opposed to non thermoelastic
martensitic transformations occurring, for example, in steel. In that case, the
energy barrier for phase transition is much higher, thus the reverse martensite
to austenite phase transition cannot occur and an essential condition for the
display of shape memory properties is removed.
Some Ni-Ti alloys display a two step phase transition. Upon cooling, austenite transforms first into a trigonal structure known as the R-phase, which in turn
changes into martensite as the temperature continues to decrease. The austenite to R-phase transition has a very small hysteresis, it is associated to a limited
recoverable shape change (about 1% elongation versus the 10% elongation associated to martensite) and a significant change in electrical resistivity.
3.1.2
Micromechanical interpretation of shape memory alloy properties
The shape memory alloy behaviour can be interpreted in light of the described
phase transitions and reorientation of martensite variants. The loading paths
associated to the shape memory effect and superelasticity are depicted in Figures
3.9 and 3.10, respectively, while the numbers in the text refer to those reported
in Figures 3.1 and 3.2.
As the shape memory effect occurs at low temperatures, the material microstructure is initially composed of twinned martensite. Upon loading, the ap-
114
Shape memory alloys
s
detwinned
martensite
twinned
martensite
austenite
T
Figure 3.9: Loading path and phase transitions for the shape memory effect
s
detwinned
martensite
austenite
T
Figure 3.10: Loading path and phase transitions for superelasticity
3.2 Shape memory alloy applications
115
plied stress reaches the critical value for detwinning (point 2) and the martensite
reorientation starts, associated to a macroscopic deformation (plateau 2–3). At
the end of the plateau, the martensite is completely detwinned and further loading only causes elastic deformation of the new microstructure (slope 3–4). Upon
unloading, the macroscopic deformation is retained as all variants of martensite are equally stable (point 5). When the material is heated above a critical
temperature, the martensite to austenite phase transition starts, allowing to recover the macroscopic deformation (points 6–7). During cooling, the austenite
to martensite transition occurs under no load, thus no macroscopic change can
be observed (points 7–1).
In superelasticity, the material microstructure is initially austenitic. During loading, the critical stress for phase transition is reached (point 2) and the
material transforms directly into detwinned martensite (plateau 2–3). As described in the previous paragraph, once the phase transition is complete further
loading only causes elastic deformation of the detwinned martensite (slope 3–4).
Austenite is the only stable phase at high temperature and no stress, therefore
during unloading the critical stress for the reverse phase transition is reached
and the macroscopic deformation is recovered (plateau 5–6).
3.2
Shape memory alloy applications
Due to their good structural characteristics, typical of metals, and their unique
functional properties, shape memory alloys can be successfully employed in a
variety of applications. While biomedical devices based on both the shape memory effect and superelasticity are rather common nowadays, issues related to cost
and difficulties in modelling are still an obstacle to a widespread use of these
materials in other engineering fields. In the following, some shape memory alloy
applications are presented and briefly discussed. Applications involving shape
memory alloy elements embedded in a structure, on the other hand, are discussed in Chapter 4 and not dealt with here. For a review on shape memory
alloy applications, see Van Humbeeck (2001).
116
3.2.1
Shape memory alloys
Couplings and fasteners
The stress developed by shape memory alloys when shape recovery is prevented
(constrained recovery) can be used to build couplings and fasteners. Indeed, the
first large scale applications of the shape memory effect was a hydraulic coupling
used to join titanium tubing in the Grumman F-14 aircraft in 19712 (Otsuka and
Wayman, 1999). The simplest fastener is a metal ring supplied in the expanded
martensitic condition. When heated, the ring shrinks and applies a uniform
radial pressure to the elements to be fastened. The main advantages of such
devices are the reliability, simplicity of assembly (allowing their use in difficult to
access areas) and the possibility of joining different materials together. On the
other hand, they are generally more expensive than other alternative solutions
and their operation temperature range is limited by the onset of the martensitic
phase transition.
Another device exploiting recovery stress is the static rock breaker: compressed shape memory alloy elements are placed into boreholes and heated to
develop stresses of the order of hundreds of megapascals. The resulting procedure is safer, silent and produces no vibration or dust, moreover it can be used
in difficult environments such as water or other liquids.
3.2.2
Actuators
Thermal shape recovery can be used for actuation purposes3 , as anticipated in
the previous section. Shape memory alloy actuators are simple and lightweight
with respect to traditional actuators, such as motors, and offer a high power to
weight ratio. They can be designed without friction and vibrating parts, avoiding the production of dust particles and sparks, thus they are extremely silent
and apt to work in harsh environments. On the other hand, their efficiency
is quite low and their operating bandwidth is significantly limited by the time
required for cooling. Indeed, heating can be achieved quickly and easily, for example through Joule heating, but cooling is hardly controllable and may require
a long time, depending upon the size of the components. Two-way memory can
2 Nowadays, shape memory alloy fasteners are produced by various companies, such as
Intrinsic Devices Inc. (www.intrinsicdevices.com).
3 In Italy, shape memory actuators for industrial applications are produced by Saes Getters
S.p.A. (www.saesgetters.com).
3.2 Shape memory alloy applications
117
be used for actuation; alternatively, a bias element needs to be introduced to
restore the deformed position upon cooling.
The limited size and simplicity of actuation make shape memory alloy especially suitable for robotics applications. Cooling time is considerably shortened
in micro-actuators, improving the actuator bandwidth. High precision manipulation of micro-objects can be obtained with shape memory alloy microgrippers
(Krevet and Kohl, 2003). Compact and lightweight actuators whose movement
mimics biological tendons’ behaviour are an ideal candidate for the development
of prosthetic robot hands (Price et al., 2007).
Space applications, requiring a tight control of the weight of the apparatus to
be launched in orbit, benefit from the high power to weight ratio of shape memory alloy actuators. Moreover, shape memory alloy actuators allow to achieve
highly controllable and smooth movements, significantly reducing acceleration
and preventing unwanted movements in a zero gravity environment. Envisaged applications include release mechanisms (Fragnito and Vetrella, 2002) and
deployment of large lightweight structures (Pollard and Jenkins, 2005). A pioneering shape memory alloy space application was the actuation of a rotating
arm placed on the Mars Pathfinder, launched by NASA on December 4, 1996
(Jenkins and Landis, 1995).
Recently, the use of shape memory alloy actuators has started to penetrate
the automotive industry. In Italy, the Fiat research centre (CRF)4 is investigating a number of different actuator applications, including air conditioning,
locks release, rear-view mirrors positioning and anti-dazzle devices.
The applications described up to now require external heating of the shape
memory alloy element to trigger the actuation. A different concept employs the
temperature of the surrounding environment as the actuation mechanism. In
this case, shape memory alloys act as both sensors and actuators, for example
actuating a fire suppression system once the room temperature has reached a
critical level (Zhuiykov, 2008).
3.2.3
Biomedical devices
Biomedical devices based both on the shape memory effect and superelasticity
are the most common and commercially available applications of shape memory
4 www.crf.it
118
Shape memory alloys
alloys5 . Indeed, biomedical applications are less affected by the cost limitations
which are still preventing a wider diffusion of shape memory alloys in other engineering fields. The high biocompatibility of Nitinol allows to use it as implant
material. Moreover, the extremely stable body temperature can either ensure
a controllable behaviour of superelastic devices or be used as the mechanism to
trigger shape recovery.
Superelastic eyeglass frames are rather common nowadays. An important
feature of superelasticity is the unloading shape recovery, occurring at nearly
constant stress (Figure 3.2). Orthodontic wires providing a steady force regardless of large teeth movements are based on this principle, as well as orthopaedic
devices such as bone anchors and staples. Superelastic guidewires and surgical
instruments for microinvasive surgery allow miniaturisation, simplicity of design
and ease of sterilisation.
The best known biomedical application of shape memory alloys are stents.
These are small tubular shaped devices inserted to prevent occlusion of arteries. During implant, the stent is collapsed to a small diameter and put over a
catheter. Once it reaches the required position, it is expanded and acts as a
scaffolding, keeping the artery open and facilitating the blood flow. Stainless
steel stents are expanded using a balloon catheter, whereas shape memory alloy
stents can exploit superelasticity or the shape memory effect to self-expand in
the operating shape.
3.2.4
Civil engineering applications
Shape memory alloy applications in civil engineering are still significantly limited
by the high cost and small size of shape memory alloy components. However,
a number of applications are being proposed in the literature (Janke et al.,
2005), mostly exploiting the shape memory alloy damping capacity for energy
dissipation. The more recent idea to use shape memory alloy bars as concrete
reinforcement falls within the topic of shape memory alloy composites, thus it
is not discussed here.
Shape memory alloys have a significant damping capacity, related to the
hysteresis present in both superelastic and shape memory stress-strain curves.
5 Some shape memory alloy companies, such as Memry Corp. (www.memry.com), are
specialised in biomedical devices.
3.2 Shape memory alloy applications
119
This feature can be used in earthquake engineering to develop passive control
devices which reduce the response of a structural system during dynamic excitation (Casciati et al., 1998). The cost issue can be partially overcome by
using copper based shape memory alloys (Casciati and Faravelli, 2004). Seismic
protection devices can be included in new buildings or added as a retrofit to existing and historical structures. The development of shape memory alloy passive
devices for new buildings and historical constructions have been the subjects of
two European research projects.
The MANSIDE (1999) project (Memory Alloys for New Seismic Isolation
and Energy Dissipation Devices) involved researchers from Italy, Belgium, Sweden and Greece on the seismic protection of new buildings via shape memory
alloy isolators and dissipation devices. The complete shape recovery at zero
load in superelastic devices was used to develop recentring units, which allow
to recover the large displacements in isolators at the end of the seismic action.
Dissipation devices were developed using both superelastic and martensitic elements. Hybrid solutions were also proposed involving the use of traditional
steel dissipation devices together with shape memory alloy recentering units.
Guidelines for shape memory alloy-based seismic devices were published at the
end of the project (Dolce and Nicoletti, 1999).
The ISTECH (2000) project (Innovative Stability Techniques for the European Cultural Heritage) led to two pioneering shape memory alloy applications
in Italian historical buildings. The idea is to use shape memory alloy devices
to prevent two classical damage mechanisms of masonry buildings: out of plane
collapse and diagonal cracks. The first mechanism is the most dangerous and
can be prevented by improving the connections between the walls placed in
different directions. This is usually done with steel connections, whose linear
elastic behaviour might end up exerting excessive forces and damaging historical
masonry. Superelastic devices, on the other hand, can limit the forces applied
on the masonry by tuning the upper and lower transformation plateau. This
concept was applied in the restoration of Assisi Upper Basilica, whose pediment
had been damaged during the earthquake which hit central Italy on the 26th
and 27th September 1997. The second damage mechanism can be prevented by
pre-compressing the masonry walls. With respect to traditional steel devices,
superelastic elements exert only a limited pre-compression force and dissipate
part of the seismic action due to their damping capacity. This concept was
120
Shape memory alloys
applied in the seismic retrofitting of the San Giorgio in Trignano bell tower in
San Martino in Rio (Reggio Emilia), damaged by an earthquake on the 15th
October 1996.
Most civil engineering applications involve superelastic passive devices, which
are deemed to be more reliable as they require no external energy input. A recent application, named the Spider project, makes use of the shape memory
effect to create a self erecting structure.
3.3
Modelling of shape memory alloys
Shape memory alloys display a strong thermomechanical coupling related to the
presence of martensitic phase transitions. Investigating these materials at the
macroscopic level requires the definition of new constitutive laws6 which are able
to represent the coupling. Generally, two different approaches are possible to
define the constitutive laws, namely a micromechanical and a phenomenological
approach. In the first case, the material behaviour at the microscopic level is
described, then the macroscopic constitutive laws are obtained by averaging the
investigated phenomena over a Reference Volume Element (RVE). In the phenomenological approach, on the other hand, only the macroscopic behaviour is
taken into consideration and the material state is assumed known if some variables (state variables) are known. The state variables describe the macroscopic
phenomena and may or may not be related to the actual microscopic material
characteristics. A classical example is that of plasticity in metals: using a micromechanical approach, one would describe the phenomenon by investigating
the movement of defects and dislocations inside the material, while within a
phenomenological approach a new variable, the plastic strain, is introduced to
macroscopically describe the material behaviour. A review of various micromechanical and phenomenological constitutive models for shape memory alloys
was recently published by Lagoudas et al. (2006) and Patoor et al. (2006).
The modelling of shape memory alloys has been a topic of interest ever
since the discovery of the shape memory effect. The approaches outlined in the
previous paragraph have been exploited by researchers to propose many models,
both one-dimensional and three-dimensional. However, the modelling of shape
6 ”Constitutive equations [. . . ] describe the macroscopic behaviour resulting from the internal constitution of the material” (Malvern, 1969).
3.3 Modelling of shape memory alloys
121
memory alloys is still an open problem, given the many recent literature articles
on the topic. The reason for the active interest, besides the many cutting
edge applications of shape memory alloys, resides in the utter complexity of
the phenomena to be described. Thus, no models have been formulated that
are able to describe thoroughly the shape memory alloy behaviour, but a wide
variety of models exists among which one can choose in order to to describe
a given set of phenomena. Some features of the shape memory alloy material
behaviour have not yet been fully explained and, to the author’s knowledge, are
not included in any published constitutive model, as it will be shown later in
this chapter.
In this section, a number of one-dimensional and three-dimensional phenomenological constitutive models for shape memory alloys are considered and
compared, concentrating in particular on plasticity-like models. Indeed, due to
the presence of a stress plateau similar to the one observed in plasticity, many
shape memory alloy models are based on a plasticity-like formalism. Other approaches, however, can be followed. Within the phenomenological approach,
another possibility is to consider Ginzburg-Landau models, such as the one
proposed by Fabrizio et al. (2008). A successful phenomenological model is
required to be thermodynamically consistent, therefore in the following paragraph a short outline of continuum thermodynamics and plasticity theories is
presented, mostly inspired from the book by Lemaitre and Chaboche (1994).
3.3.1
Outline of phenomenological thermodynamics
In phenomenological thermodynamics, the thermodynamic state of a material
at a given point and instant is completely described by a set of state variables
at that instant, that depend only upon the point considered. Different material behaviours can thus be described by choosing the appropriate set of state
variables, whose evolution in time constitutes a thermodynamic process. Indeed, thermodynamically admissible processes must satisfy the second principle
of thermodynamics, thus the entropy inequality constitutes a constraint to the
admissible constitutive laws for real materials (Coleman and Noll, 1963).
State variables can be divided into observable variables (typically the total
strain ε and the temperature T ), which are the only ones involved in non dissipative phenomena, and internal variables. Internal variables are introduced
122
Shape memory alloys
to keep track of the history of the material, which in dissipative phenomena
influences the material current behaviour. Their choice is largely subjective
and influences the kind of phenomena which a given model is able to represent.
Some examples of internal variables often introduced in shape memory alloy
constitutive models are the martensite volume fraction and the total transformation strain. The different choices of internal variables in shape memory alloy
models is discussed further into this chapter, while in this paragraph they will
be generally denoted by (V1 , . . . , Vk ).
Once the state variables have been chosen, they are used to postulate the
existence of a thermodynamic potential, or free energy. Various free energies can
be defined whose difference consists in the choice of the independent variables.
For example, Helmholtz free energy is a function of the strain, temperature and
internal variables:
Ψ = Ψ (ε, T, V1 , . . . , Vk ) ,
(3.1)
while Gibbs free energy can be obtained from Helmholtz free energy via a Legendre transform, making stress the independent variable:
G = G (σ, T, V1 , . . . , Vk ) .
(3.2)
Knowledge of the thermodynamic potential yields the constitutive equations by
satisfying the second principle of thermodynamics. The following derivation
is carried out with reference to Helmholtz free energy, but one can proceed
equivalently with different potentials.
The second principle of thermodynamics in local form can be expressed as
follows:
q
r
ρṡ + ∇ · − ≥ 0,
(3.3)
T
T
where ρ is the density, s is the specific entropy, q is the heat flux and r the
external heat source. Equation (3.3) does not explicitly contain all the state
variables, but can be reformulated by taking into account the first principle of
thermodynamics (conservation of energy) and the principle of virtual power,
obtaining:
∇T
σ · ε̇ + ρ (T ṡ − ė) − q ·
≥ 0,
(3.4)
T
where e is the internal energy of the material.
3.3 Modelling of shape memory alloys
123
Helmholtz free energy is defined as
Ψ = Ψ (ε, T, V1 , . . . , Vk ) = e − T s,
(3.5)
thus the term ė in Equation (3.4) can be substituted, yielding
³
´
∇T
σ · ε̇ + ρ −sṪ − Ψ̇ − q ·
≥ 0,
T
(3.6)
The rate of change of the free energy Ψ̇ can be expressed in terms of partial
derivatives with respect to the state variables, substitution in the previous equation yields
¶
µ
¶
µ
∂Ψ ˙
∂Ψ
∂Ψ ˙
q
∂Ψ
Ṫ − ρ
· ε̇ − ρ s +
V1 − . . . − ρ
Vk − · ∇T ≥ 0. (3.7)
σ−ρ
∂ε
∂T
∂V1
∂Vk
T
As this inequality must be satisfied by every admissible thermodynamic process,
by considering an elastic deformation at constant temperature which does not
involve an evolution of the internal variables one obtains
σ=ρ
∂Ψ
,
∂ε
(3.8)
which is the constitutive equation for the stress. Analogously, for entropy one
can obtain
∂Ψ
s=−
.
(3.9)
∂T
As for the internal variables, it is possible to define the associated thermodynamic driving forces:
∂Ψ
Ak = ρ
.
(3.10)
∂Vk
Equation (3.7) can thus be written as
−A1 V˙1 − . . . − Ak V˙k −
q
· ∇T ≥ 0.
T
(3.11)
This is the dissipation inequality, stating that the total dissipation must be
greater than zero for real materials. The total dissipation is obtained as the
sum of mechanical dissipation, associated with the evolution of the internal
variables, and thermal dissipation.
124
Shape memory alloys
Defining constitutive laws via a thermodynamic potential automatically yields
the thermomechanical coupling. Indeed, while temperature terms are introduced in the mechanical problem via the stress constitutive law, the entropy
constitutive law allows to introduce mechanical dissipation effects in the heat
equation. Indeed, taking into account the first principle of thermodynamics,
Fourier’s law and the entropy constitutive equation, one can write
ρcṪ − k∇ · (∇T ) = b,
(3.12)
where c is the specific heat capacity, k is the thermal conductivity and b represents the heat source:
¶
µ 2
∂Ψ
∂ Ψ
∂2Ψ
V̇k −
V̇k .
(3.13)
b = r + ρT
· ε̇ +
∂T ∂ε
∂T ∂Vk
∂Vk
Here, the external heat source r was previously defined, the parenthesis contains terms related to thermomechanical coupling and the last term introduces
mechanical dissipation caused by the evolution of the internal variables.
The constitutive equations introduced so far do not give any information as
to the evolution of the internal variables. It is therefore necessary to introduce
kinetic laws, which must again be in accord with the dissipation inequality. A
way to define the kinetic laws is via the introduction of a dissipation potential,
which needs to satisfy some properties in order to ensure the positiveness of the
mechanical dissipation. Some details on the definition of kinetic laws for plasticity models, which are formally similar to those used in many shape memory
alloy models, are given in the following.
3.3.2
Outline of plasticity theories
In plasticity models, the strain is usually divided into elastic and inelastic parts:
ε = εe + εin . The inelastic strain can be further decomposed into thermal and
plastic strain: εin = εth + εp , which constitutes an internal variable of the
plasticity model. In shape memory alloy modelling, the transformation strain
εtr associated to detwinned martensite is generally used in place of the plastic
strain.
The kinetic laws require the definition of a yield surface f and a plastic
potential F , which are written in terms of the thermodynamic driving forces.
3.3 Modelling of shape memory alloys
125
The yield surface f , often called transformation surface in shape memory alloy
models, governs the onset of plastic deformations, while the plastic potential
F defines the direction of plastic flow. In associated plasticity, f ≡ F , but
a more general case of non associated plasticity is considered in the following
presentation.
The kinetic laws, or flow rules, are written in terms of the plastic potential
F as:
∂F
∂F
ε̇p = ζ̇
, −V˙k = ζ̇
(3.14)
∂σ
∂Ak
Here, the derivative of the plastic potential with respect to the dual variables
defines the direction of increment of the internal variables, while ζ̇ is the plastic
multiplier.
The yield surface defines the limit of the elastic domain. Inside the surface,
that is when f < 0, the internal variables do not vary, therefore the plastic
multiplier ζ̇ = 0. On the surface (f = 0), the internal variables change and the
plastic multiplier is non zero, while points outside the surface are not admissible.
The described cases can be summarised in the Kuhn-Tucker condition:
ζ̇ ≥ 0, f ≤ 0, ζ̇f = 0
(3.15)
The fulfilment of this condition, also known as consistency condition, allows to
evaluate the plastic multiplier and, therefore, the change in the internal variables
via Equation (3.14). Note that the transformation surface may change as the
internal variables evolve, thus generating a hardening behaviour.
In associated plasticity, if the limit function is convex and the normality rule
is fulfilled, the satisfaction of the dissipation inequality (3.11) is automatically
guaranteed.
In the following sections, various one-dimensional and three-dimensional
models for shape memory alloys are introduced and discussed within the approach of phenomenological thermodynamics. It should be underlined that, in
order to facilitate comparison between the different models, a common notation
has been defined and is used consistently throughout the models description.
3.3.3
One-dimensional models
One-dimensional models are still the most widely used constitutive relations
for shape memory alloys. Indeed, three-dimensional model are quite complex to
126
Shape memory alloys
handle and in some cases not strictly necessary, as many applications involve the
use of shape memory wires or ribbons. Moreover, not much experimental data is
available to validate a full three-dimensional shape memory alloy behaviour. For
all these reasons, this paragraph is devoted to the presentation and discussion
of some of the most common shape memory alloy one-dimensional constitutive
models.
As already pointed out previously, the models discussed in this section are
defined within the context of phenomenological thermodynamics and plasticity
theory. Considering only one-dimensional behaviour, however, greatly simplifies
the mathematical treatment of the problem. In particular, the yield surface
degenerates into a line. Thus, the presentation of the one-dimensional models
is somewhat simplified with respect to the three-dimensional models which are
discussed later.
Two groups of one-dimensional models are discussed in some details in the
following. The first group contains the models proposed in the works by Tanaka
(1986), Liang and Rogers (1990) and Brinson (1993), the second is a group of
models proposed by Auricchio and Sacco (1997, 1999b, 2001) and Auricchio
et al. (2003). This is by no means an exhaustive discussion, however the choice
was dictated by different reasons, which are detailed in the following.
The models by Tanaka (1986), Liang and Rogers (1990) and Brinson (1993)
were among the first to be proposed in the literature. They are very simple
constitutive laws in which the martensite volume fraction is taken as internal
variable. These models are still very popular because of their simplicity and
ability to represent both shape memory effect and superelasticity. The interest
they still raise is proven by the several recent works concerned with their comparative evaluation. In particular, the works of Epps and Chopra (1997) and
Prahlad and Chopra (2001) investigate the correspondence of the three material
models with experimental data, while Żak et al. (2003) comparatively evaluate
the behaviour of various models for use with active composites. A correction to
an inconsistency present in Brinson’s model was recently proposed by Buravalla
and Khandelwal (2007). A modified Brinson’s model able to account for the experimental behaviour in constrained recovery is presented later in this chapter
(Daghia et al., 2006).
The models by Auricchio and Sacco were proposed more recently in the
context of generalised plasticity (Lubliner and Auricchio, 1996). They are a
3.3 Modelling of shape memory alloys
127
family of models having a similar structure, in which different choices of the
internal variables allow to represent different features of the shape memory
alloy behaviour, such as superelasticity, shape memory effect, two-way shape
memory effect and training. The possibility to describe different properties by
simply including new internal variables within an existing modelling framework
is particularly attractive, as this allows to tune the model complexity according
to the investigated problem.
Models proposed by Tanaka, Liang and Rogers, Brinson
The models proposed by Tanaka (1986), Liang and Rogers (1990) and Brinson
(1993) are based on the Helmholtz free energy, even though its functional form is
not explicitly given. The internal variable considered by Tanaka and Liang and
Rogers is the martensite volume fraction, while in Brinson’s model the twinned
and detwinned martensite fractions are introduced as separate variables, thus
allowing to model the shape memory effect at low temperatures. In the notation
adopted here, the detwinned and twinned fractions are denoted by ξ d and ξ t
respectively, while in models where only one martensite fraction is involved it
is denoted by ξ d . As all models are one-dimensional, the transformation strain
is simply defined as εtr = εL ξ d , where εL is a material parameter denoting the
maximum strain associated to the phase transformation.
The constitutive law for stress is the following:
σ̇ = E ε̇ + ΘṪ + Ωξ˙d
(3.16)
where E, Θ and Ω are material moduli, which in Brinson’s model are considered
¡
¢
to vary as a function of the total martensite fraction ξ d + ξ t .
The onset of phase transformation is defined with the help of experimental
phase diagrams, reported in Figure 3.11. The zero stress transformation temperatures, Mf , Ms , As and Af , the slopes of the transformation lines in the
stress-temperature plane, CM and CA , and the critical stresses for detwinning,
σs and σf , are all material parameters which need to be determined experimentally. Note that, even though the parameters are apparently the same, they
need to be chosen differently depending upon the model in order to predict a
similar material behaviour.
Phase transformation occurs within the zones of the phase diagram marked
in grey, as long as the loading path follows the direction of the arrows. Outside
128
Shape memory alloys
s
1
CM
d
x
xd =1
CA
d
x
d
x =0
d
0<x <1
Mf Ms
As
Af
T
(a) Tanaka (1986) and Liang and Rogers (1990)
s
1
2
xd =1
xt =0
sf
3
ss
CM
d
x
x
d
x
d
d
0<x <1
0<xt <1
t
xd+x=1
CA
t
x
0<x <1
0<xt <1
d
x =0
xt =0
4
Mf Ms A s
Af
T
(b) Brinson (1993)
Figure 3.11: Phase diagrams for Tanaka (1986), Liang and Rogers (1990) and
Brinson (1993) models
3.3 Modelling of shape memory alloys
129
the phase transformation zones, the volume fractions of the different phases
are constant, their values depending upon the previous loading history of the
material. The possible values assumed by the different volume fractions in each
area of the phase diagram are marked in Figure 3.11.
Within the phase transformation zones, the kinetic laws are given as follows:
• Tanaka:
– forward A → M transformation:
ξ d = 1 − exp [AM (T − Ms ) + BM σ] ,
(3.17)
– reverse M → A transformation:
ξ d = exp [AA (T − As ) + BA σ] ,
(3.18)
where AM , AA , BM and BA are material constants in terms of the transition temperatures;
• Liang and Rogers:
– forward A → M transformation:
·
µ
¶¸
1 − ξ0d
π
σ
1 + ξ0d
d
ξ =
cos
T − Mf −
+
,
2
Ms − Mf
CM
2
– reverse M → A transformation:
·
µ
¶¸
ξ0d
π
σ
ξd
d
ξ =
cos
T − As −
+ 0,
2
Af − As
CA
2
(3.19)
(3.20)
where ξ0d is the martensite fraction prior to entering the phase transformation zone;
• Brinson:
– forward (A, M t ) → M d transformation (T > Ms ):
½
¾
π
1 + ξ0d
1 − ξ0d
cos
[σ − σf − CM (T − Ms )] +
,
ξd =
2
σs − σf
2
(3.21)
t
¡
¢
ξ
0
ξ t = ξ0t −
(3.22)
ξ d − ξ0d ,
1 − ξ0d
130
Shape memory alloys
– forward (A, M t ) → M d transformation (T < Ms ):
·
¸
1 − ξ0d
π
1 + ξ0d
d
ξ
=
cos
,
(σ − σf ) +
2
σs − σf
2
¢
ξ0t ¡ d
ξ t = ξ0t −
ξ − ξ0d ,
d
1 − ξ0
– forward A → M t transformation:
½
·
¸
¾
1 − ξ0d − ξ0t
π
t
ξ =
cos
(T − Mf ) + 1 + ξ0t ,
2
Ms − Mf
¡
¢
– reverse M t , M d → A transformation:
½
·
µ
¶¸
¾
ξ0d
π
σ
ξd =
cos
T − As −
+1 ,
2
Af − As
CA
½
·
µ
¶¸
¾
t
ξ0
π
σ
t
ξ =
cos
T − As −
+1 ,
2
Af − As
CA
(3.23)
(3.24)
(3.25)
(3.26)
(3.27)
where ξ0d and ξ0t are the detwinned and twinned martensite fractions prior
to entering the phase transformation zone.
Later, in Lagoudas et al. (2006), the same kinetic laws were shown to derive
from a particular choice of the hardening functions within the definition of the
free energy.
Models proposed by Auricchio and Sacco
The models proposed in various works by Auricchio and Sacco are derived within
the framework of generalised plasticity (Lubliner and Auricchio, 1996). In each
model, an appropriate choice of the internal variables allows to represent different features of the shape memory alloy behaviour. This way, depending upon
the problem considered, the complexity of the model can be tailored by including
only the required variables and equations.
Starting from a simple model for superelasticity at constant temperature
(model AS1, Auricchio and Sacco, 1997), the following models are improved by
including the shape memory effect (model AS2, Auricchio and Sacco, 1999b),
the coupling with the heat equation (model AS3, Auricchio and Sacco, 2001)
3.3 Modelling of shape memory alloys
131
Table 3.1: Comparison between the different models proposed by Auricchio and
Sacco
MODEL
AS1
AS2
AS3
obser.
variables
ε
ε, T
ε, T
int.
variables
ξd
ξd , ξt , β
ξd
non-constant
parameters
yes
yes
no
coupling
no
no
yes
AS4
ε, T
ξ d , β,
γM , γA
yes
no
modelling
capabilities
SE at T = const.
SE, SME
SE at different
loading rates
SE, SME at T > Ms ,
TWME, training
and finally the two-way memory effect and training (model AS4, Auricchio et al.,
2003). The different choices operated in each model, together with the phenomena they are able to represent, are summarised in Table 3.1 and detailed in the
following.
Apart from AS1, all models are strain and temperature driven, that is the
conditions for the onset of phase transition are given in terms of strain and
temperature (rather than the usual stress and temperature). This choice is
aimed at simplifying numerical implementation of the constitutive models into
displacement-based finite element codes. Indeed, model AS2 is actually implemented in the commercial finite element code ADINA (ADINA, 2006). For the
same reason, all models are given in both the time-continuous and time-discrete
frame and a predictor-corrector integration algorithm is suggested. Only the
time-continuous model formulation is discussed here, for further details refer to
the cited articles.
The volume fractions of the different constituents are introduced as internal variables. A single martensite fraction ξ d is used in models AS1, AS3 and
AS4, which deal exclusively with phenomena occurring at temperatures higher
than martensite start (T > Ms ), while the twinned and detwinned fractions,
ξ t and ξ d , are defined separately in model AS2 to take into account the shape
memory effect occurring at low temperatures. This last choice is similar to the
one operated in Brinson’s model. Models AS2 and AS4 introduce the internal variable β to describe the reorientation of detwinned martensite. In AS4,
the two-way memory effect and training are modelled by introducing the irre-
132
Shape memory alloys
s
s
sMf
sMs
CM
sf
sAs
CA
ss
AS1
sAf
T=const
Mf Ms As Af
AS2
T
(a) Model AS1
s
(b) Model AS2
s
sMfu
AS3
AS4
sMsu
sAsu
k
T
sf 1
sf 2
ss1
ss2
k
sAfu
CM
CA
Ms
Tu
(c) Model AS3
T
As1 Af 1
As2 Af 2
T
(d) Model AS4
Figure 3.12: Phase diagrams for Auricchio and Sacco’s models
versible martensite fraction ξR , that is the fraction of martensite which is not
able to transform back into austenite during reverse transformation. The total
martensite fraction ξ d is thus divided into a reversible and an irreversible part:
ξ d = ξ +ξR . The evolution of the irreversible martensite fraction and the change
in the transformation zones introduced in model AS4 are related to the total
amount of forward and reverse transformation accumulated throughout the cycles. In order to keep track of these quantities, the internal variables γM and
γA are introduced in model AS4.
In the spirit of plasticity models, strain is additively separated into the elastic
and inelastic parts. The inelastic strain is a function of the temperature (thermal
3.3 Modelling of shape memory alloys
133
strain, εth ) and the detwinned martensite fraction (transformation strain, εtr ):
• AS1: ε = εe + εL ξ d sgn (σ);
• AS2: ε = εe + ξ d β + α (T − T0 );
• AS3: ε = εe + εL ξ d sgn (σ) + α (T − T0 );
¡
¢
• AS4: ε = εe + ξ d − ξR β + ξR κ + α (T − T0 ).
Here, the term sgn (σ) is introduced to take into account the different directions
of detwinning (tension and compression). Note that the behaviour in tension
and compression of shape memory alloys is generally different, thus different
material parameters need to be defined. For example, εL specialises into ε+
L
and ε−
L , in tension and compression respectively. The parameter κ, introduced
−
in AS4 to model training, is set to ε+
L if training is in tension, −εL if training is
in compression, respectively. The term β, describing martensite reorientation,
is defined as
β̇ = Hss γ [εL sgn (σ) − β] (|σ| − σ ss ) ,
(3.28)
with
½
Hss =
1 when |σ| > σ ss
0 otherwise
(3.29)
Here, σ ss is a limit stress that activates the reorientation process, while γ is
a material parameter measuring the velocity of reorientation. Note that β is
introduced only in AS2 and AS4, which are designed to model the shape memory
effect and thus require a description of the change of martensite orientation from
tension to compression and viceversa. On the other hand, no reorientation is
possible in AS1 and AS3, focused on superelasticity. In principle, model AS3
would be able to describe the shape memory effect occurring at temperatures
above Ms , since both temperature and detwinned martensite fraction are model
variables. However, the absence of the reorientation term β reduces the model
to superelasticity description only.
The constitutive equation for the stress is expressed as
σ = Eεe
(3.30)
where the elastic modulus E is either constant (model AS3) or given as s function of the total martensite fraction. Different homogenisation possibilities are
134
Shape memory alloys
explored in model AS1, while the choice retained in the following models is
derived from the Reuss scheme:
¡
¢
E ξd + ξt =
EA EM
.
EM + (ξ d + ξ t ) (EA − EM )
(3.31)
All kinetic laws defining the evolution of the different phase fractions have
a similar structure, given by:
ξ˙ = aξ0 H
Ḟ
b
(F − Ff )
.
(3.32)
Here, a and b are material parameters which can be different for the various
phase transformations. If they are set to one, the kinetic laws become linear
(models AS1, AS2 and AS3). Similarly to Hss in (3.29), H is an indicator
function which is 1 within the corresponding transformation zone, 0 otherwise.
This term allows to define which transformations are active at a given moment.
It should be noted that, because of the general framework in which the model
is developed, the transformation zones may or may not be disjoint.
In the following, the terms appearing in Equation (3.32) are defined for each
model and phase transition:
• AS1:
– forward A → M transformation:
F
=
|σ| ,
(3.33)
Fs
=
σM s ,
(3.34)
Ff
=
(3.35)
H
=
σM f ,
½
1 if (Ḟ > 0) ∧ (Fs < F < Ff )
0 otherwise,
ξ
=
(3.37)
ξ0
=
ξd,
¡
¢
− 1 − ξd ,
(3.36)
(3.38)
3.3 Modelling of shape memory alloys
135
– reverse M → A transformation:
F
=
|σ| ,
(3.39)
Fs
=
σAs ,
(3.40)
Ff
=
(3.41)
H
=
σAf ,
½
1 if (Ḟ < 0) ∧ (Ff < F < Fs )
0 otherwise,
ξ
=
ξd,
(3.43)
=
d
(3.44)
ξ0
ξ ,
(3.42)
where σM s , σM f , σAs and σAf are the stresses at the beginning and end
of the forward and reverse transformation at a fixed temperature T . This
corresponds to considering a section of the phase diagram, as shown in
Figure 3.12(a);
• AS2:
– forward (A, M t ) → M d transformation (T > Ms ):
F
=
Fs
=
Ff
=
H
=
CM T
,
E
σs − CM Ms
εL ξ d +
,
E
σf − CM Ms
εL +
,
EM
½
1 if (Ḟ > 0) ∧ (Fs ≤ F ≤ Ff ) ∧ (T ≥ Ms )
0 otherwise,
|ε| −
(3.45)
(3.46)
(3.47)
(3.48)
∗ detwinned martensite evolution:
ξ
=
ξ0
=
ξd,
¡
¢
− 1 − ξd ,
(3.49)
=
ξt,
(3.51)
=
t
(3.52)
(3.50)
∗ twinned martensite evolution:
ξ
ξ0
ξ,
136
Shape memory alloys
– forward (A, M t ) → M d transformation (T < Ms ):
F
Fs
Ff
H
= |ε| ,
(3.53)
σ
s
= εL ξ d + ,
(3.54)
E
σf
= εL +
,
(3.55)
EM
½
1 if (Ḟ > 0) ∧ (Fs ≤ F ≤ Ff ) ∧ (T ≤ Ms )
=
(3.56)
0 otherwise,
∗ detwinned martensite evolution:
ξ
=
ξ0
=
ξd,
¡
¢
− 1 − ξd ,
(3.57)
= ξt,
(3.59)
(3.58)
∗ twinned martensite evolution:
ξ
ξ0
t
= ξ,
(3.60)
– forward A → M t transformation:
F
=
T,
(3.61)
Fs
=
Ms ,
(3.62)
Ff
=
(3.63)
H
=
Mf ,
½
1 if (Ḟ < 0) ∧ (Ff < F < Fs )
0 otherwise,
ξ
=
ξt,
¡
¢
− 1 − ξd − ξt ,
(3.65)
ξ0 =
¡ t
¢
– reverse M , M d → A transformation:
F
=
Fs
=
Ff
=
H
=
CA T
,
E
C A As
,
εL ξ d −
E
CA Af
−
,
EA
½
1 if (Ḟ < 0) ∧ (Ff < F < Fs )
0 otherwise,
|ε| −
(3.64)
(3.66)
(3.67)
(3.68)
(3.69)
(3.70)
3.3 Modelling of shape memory alloys
137
∗ detwinned martensite evolution:
ξ
ξ0
=
ξd,
(3.71)
=
d
ξ ,
(3.72)
=
ξt,
(3.73)
=
t
(3.74)
∗ twinned martensite evolution:
ξ
ξ0
ξ.
Note that the whole phase diagram is described by model AS2 (see Figure
3.12(b));
• AS3:
– forward A → M transformation:
F
=
Fs
=
Ff
=
H
=
ξ
=
ξ0
=
k
(T − Tu ) ,
E
σM su
εL ξ d +
,
E
σM f u
εL +
,
E
½
1 if (Ḟ > 0) ∧ (Fs < F < Ff )
0 otherwise,
|ε| −
ξd,
¡
¢
− 1 − ξd ,
(3.75)
(3.76)
(3.77)
(3.78)
(3.79)
(3.80)
– reverse M → A transformation:
k
(T − Tu ) ,
E
σAsu
εL ξ d +
,
E
σAf u
,
E
½
1 if (Ḟ < 0) ∧ (Ff < F < Fs )
0 otherwise,
F
=
Fs
=
Ff
=
H
=
ξ
=
ξd,
(3.85)
=
d
(3.86)
ξ0
|ε| −
ξ .
(3.81)
(3.82)
(3.83)
(3.84)
138
Shape memory alloys
Here, Tu is an upper bound temperature and the stresses at which the
forward and reverse transformations begin and end at temperature Tu are
denoted by σM su , σM f u , σAsu and σAf u , respectively. This is just a different way of characterising the phase diagram (see Figure 3.12(c)). Moreover, as in model AS3 the functional form of the free energy is explicitly
given, the slope of the phase transformation lines in the stress-temperature
space is derived from the definition of the thermodynamic driving force,
resulting
CM = CA = k = −
∆s
.
εL
(3.87)
Thus, the onset of phase transition is related to the change in entropy
between the two phases ∆s and to the maximum transformation strain.
This is similar to what observed in the three-dimensional model by Popov
and Lagoudas (2007), discussed further on in this chapter;
• AS4:
– forward A → M transformation:
F
= εsgn(σ) −
Fs
=
Ff
H
ξ
ξ0
£¡
CM T
,
E
¢
¤
σs − CM Ms
ξ d − ξR β + ξR κ sgn(σ) +
,
E
σf − CM Ms
= [(1 − ξR ) β + ξR κ] sgn(σ) +
,
EM
½
1 if (Ḟ > 0) ∧ (Fs < F < Ff )
=
0 otherwise,
= ξd,
¡
¢
= − 1 − ξd
(3.88)
(3.89)
(3.90)
(3.91)
(3.92)
(3.93)
3.3 Modelling of shape memory alloys
139
– reverse M → A transformation:
F
Fs
Ff
H
ξ
ξ0
= εsgn(σ) −
¢
¤
CA As
ξ d − ξR β + ξR κ sgn(σ) −
,
E
CA Af
= ξR κsgn(σ) −
,
EA
½
1 if (Ḟ < 0) ∧ (Ff < F < Fs )
=
0 otherwise,
=
£¡
CA T
,
E
= ξd,
d
= ξ − ξR .
(3.94)
(3.95)
(3.96)
(3.97)
(3.98)
(3.99)
Note that the material parameters σs , σf , As and Af are not constant
in this model, but evolve from the virgin to the trained material (see
Figure 3.12(d)). Their evolution, as well as the evolution of the irreversible
martensite fraction ξR , is given in the following.
All terms described up to now are common to all four models. In addition,
models AS3 and AS4 have special features detailed here.
In model AS3, complete thermomechanical coupling is defined by explicitating the functional form of the Helmholtz free energy. Then the heat equation
is derived as described in Section 3.3.1 and automatically contains the coupling
terms. The free energy is
·
¸
£
¤
T
d
Ψ = (eA − T sA ) + ξ (∆e − T ∆s) + C (T − T0 ) − T ln
+
T0
¤2
£
¤
1 £
(3.100)
+ E ε − εL ξ d sgn(σ) − (T − T0 ) ε − εL ξ d sgn(σ) Eα,
2
where the subscript A denotes quantities referred to the austenite phase, ∆e
and ∆s are the differences in internal energy and entropy between martensite
and austenite, T0 denotes the temperature of the reference state.
The constitutive equation for the stress has already been introduced, while
the constitutive equation for the entropy, necessary to write the coupled heat
equation, is here recalled:
s=−
£
¤
∂Ψ
T
= sA + ξ d ∆s + C ln
+ ε − εL ξ d sgn(σ) Eα.
∂T
T0
(3.101)
140
Shape memory alloys
Assuming that the external contribution is null, the heat source is given by the
sum of the thermomechanical coupling terms and the mechanical dissipation
(see Equation (3.13)):
b =
n
o
T −Eαε̇ + [εL sgn(σ)Eα − ∆s] ξ˙d +
+ (εL |σ| + T ∆s − ∆e) ξ˙d .
(3.102)
The heat equation is then given by Equation (3.12). Note that the term in
the last bracket is the thermodynamic driving force associated to the internal
variable ξ d , which is used to define the kinetic equations for the variable itself.
By simultaneously solving the thermal and mechanical problems, model AS3
allows to study the effect of the loading rate on the shape memory alloy response.
If the loading is fast and the heat generated by the phase transformation is not
given time to dissipate, the phase transition zone of the stress-strain diagram
changes significantly: in particular, the slope of the plateau zone is higher the
fastest the loading rate. This result is similar to that described in Motahari and
Ghassemieh (2007), where a simple model considering isothermal and adiabatic
response (that is the two limit cases) is proposed. The influence of the loading
rate on the shape memory alloy behaviour is particularly significant in seismic
applications, where fast loads are involved and an accurate prediction of the
dissipation is important.
In model AS4, new internal variables and relations are introduced in order to
predict the training and the two-way memory effect. These effects are related to
the total amount of forward and reverse transformation accumulated throughout
the loading cycles. In order to keep track of these quantities, two new internal
variables are introduced, γM and γA . Their evolution is the following:
¯
¯
¯
¯
¯ d ¯
¯ d ¯
γ̇M = ¯ξ˙AM
(3.103)
¯ , γ̇A = ¯ξ˙M
A¯
d
d
where ξAM
and ξM
A represent the martensite fraction during the forward (A →
M ) and reverse (M → A) transformation, respectively. Note that due to the
presence of the modulus, γM and γA are always increasing and represent, in
some sense, the amount of training that the material received.
According to the physical interpretation adopted in the paper, as the material is loaded and unloaded a fraction of the formed martensite is not able
3.3 Modelling of shape memory alloys
141
to transform back to austenite. The irreversible martensite fraction ξR is thus
introduced:
ξR = ξL [1 − exp (−br γM )] ,
(3.104)
where ξL is the maximum irreversible martensite fraction which can be formed
and br is a material parameter measuring the ability of the material to be
trained.
Moreover, experimental observations show that the shape memory alloy behaviour is not initially stable, but stabilised after a certain number of loading
cycles. In particular, the transformation zones shift to the right-bottom part
of the phase diagram (see Figure 3.12(d)). In order to model this, the material
parameters defining the transformation zones evolve according to exponential
equations:
σs
= σs1 − (σs1 − σs2 ) [1 − exp (−bsM γA )] ,
(3.105)
σf
= σf 1 − (σf 1 − σf 2 ) [1 − exp (−bf M γA )] ,
(3.106)
As
= As1 − (As1 − As2 ) [1 − exp (−bsA γM )] ,
(3.107)
Af
= Af 1 − (Af 1 − Af 2 ) [1 − exp (−bf A γM )] ,
(3.108)
where the subscripts 1 and 2 indicate the transformation parameters for virgin
and trained material, respectively, bsM , bf M , bsA and bf A are material parameters. Note that the transformation zone for the forward A → M transformation
evolves as a function of γA , which represents the total amount of reverse transformation and viceversa.
3.3.4
Three-dimensional models
Most shape memory alloy applications involve the use of devices displaying
one-dimensional behaviour, such as wires and ribbons. On the other hand,
a three-dimensional description of the material behaviour may be required in
order to model complex devices, such as stents or composite elements with embedded shape memory alloys. Even within phenomenological models, however,
the extension of one-dimensional shape memory alloy models to a full threedimensional description is not straightforward and some choices must be made
which significantly influence the model predictive capabilities. In particular, the
use of scalar or tensorial quantities as the main internal variables can drastically
change the way some microscopic phenomena are portrayed.
142
Shape memory alloys
When describing the three-dimensional shape memory alloy behaviour, all
possible orientations of the martensite variants need to be accounted for. In single crystals, only a finite number N of martensite variants can appear depending upon the crystal symmetry (e.g., in the cubic to tetragonal transformation
3 martensite variants are possible, while in the cubic to monoclinic transformation 24 variants can appear, see Boyd and Lagoudas, 1996a), thus a way
to represent all possible material configurations is to introduce N martensite
fraction internal variables. Polycrystalline shape memory alloys, on the other
hand, are made up of single crystal grains with different orientations. As the
martensite variants are defined with respect to the crystallographic axes, whose
orientation changes in the different grains, in polycrystalline shape memory alloys N internal variables are not sufficient to thoroughly describe the material
microstructure.
Due to the presence of differently oriented martensite variants, the proportional relation between the transformation strain and the detwinned martensite fraction introduced in most one-dimensional shape memory alloy models is
not obviously extended to three dimensions. In three-dimensional models, the
amount of phase transformation can still be described in terms of scalar internal
variables representing the volume fractions of the different constituents (as in
Popov and Lagoudas, 2007) or directly in terms of the tensorial transformation
strain (as in Auricchio and Petrini, 2004a). These different approaches deeply
influence the model structure, their advantages and drawbacks are outlined in
the following and detailed in the remainder of this section. In the first case,
the transformation strain evolves together with the corresponding scalar variable, thus other (tensorial) internal variables need to be introduced in order to
depict martensite reorientation. In the second case, a single tensorial variable
depicts phase transformation, detwinning and martensite reorientation, but the
model is not able to distinguish among different phases if no macroscopic deformation is associated to them. The discussion suggests to adopt an alternative
approach, based on the introduction of two independent internal variables to
define the crystal structure and arrangement: a scalar variable could be used to
keep track of the martensite fraction, while its orientation could be described
by an independent tensorial variable denoting the mean martensite strain. The
roles played by the two different variables are similar, though not identical,
to those of the detwinned martensite volume fraction ξ d and the reorientation
3.3 Modelling of shape memory alloys
143
parameter β in the models by Auricchio and Sacco, described in the previous
section. A similar approach was recently adopted in the model proposed by
Peultier et al. (2007).
Another issue to be dealt with in three-dimensional plasticity-like shape
memory alloy models is the definition of the transformation surface. Provided
some mathematical properties, such as convexity, are satisfied, different surfaces
can be defined within a given model without altering its structure (see, for example, Qidway and Lagoudas, 2000). Symmetric transformation surfaces are
usually defined in terms of the second deviatoric stress invariant (Auricchio and
Petrini, 2002; Popov and Lagoudas, 2007), while the third invariant can be introduced to account for tension-compression asymmetry (Auricchio and Petrini,
2004a; Qidway and Lagoudas, 2000). Indeed, the different behaviour in tension
and compression is widely accounted for both by experimental observations and
model predictions (Lexcellent and Schlömerkemper, 2007).
Popov and Lagoudas model
The model recently proposed by Popov and Lagoudas (2007) belongs to a family
of models developed by Lagoudas and coworkers starting from the mid 1990s
(Bo and Lagoudas, 1999a,b,c; Bo et al., 1999; Boyd and Lagoudas, 1996a,b;
Lagoudas and Bo, 1999; Lagoudas et al., 1996; Popov and Lagoudas, 2007; Qidway and Lagoudas, 2000). In these works, micromechanical considerations and
an averaging process over the Reference Volume Element (Bo and Lagoudas,
1999a; Boyd and Lagoudas, 1996a) constitute the rational basis for the development of phenomenological models. As in the works by Auricchio and Sacco, all
models developed by Lagoudas and coworkers have the same structure, while
the introduction or removal of the appropriate internal variables allow to predict different phenomena. In the following, the most recent version of the model
(Popov and Lagoudas, 2007) is discussed in details, while some comments are
given on the choices operated in the previous works.
In all models, internal variables are introduced to describe the microstructure in terms of mass fractions of the different phases (as the phase transformation is mostly volume-preserving, the mass and the volume fractions are
considered equivalent). As discussed earlier for one-dimensional models, a single internal variable ξ d may be chosen to define the whole martensite fraction
(Bo and Lagoudas, 1999a; Boyd and Lagoudas, 1996a; Qidway and Lagoudas,
144
Shape memory alloys
s
d
x
d3
d2
x
x
d2
x
t
x
d
t1
x
t1
x
t
1-x -x
T
Figure 3.13: Internal variables for Popov and Lagoudas (2007) model
2000), alternatively two variables ξ d and ξ t may be introduced for detwinned
and twinned martensite, respectively (Boyd and Lagoudas, 1996a). In Popov
and Lagoudas (2007), the phase volume fractions are distinguished according
to their origin (see Figure 3.13). Three internal variables are thus introduced:
ξ t1 is the twinned martensite fraction, generated via phase transformation from
austenite, ξ d2 and ξ d3 are the detwinned martensite fractions generated via
phase transformation from austenite and via detwinning from twinned martensite, respectively. Thus, the total twinned and detwinned martensite fractions
are
ξ t = ξ0t + ξ t1 − ξ d3 , ξ d = ξ0d + ξ d2 + ξ d3 .
(3.109)
This choice allows to keep track of detwinning and phase transformation as
separate phenomena.
Different effects contribute to the overall inelastic strain: thermal expansion, phase transformation, detwinning of twinned martensite, reorientation of
detwinned martensite and plastic strain. Each model includes different terms
3.3 Modelling of shape memory alloys
145
from this set.
Having introduced the phase volume fractions as internal variables, the
transformation strain εtr is directly related to ξ d via an evolution equation of
the type ε̇tr = Λξ˙d , where the tensor Λ is a function of the applied stress during
the forward transformation and of the current transformation strain during the
reverse transformation. This ensures complete disappearance of the transformation strain when ξ d = 0. This relation is obtained through micromechanical
considerations in Boyd and Lagoudas (1996a) and Bo and Lagoudas (1999a)
and derived as a consequence of the principle of maximum transformation dissipation in Qidway and Lagoudas (2000). In the model by Popov and Lagoudas
(2007), detwinning and transformation strain are defined by two flow rules and
may be regulated by different tensors:
ε̇d = Λd ξ˙d3 , ε̇tr = Λtr ξ˙d2 .
Tensors Λd and Λtr are specified as
r
3
dev (σ)
Λd =
εL,d
,
2
kdev (σ)k
 q
dev(σ)
3

˙t2 > 0

2 εL,tr kdev(σ)k for ξ
q
Λtr =
dev(εtr +εd ) for ξ˙t2 < 0 ,
3


ε
L,tr
2
kdev(εtr +εd )k
(3.110)
(3.111)
(3.112)
where dev (·) denotes the deviatoric part of a tensor, εL,d and εL,tr are material
parameters defining the maximum uniaxial strain in detwinning and transformation, respectively.
As the transformation and detwinning strain evolve together with the corresponding martensite fractions and no other strain variable related to the material microstructure is introduced, the model by Popov and Lagoudas (2007)
does not allow for reorientation of detwinned martensite. This constitutes a
significant limitation, as it is not possible to model tension-compression loading
cycles below the austenite finish temperature or loading paths in which the load
direction changes over time. This limitation could be overcome by introducing
new internal variables, such as the detwinned martensite reorientation strain εr .
This way, three different tensors would be used to represent the transformation,
detwinning and reorientation strains, namely εtr , εd and εr .
146
Shape memory alloys
The idea outlined to model detwinned martensite reorientation is, obviously,
not the only possible approach. Detwinning and reorientation are, in fact, the
same phenomenon. In both cases, indeed, the martensite crystals change their
orientation because of a change in the applied load: in the first case, twinned
martensite becomes detwinned, while in the second case the change is between
different detwinned martensite variants. As the twinned and detwinned martensite are described with different internal variables in Popov and Lagoudas (2007),
detwinning and reorientation must be described separately. On the other hand,
a single martensite volume fraction internal variable ξ d would lead to a single
reorientation tensor εr to model reorientation of both twinned and detwinned
martensite (see Boyd and Lagoudas, 1996a).
Finally, plastic deformations are introduced in Bo and Lagoudas (1999a,b,c)
and Lagoudas and Bo (1999) to account for the irreversible phenomena related
to two-way memory and training. This choice implies a different microstructural
interpretation with respect to the one adopted by Auricchio et al. (2003), who
depicted the same phenomena by introducing the accumulation of irreversible
martensite.
In order to obtain the constitutive equations, a functional form of the free
energy is explicitly defined in every model. Gibbs free energy is considered as
most experimental tests are carried out in stress control. From micromechanical
considerations, the free energy is made up of two parts:
G = Gpure + Gmix ,
(3.113)
where Gpure is the thermoelastic energy of the pure phases, while Gmix represents
the energy of mixing, responsible for the transformation behaviour. Indeed, it
has been shown by Lagoudas et al. (1996, 2006) that many earlier models can
be interpreted within the same context with an appropriate choice of the term
Gmix . In Popov and Lagoudas (2007), the thermoelastic energy of a general
phase α = {A, M } is given as follows:
Gα
1
1
1
σ · Fα σ − σ · αα (T − T0 ) − σ · (εtr + εd ) +
2ρ
ρ
ρ
·
¸
T
(3.114)
+cα (T − T0 ) − T ln
− s0α (T − T0 ) + e0α ,
T0
= −
where F is the flexibility tensor, α is the thermal expansion coefficients tensor,
c is the specific heat capacity, s0 and e0 the entropy and internal energy at the
3.3 Modelling of shape memory alloys
147
reference state and the subscript α refers to the general phase α. The energy
Gpure is therefore obtained as
¢
¡
Gpure = GA + ξ t + ξ d (GM − GA ) =
1
1
1
= − σ · Fσ − σ · α (T − T0 ) − σ · (εtr + εd ) +
2ρ
ρ
ρ
·
¸
T
+c (T − T0 ) − T ln
− s0 (T − T0 ) + e0 ,
(3.115)
T0
where the material parameters are a function of the martensite fraction:
¢
¡
¢
¡
F = F ξ t + ξ d = FA + ξ t + ξ d ∆F,
(3.116)
¡ t
¢
¡ t
¢
d
d
α = α ξ + ξ = αA + ξ + ξ ∆α,
(3.117)
¡ t
¢
¡ t
¢
d
d
c = c ξ + ξ = cA + ξ + ξ ∆c,
(3.118)
¡ t
¢
¡ t
¢
d
d
s0 = s0 ξ + ξ = s0A + ξ + ξ ∆s0 ,
(3.119)
¡ t
¢
¡ t
¢
d
d
e0 = e0 ξ + ξ = e0A + ξ + ξ ∆e0 .
(3.120)
The free energy of mixing is the following:
Z
i
1 t h ˙t1
Gmix =
f1 ξ (τ ) + f2 ξ˙d2 (τ ) + f3 ξ˙d3 (τ ) dτ,
ρ 0
where
³
³ ´´
f1 = f1 ξ t1 , ξ d2 , ξ d3 , sgn ξ˙t1 ,
³
³ ´´
f2 = f2 ξ t1 , ξ d2 , ξ d3 , sgn ξ˙d2 ,
¡
¢
f3 = f3 ξ t1 , ξ d2 , ξ d3
(3.121)
(3.122)
(3.123)
(3.124)
are general hardening functions which may be different for the forward and
reverse transformation. The choice of the hardening functions depends upon
the material considered and should be part of material characterisation. In
Popov and Lagoudas (2007) they are simply chosen as a linear function of the
martensite fractions:
½ + t
∆1 ξ for ξ˙t1 > 0
f1 =
(3.125)
t
˙t1 < 0 ,
∆−
1 ξ for ξ
½ + d
∆2 ξ for ξ˙d2 > 0
f2 =
(3.126)
d
˙d2 < 0 ,
∆−
2 ξ for ξ
f2 = ∆3 ξ d for ξ˙d3 > 0,
(3.127)
148
Shape memory alloys
±
where ∆±
1 , ∆2 and ∆3 are material parameters which serve as scaling factors
so that fi (0) = 0 and fi (1) = 1. Note that the superscripts + and − denote
the forward and reverse transformations, respectively.
The constitutive equations are chosen to satisfy the second principle of thermodynamics. They are
ξ
d2
σ
=
s
=
F(−1) [ε − α (T − T0 ) − εtr − εd ] ,
1
T
σ · α + c ln
+ s0 .
ρ
T0
(3.128)
(3.129)
The thermodynamic driving forces associated to the internal variables ξ t1 ,
and ξ d3 are the following
πt1 = π̃ − f1 , πd2 = π̃ + σ · Λtr − f2 , πd3 = σ · Λd − f3 ,
(3.130)
where
1
π̃ = σ · ∆Fσ + σ · ∆α (T − T0 ) + ρ∆s0 T +
2
·
¸
T
− ρ∆c (T − T0 ) − ln
− ρ∆e0 .
T0
(3.131)
The transformation and detwinning surfaces are defined as follows:
F1+ = πt1 − Y1+ = 0,
F1− = −πt1 − Y1− = 0,
(3.132)
F2+
F2−
(3.133)
= πd2 −
Y2+
= 0,
= −πd2 −
Y2−
F3 = πd3 − Y3 = 0,
= 0,
(3.134)
where Y1± , Y2± and Y3 are material parameters. The corresponding consistency
conditions are defined directly in terms of the internal variables evolution:
ξ˙t1 ≥ 0, F1+ ≤ 0,
ξ˙t1 ≤ 0, F1− ≤ 0,
ξ˙d2 ≥ 0, F2+ ≤ 0,
ξ˙d2 ≤ 0, F − ≤ 0,
2
ξ˙d3 ≥ 0, F3 ≤ 0,
F1+ ξ˙t1 = 0,
F1− ξ˙t1 = 0,
F + ξ˙d2 = 0,
2
F2− ξ˙d2 = 0,
F3 ξ˙d3 = 0,
(3.135)
(3.136)
(3.137)
(3.138)
(3.139)
thus ξ˙t1 , ξ˙d2 and ξ˙d3 are the consistency parameters for the model. The evolution of the internal variables can thus be obtained directly by enforcing the
3.3 Modelling of shape memory alloys
149
s
d
x =1
xt =0
d
0<x <1
xt =0
t1
x
d2
x
t1
x
sf
ss
d2
x
d3
x
d
0<x <1
0<xt <1
t
xd+x=1
d
x =0
xt =0
d
0<x <1
0<xt <1
Mf Ms
Ats Atf
Ads Adf
T
Figure 3.14: Phase diagram for Popov and Lagoudas (2007) model
consistency conditions, then the corresponding inelastic strain evolution can be
evaluated from the flow rules (3.110). Note that the martensite fractions must
satisfy the following constraints:
0 ≤ ξ t ≤ 1,
(3.140)
d
(3.141)
0 ≤ ξ ≤ 1,
¡
¢
0 ≤ ξ d + ξ t ≤ 1,
(3.142)
thus the consistency condition for a given transformation holds only until the
corresponding transformation is complete.
The model described up to now requires the definition of a number of parameters. In particular, the parameters defining the transformation and hardening
±
±
behaviour, namely (∆±
i , ∆3 , Yi , Y3 , fi , f3 ) where i = {1, 2}, are related to
the uniaxial phase diagram in the stress-temperature space (see Figure 3.14).
Note that the phase diagram depicted in Figure 3.14 is different from those assumed in the models discussed up to now. Indeed, as different internal variables
150
Shape memory alloys
are defined for phase transformations and detwinning, five independent strips
are depicted in the phase diagram, each associated to one of the transformation surfaces (3.132), (3.133) and (3.134). Even though the relative position of
the transformation strips is restricted by thermodynamic considerations, this
approach allows a certain freedom to define the transformation behaviour. In
particular, a novelty introduced in this model is the difference between the reverse transformation temperatures for twinned and detwinned martensite (Ads
and Adf versus Ats and Atf ). This distinction is motivated by calorimetric observations and it is further discussed in Section 3.4. Moreover, the detwinning
strip does not end at the martensite start temperature but extends all the way
to the reverse transformation. This prevents the existence of twinned martensite at high stress levels, which was possible for certain loading paths in the
previous phase diagrams. Other differences with respect to previously proposed
phase diagrams are the presence of a vertical twinned martensite to austenite
transformation strip and the possibility to either extend to zero stress or limit
at a critical stress value the detwinned martensite forward transformation, according to the level of training of the material. Once the phase diagram is
experimentally defined, the transformation and hardening parameters can be
determined by simulating various simple loading paths. The calculations are
straightforward and they are not reported.
The remaining model parameters include the characteristics of the single
phases and the maximum uniaxial transformation and detwinning strain: (Fα ,
αα , cα , s0α , e0α , εL,d , εL,tr ), where α = {A, M }. These parameters have a direct
physical interpretation and can be determined via mechanic and calorimetric
tests.
The compliances and thermal expansion coefficients of the two phases can be
determined by standard tests. The maximum transformation and detwinning
strain are easily evaluated via uniaxial tests with complete transformation or
detwinning.
The specific heat capacity of the two phases can be determined from calorimetric measurements. In particular, it is often safe to assume the same specific
heat for austenite and martensite, thus in the following calculations we set
∆c = 0.
The difference in internal energy and entropy between the two phases, ∆e0
and ∆s0 respectively, can be determined via differential scanning calorimeter
3.3 Modelling of shape memory alloys
Heat flow
Q
ex-f
Q
151
ex
baseline
Mf
T
f
Ms
As
T
r
Af
Q
ex-r
T
Figure 3.15: Schematic representation of a DSC measurement for shape memory
alloys
measurements (Bo et al., 1999; Lagoudas and Bo, 1999). The differential scanning calorimeter (DSC) allows to measure the heat flow through a sample of
material when the material is heated or cooled at zero stress. A schematic
representation of a DSC measurement is reported in Figure 3.15. The baseline
represents the heat flow in the absence of phase transformation, while the peaks
and troughs occurring during phase transformation are related to the latent
heat. In particular, the rate of the excess heat is given by Q̇ex = Q̇ − cṪ and
it is represented by the distance between the curve and the baseline. Integrating over the temperature yields the total latent heat Qex for the forward or
reverse transformation, represented by the shaded areas in Figure 3.15. Notice
that the DSC curves can be used to determine the zero stress transformation
temperatures Ms , Mf , As and Af .
The difference in internal energy between austenite and martensite, ∆e0 , is
related to the entropy difference ∆s0 and it can be determined by defining the
equilibrium temperature T eq . The equilibrium temperature is the temperature
152
Shape memory alloys
at which the driving force at zero stress is null (Raniecki and Lexcellent, 1994),
thus:
ρ∆s0 T eq − ρ∆e0 = 0.
(3.143)
³
´
Assuming that the equilibrium temperature is given by T eq = Ms + Atf /2,
one obtains
Ms + Atf
ρ∆e0 = ρ∆s0
.
(3.144)
2
The entropy difference is related to the total latent heat Qex . As already
pointed out, Qex can be measured experimentally via DSC. Its analytical value
is obtained in the following starting from the heat equation, Equation (3.12).
The heat source, Equation (3.13), can be rewritten in terms of the Gibbs free
energy as
µ 2
¶
∂ G
∂G
∂2G
b = r + ρT
V̇k −
V̇k .
(3.145)
· σ̇ +
∂T ∂σ
∂T ∂Vk
∂Vk
Then, considering Equation (3.12) and the rate of specific heat change ρQ̇ =
r − ∇ · q it is possible to evaluate the rate of the excess heat:
µ 2
¶
∂ G
∂2G
ρQ̇ex = −ρT
· σ̇ +
V̇k + Ak V̇k .
(3.146)
∂T ∂σ
∂T ∂Vk
During a DSC test on a sample containing no detwinned martensite, the stress
is null and the only evolving internal variable is ξ t1 . The rate of excess heat
thus becomes
µ
¶
∂2G
˙t1
ρQ̇ex = −ρT
+
π
(3.147)
t1 ξ .
∂T ∂ξ t1
During forward transformation, the first consistency condition (3.135) becomes
+
˙t1 = 0,
πt1 = Y1+ , F1+, T Ṫ + F1,ξ
t1 ξ
thus combining the previous expressions yields
¶ +
µ
F1,T
∂2G
+
ex
+ Y1
Ṫ .
ρQ̇ = − −ρT
+
∂T ∂ξ t1
F1,ξ
t1
(3.148)
(3.149)
The functional form of the free energy previously introduced enables us to write
ρQ̇
ex
+
¡
¢ F1,T
+
Ṫ .
= − ρT ∆s0 + Y1
+
F1,ξ
t1
(3.150)
3.3 Modelling of shape memory alloys
153
Integrating yields the total latent heat for the forward transformation
Qex−f = ρ∆s0 T̄ f − Y1+ ,
(3.151)
where the temperature T̄ f is an average temperature for the forward phase
transformation, defined as
Z
T̄ f = −
T
∞
Z
+
F1,T
−∞
+
F1,ξ
t1
∞
Ṫ =
T
−∞
+
F1,T
+
F1,ξ
t1
Ṫ .
(3.152)
An approximation of T̄ f can be determined by considering a temperature which
divides the area representing the latent heat in two equal parts. Notice that
Z
−∞
∞
Z
+
F1,T
−∞
Ṫ =
+
F1,ξ
t1
ξ˙t1 = 1,
(3.153)
∞
as the twinned martensite fraction goes from zero to one during the complete
forward transformation. In an analogous way, for the reverse transformation it
is possible to obtain
Qex−r = −ρ∆s0 T̄ r − Y1− ,
(3.154)
where
Z
T̄ r =
∞
T
−∞
−
F1,T
−
F1,ξ
t1
Ṫ .
(3.155)
Neglecting irreversible phenomena, the total latent heat released during forward
transformation
¯ should
¯ be equal to the one absorbed during reverse transformation, that is ¯Qex−f ¯ = |Qex−r |. This yields
¡
¢
Qex−f − Qex−r = 2Qex−f = ρ∆s0 T̄ f + T̄ r − Y1+ + Y1− .
(3.156)
From the expressions for Y1± reported in Popov and Lagoudas (2007) and Equation (3.144) we obtain Y1+ − Y1− = 0, thus ∆s0 can be evaluated as follows
ρ∆s0 =
2Qex−f
.
T̄ f + T̄ r
(3.157)
It is interesting to notice that, as already anticipated in the discussion of the
models proposed by Auricchio and Sacco earlier in this chapter, the entropy
difference is also related to the slope of the A → M d transformation lines in
154
Shape memory alloys
the uniaxial phase diagram. Indeed, neglecting the difference in elastic moduli
and thermal expansion coefficients between the two phases, the uniaxial transformation lines become
ρ∆s0 T − ρ∆e0 + σεL,tr − f2 ± Y2∓ = 0,
(3.158)
thus, in the stress-temperature plane, the slope is
k=−
ρ∆s0
.
εL,tr
(3.159)
Indeed, due to the different properties between the two phases, the transformation curves are not exactly linear, but this effect is evident mostly at high stress
levels.
Auricchio and Petrini model
The model proposed by Auricchio and Petrini (2004a) was developed starting
from the work of Souza et al. (1998). First, some improvements to the timecontinuous model and the integration algorithms were proposed (Auricchio and
Petrini, 2002), then complete thermomechanical coupling was introduced (Auricchio and Petrini, 2004a,b). More recently, a modified model introducing the
effect of permanent deformations was proposed (Auricchio et al., 2007). Here,
the 2004 version of the model is discussed in details and some comments are
included concerning the other works.
This model is cast in the framework of generalised standard materials. This
framework, along with the choice of convex potentials, allows to demonstrate the
well-posedness of the model (which automatically satisfies the second principle
of thermodynamics) and thus to have a robust time-continuous model to be
used for subsequent numerical developments.
The transformation strain tensor εtr is chosen as internal variable. Experimental evidence suggests that martensitic transformations preserve the volume,
thus the transformation strain is a purely deviatoric tensor. The choice of a
tensorial internal variable makes the model by Auricchio and Petrini dramatically different from the models proposed by Lagoudas and coworkers, previously
discussed. Indeed, the martensite volume fraction is not defined here and the
phase transformation is viewed only through its effect on the transformation
3.3 Modelling of shape memory alloys
155
strain. For this reason, the model can only distinguish between an undeformed
parent phase (which could be either austenite or twinned martensite) and a
product phase which shows some degree of detwinning. This feature may be
viewed as a limitation of the model, as it does not allow to account for the
different material properties of austenite and martensite and to model purely
temperature-induced phase transformation (such as the one occurring during a
DSC test). On the other hand, the transformation strain evolution law does not
need to be related to the evolution of a scalar parameter, thus phase transformation, detwinning of twinned martensite and reorientation between different
detwinned martensite variants can be described within the same evolution law.
Another internal variable is introduced in the modified model (Auricchio et al.,
2007) in order to keep track of the permanent inelastic strain.
A convex free energy function is postulated. As the time-discrete version
of the model is going to be used for finite element simulations, a strain-driven
problem is formulated by using Helmholtz free energy:
Ψ
=
1
2
2
K [tr (ε)] + G kdev (ε) − εtr k − 3αKtr (ε) (T − T0 ) +
2 ·
¸
T
+c (T − T0 ) − T ln
− s0 T + e 0 +
T0
1
2
(3.160)
+k hT − Mf i kεtr k + h kεtr k + IεL (εtr ) ,
2
where tr (·) denotes the trace of a tensor, K and G are the bulk and shear
moduli, α is the thermal expansion coefficient, k and h are material parameters
and h·i denotes the positive part of the argument. Here, the first line contains
the thermoelastic volumetric and deviatoric terms; the second line contains the
energy related to temperature change; finally, the third line contains the terms
related to phase transformation and hardening. In particular, the last term is
an indicator function used to limit the norm of the transformation strain:
½
IεL (εtr ) =
0
if kεtr k ≤ εL
.
+∞ if kεtr k > εL
(3.161)
156
Shape memory alloys
The constitutive equations are derived using standard arguments:
tr (σ)
=
3K [tr (ε) − 3α (T − T0 )] ,
(3.162)
dev (σ)
=
(3.163)
s
=
2G (dev (ε) − εtr ) ,
T
hT − Mf i
3αKtr (ε) + c ln
+ s0 − k kεtr k
.
T0
kT − Mf k
(3.164)
The thermodynamic driving force of the transformation is
X = dev (σ) − [k hT − Mf i + h kεtr k + γ]
where
∂ kεtr k
,
∂εtr
γ = 0 if kεtr k < εL ,
γ ≥ 0 if kεtr k = εL .
(3.165)
(3.166)
Note that the bracketed term in Equation (3.165) has a role similar to that
of the back-stress in classical plasticity, defining the position of the centre of
the elastic domain. As the derivative of the transformation strain norm is not
defined for kεtr k = 0, in the time-discrete version of the model the Euclidean
norm is substituted with a regularised norm:
kεtr kreg = kεtr k −
δ (δ+1)/δ
(δ−1)/δ
(kεtr k + δ)
,
δ−1
(3.167)
where δ is a regularisation parameter.
A transformation function which accounts for tension-compression asymmetry is defined:
p
J3
F (X) = 2J2 + m − R,
(3.168)
J2
where J2 and J3 are the second and third invariant of the tensor X, m and
R are material parameters related to the critical detwinning stresses in tension
and compression, σt and σc :
r
r
27 σc − σt
2 σc σt
m=
, R=2
.
(3.169)
2 σc + σt
3 σc + σt
It should be m ≤ 0.46 in order to ensure convexity of the transformation function. Notice that, in previous versions of the model, a symmetric transformation
function was defined
F (X) = kXk − R,
(3.170)
3.3 Modelling of shape memory alloys
157
and R was defined in terms of the forward and reverse transformation temperatures:
R = k (Af − Mf ) .
(3.171)
As in associated plasticity, the transformation function and the plastic potentials
coincide and the flow rule for the transformation strain is the following:
ε̇tr = ζ̇
∂F (X)
,
∂σ
(3.172)
along with the Kuhn-Tucker condition
ζ̇ ≥ 0, F ≤ 0, ζ̇F = 0.
(3.173)
The role of the material parameters introduced in the Auricchio and Petrini
(2004a) model is discussed here. As the model does not distinguish between
austenite and twinned martensite, a single set of elastic and thermal properties is
considered: (K, G, α, c, s0 , e0 ). In particular, the entropy difference ∆s0 between
the two phases, which in the model by Popov and Lagoudas (2007) was seen
to be responsible for the slope of the transformation lines in the uniaxial phase
diagram (see Equation (3.159)), is not portrayed in this model. Its role is played
by the parameter k: for temperatures above Mf , indeed, the centre of the elastic
domain linearly depends on the temperature. For temperatures below Mf , on
the other hand, the position of the elastic domain is independent of temperature,
thus representing the horizontal martensite detwinning line. The parameter h
introduces linear kinematic hardening related to the amount of transformation
strain, while m and R define the radius of the elastic domain. The role of
each material parameters is clarified in Figure 3.16(a). Here, the white area
represents the elastic domain for positive transformation strain in the uniaxial
stress-temperature diagram (for negative transformation strain, the slope of the
lines is −k).
As it can be seen, the uniaxial diagram associated to this model is quite
different from most phase diagrams (see for example Figure 3.16(b)). As the
austenite to twinned martensite transformation is not described, the vertical
transformation strip present in most phase diagrams disappears. Moreover,
while the hysteresis associated to detwinning and phase transformation (dd and
dtr , respectively) are generally different, in this model they are both defined
by the choice of the parameters m and R. Indeed, in different versions of
158
Shape memory alloys
s
s
h
(m,R)
d tr
sft
k
3
sst
h
dd
(m,R)
Mf
T
As
Af
4
Mf M
s
ssc
T
3
sfc
(a) Auricchio and Petrini (2004a) model
(b) Typical phase diagram
Figure 3.16: Uniaxial phase diagrams
the model these parameters have been defined either in terms of the tensioncompression critical stresses for detwinning, as in Equation (3.169), or the phase
transformation hysteresis, as in Equation (3.171). Finally, for temperatures
between Mf and pure superelastic behaviour, the critical stress for the reverse
transformation (or detwinning) reaches very low compressive values which are
not accounted for in any other phase diagram.
3.4
Experimental characterisation and modified
phase diagram
Superelasticity and the shape memory effect are only two examples of the variety of responses displayed by shape memory alloys subjected to a general
stress-strain-temperature loading path. Most constitutive models present in the
literature, including those discussed in the previous section, predict the shape
memory alloy behaviour in complex loading based on a phase diagram, which
can be experimentally characterised with a relatively small number of simple
tests, usually performed in constant stress or constant temperature conditions.
A phase diagram which can accurately describe the shape memory alloy behav-
3.4 Experimental characterisation and modified phase diagram
159
iour in complex loading is particularly crucial when dealing with shape memory
alloys embedded in a host structure, whose loading history depends upon the
relative stiffness of the shape memory and host structure.
The phase diagram proposed by Brinson (1993) is still the most common
in the literature. Recently, however, some attempts have been made to define
modified phase diagrams taking into account other features of the shape memory
alloy behaviour (He et al., 2006; Popov and Lagoudas, 2007). In particular, two
assumptions underlying the Brinson’s phase diagram are questioned, namely
the existence of a limit stress value for the formation of detwinned martensite
and the definition of the reverse martensite to austenite transformation strip.
In the following, the two mentioned issues are discussed based on experiments carried out on untrained Nitinol wires. Though further evidence of the
phenomena discussed in the following can be found in recent literature works,
such material behaviour is generally not incorporated in constitutive models.
The present experimental work is thus intended as a starting point for the development of a new constitutive model incorporating the observed phenomena.
3.4.1
Detwinned martensite formation at low stress levels
Shape memory alloy wires embedded in a composite matrix experience a complex loading history. Indeed, in this case shape recovery is totally or partially
hindered by the presence of the matrix. A limit condition which can be used
to simulate embedded shape memory alloys is to completely prevent the wires
deformation. This condition is usually termed as constrained recovery.
In constrained recovery, the shape memory alloy is deformed at low temperature and then externally constrained. As the temperature increases, the
reverse phase transition starts and a recovery stress is developed in the material
as it tries to recover its original shape (slope 2-3). During cooling in presence
of stress, detwinned martensite is reformed and the recovery stress is totally
or partially released (slope 4-5). Figure 3.17 reports two schematic constrained
recovery loading paths as depicted in the Brinson’s model phase diagram. In
Figure 3.17(a), the material is loaded and subjected to a constrained temperature cycle, while in Figure 3.17(b) the material is prestrained and then unloaded
prior to the temperature cycle.
The stress developed in constrained recovery is crucial in active composites
160
Shape memory alloys
s
s
1
2
4
sf
1
2
3
4
3
sf
1=6
5
2
6
ss
5
ss
4
1
Mf Ms A s
(a)
Af
T
4
2
Mf Ms A s
Af
T
(b)
Figure 3.17: Constrained recovery sketch in Brinson’s phase diagram
applications, as it is discussed in Chapter 4. The amount of recovery stress developed during cooling and its release during heating are important, for example,
in structural shape control. Indeed, if the stress is not completely released, the
original structural shape cannot be fully recovered (see Figure 4.3 on the right
in Chapter 4).
Some experimental data concerning constrained recovery are available in the
literature (Prahlad and Chopra, 2001), however most of them concentrate on
the heating branch (Li et al., 2003). In the following, some tests on Nitinol wires
in constrained recovery are reported. The tests were carried out at the Center
for Intelligent Material Systems and Structures of the Virginia Polytechnic Institute and State University. The results presented here are intended only as
preliminary observations on low temperature phenomena in constrained shape
memory alloys, while a thorough experimental investigation is, to the author’s
knowledge, still lacking.
Results regarding prestrained and prestrained-unloaded samples respectively
are reported in Figure 3.18. As it can be seen, in both cases the stress was
completely released upon cooling. This suggests that in constrained recovery
detwinned martensite is formed even at very low stress levels. According to
Popov and Lagoudas (2007), the ability to develop detwinned martensite at low
stress levels is related to the shape memory alloy training.
3.4 Experimental characterisation and modified phase diagram
(a)
(b)
Figure 3.18: Nitinol wires in constrained recovery
161
162
Shape memory alloys
Most constitutive models discussed in the previous section are not able to
model complete release of the recovery stress, as most phase diagrams envisage a
limit stress below which no further detwinned martensite can be formed. An exception is the model by Popov and Lagoudas (2007), in which the critical stress
for detwinned martensite formation can be arbitrarily introduced or removed
thanks to the independence of the twinned and detwinned martensite transition
zones (see Section 3.3). As one-dimensional models are still very popular in the
engineering literature, a simple modification to the Brinson’s model allowing to
include constrained recovery stress release at low stress levels is here proposed
(Daghia et al., 2006).
According to Brinson’s model, the residual stress at cooling is always greater
than σs . Consider Figure 3.17: once the temperature falls below Ms , a further
increase in detwinned martensite can only occur if the stress increases; however,
due to the constant strain condition, more detwinned martensite would cause
the stress to decrease, therefore the only possible equilibrium is at constant
stress and constant detwinned martensite. Between Ms and Mf the phase
transformation continues, but the martensite developed is twinned and does
not contribute to the change in the shape memory alloy stress state.
The required modification can be introduced without changing the shape of
the phase diagram, by admitting a certain degree of detwinning in the temperature martensite formed in the presence of stress. This amounts to changing the
kinetic laws for the vertical transformation strip. The new kinetic laws must
satisfy some requirements:
• detwinned martensite is formed only as a part of the temperature martensite and it is not originated by stress;
• at zero stress there should be no detwinned martensite formed.
Given the listed criteria, the natural way of describing the detwinned martensite increase between Ms and Mf is as a function of the temperature induced
martensite and the stress level. The temperature induced martensite is given
by Equation (3.25) and it can be expressed as
ξ t = ξ0t + f (T ) .
(3.174)
Assuming that, depending on the applied stress, a part of the newly formed
3.4 Experimental characterisation and modified phase diagram
163
Table 3.2: Nitinol material properties
Mf
◦
C
-12
CA
MPa/◦ C
6.2
CM
MPa/◦ C
6.2
σs
MPa
70
Ms
◦
C
20
As
C
28
◦
σf
MPa
170
Af
◦
C
43
EM
GPa
11
EA
GPa
28
Θ
MPa/◦ C
0
εL
0.065
martensite is detwinned, the previous equation can be modified as follows:
ξ d = ξ0d + g (σ) f (T ) , ξ t = ξ0t + [1 − g (σ)] f (T ) ,
(3.175)
where g (σ) is a function of the applied stress, which needs to be null at σ = 0
to ensure satisfaction of the second requirement. A simple choice for g (σ) can
be taken as
½
g (σ) =
1
σ/σs
if σs ≤ σ ≤ σf
,
if σ < σs
(3.176)
however different functions could be chosen according to the experimental data,
introducing or removing the limit stress for detwinned martensite formation.
The original and modified Brinson’s models are used to simulate the observed
experimental results. The shape memory alloy material parameters used in the
simulation are reported in Table 3.2 (notice that a mean elastic modulus was
considered for simplicity). The results are reported in Figure 3.19. As it can be
seen, the modified Brinson’s model is able to capture the constrained recovery
stress relaxation below Ms , while the original Brinson’s model shows a residual
recovery stress which cannot be eliminated by further cooling.
164
Shape memory alloys
(a)
(b)
Figure 3.19: Comparison between original and modified Brinson’s models
3.4 Experimental characterisation and modified phase diagram
3.4.2
165
Influence of the loading path on the reverse phase
transition temperatures
Recently, a number of researchers have started to question the assumption on the
existence of a single transformation strip for the martensite to austenite phase
transition. Indeed, modified phase diagrams have been proposed (He et al.,
2006; Popov and Lagoudas, 2007) displaying two separate strips for the twinned
and detwinned martensite transitions into austenite. In particular, zero-stress
phase transition temperatures for detwinned martensite are generally higher
than those for twinned martensite.
The modification to the phase diagram is motivated by calorimetric observations. Popov (2005) reports DSC measurement on various shape memory
alloy specimens, both trained and untrained. The wires, initially in twinned
martensite state, were mechanically loaded and unloaded at room temperature
to achieve complete detwinning, then DSC tests were performed starting with
a heating ramp to achieve complete phase transition into austenite. The DSC
results consistently showed an increase in the reverse phase transition temperatures during the first heating, in which the detwinned martensite was transformed into austenite. The subsequent cooling and heating cycles, involving
only austenite and twinned martensite, were consistent with those observed for
the wires which had not been subjected to mechanical loading.
A similar test was carried out within the present experimental campaign,
yielding results which were consistent with those reported by Popov (2005). A
DSC test was performed on a 0.5 mm diameter shape memory alloy wire which
had been mechanically strained to 4% elongation. Details on the wire preparation and the test itself are reported later in this section, while the result of the
DSC test is reported in Figure 3.20. Due to the presence of the intermediate
austenite to R-phase transition, the first and second cooling cycles were interrupted at 30◦ C and -20◦ C, respectively. This allowed to observe the R-phase
to austenite transition (R → A) during the second heating and the twinned
martensite to austenite transition (M t → A) during the third heating.
As anticipated, these results are consistent with those reported by Popov
(2005). In particular, the trough in the black line (first heating, M d → A)
is significantly shifted to the right, while the cooling curves are completely
superimposed. The two separate peaks in cooling represent the A → R and R →
166
Shape memory alloys
t
Heat flow (W/g)
R®M
cooling
heating
A®R
R®A
t
M ®A
first cycle
second cycle
third cycle
0
10
20
30
40
50
60
Temperature (°C)
d
M ®A
70
80
90
Figure 3.20: DSC for shape memory alloy wire with 4% elongation
M t transitions, respectively. The R → A and M t → A troughs in the second
and third heating are consistent with those obtained for the virgin specimen.
Both the results reported by Popov (2005) and the present experiment show
a shift in the zero-stress reverse phase transition temperatures related to the
different loading history of the material (virgin specimen versus mechanically
loaded specimen). The effect of the mechanical loading on the M → A transition, however, is temporary as it disappears once the material is heated above
the (current) austenite finish temperature during the first DSC heating. The
observed phenomenon is thus generally different from the permanent change
of the phase transition temperatures and enthalpy of the transformation documented in experimental and modelling studies concerning the effects of training
(Auricchio et al., 2003; Miller and Lagoudas, 2000; Wada and Liu, 2005). In
order to incorporate the non-uniqueness of zero-stress reverse phase transition
temperatures into a modified phase diagram, a thorough experimental investigation is required, including partially detwinned shape memory alloy specimens
and a variety of complex stress-temperature loading histories.
3.4 Experimental characterisation and modified phase diagram
167
A®R
Heat flow
R®M
t
t
R,M ®A
0
10
20
30
40
50
60
Temperature (°C)
70
80
90
Figure 3.21: DSC test for the virgin specimen
Experimental setup
To further investigate the effect of the loading history on the zero-stress reverse
transformation temperatures, DSC measurements were performed on untrained
Nitinol wires subjected to different loading histories. The experiments were
carried out at the Institute for Composite and Biomedical Materials of the
Italian National Research Council (CNR).
Shape memory alloy wire (diameter 0.5 mm) was provided by Memory Metalle GmbH7 . The virgin material characteristic temperatures were determined
via DSC (Figure 3.21). In particular, the formation of R-phase during cooling
is noticed.
Prior to testing, the wires were heated to 100◦ C to remove the effects of
packaging and handling. As noticed in the virgin specimen DSC, upon cooling
at room temperature the wires are in the R-phase and further cooling is required
to obtain a twinned martensite structure. For this reason, all the described
experiments were carried out on two different sets of wires, whose initial crystal
7 www.memory-metalle.de
168
Shape memory alloys
structure is composed of martensite and R-phase, respectively.
Three different loading histories were considered:
1. mechanical loading to a given elongation at room temperature;
2. thermal cycling at constant stress (restrained recovery);
3. thermal cycling at constant strain (constrained recovery).
The first and third tests were carried out using an Instron 8501 machine with
thermal chamber. In test 1, the wire was loaded and unloaded at room temperature, leaving a residual deformation; in test 3, the wire was first loaded
at room temperature to a given strain, then heated above the phase transition
temperature and cooled to room temperature while keeping the strain constant,
finally unloaded at room temperature. In test 2, the constant load was simply
applied by hanging a weight to the wire. The wire was heated electrically and
allowed to cool to room temperature, then unloaded.
The following DSC test was carried out on material taken from each wire:
• stabilise at 25.00◦ C;
• heat to 120.00◦ C at 10.00◦ C/min;
• cool to 30.00◦ C at 10.00◦ C/min;
• heat to 120.00◦ C at 10.00◦ C/min;
• cool to −20.00◦ C at 10.00◦ C/min;
• heat to 120.00◦ C at 10.00◦ C/min.
The behaviour after the first heating was quite repeatable, confirming the temporary effect of the loading history, which is erased with a single heating at zero
stress. For this reason, only the first heating is reported in the following DSC
graphs, while an example of the complete cycle was reported earlier in this section (Figure 3.20). As the width of the peaks varies considerably in the different
tests, the temperature value considered for comparison is the peak temperature.
3.4 Experimental characterisation and modified phase diagram
169
Test 1: deformation at room temperature
The first set of results concerns wires that have been strained at room temperature before the DSC test. The experiment previously discussed and reported
in Figure 3.20 belongs to this group. The first heating curve of the first and
second sets of wires (martensite and R-phase) are reported in Figure 3.22(a) and
3.22(b), respectively. Figure 3.23, moreover, reports the strain plotted against
the peak temperature values for all considered specimens.
As it can be seen from the graphs, reverse phase transition temperatures
increase with increasing strain. The martensite and R-phase wires follow a
similar trend, although the peak temperatures for the same strain are generally
different in the two cases. This difference is probably related to the different
phenomena occurring within the material during mechanical loading (R → M d
phase transition versus M t → M d detwinning). Indeed, at 10% strain both
detwinning and phase transition are over, the wires are in detwinned martensite
state and their austenite transition temperatures are the same.
On the other hand, the greatest difference between the martensite and Rphase wires can be observed in the 2% strain graphs. For ease of comparison,
they are both reported in Figure 3.24. As the R → M d phase transition is
not complete at 2% strain, the R-phase graph shows two troughs, denoting the
R → A (left trough, peak temperature 52.7◦ C) and the M → A (right trough,
peak temperature 61.5◦ C). On the other hand, the martensite graph shows a
single trough (peak temperature 66.0◦ C), not allowing to distinguish between
twinned and detwinned martensite.
Mechanical deformation at low temperature induces a stabilisation of the
martensite phase, demonstrated by the increase in the reverse phase transition
temperatures in the reported tests. In this case, the amount of stabilisation
appears to increase as the strain increases. Martensite stabilisation due to cold
deformation was previously observed, among others, by Liu and Favier (2000).
The interpretation and consequences on the phase diagram modification are
discussed later in this section.
Test 2: thermal cycling at constant stress
The second set of experiments is aimed at investigating the effect of temperature
on the martensite stabilisation. A thermal cycle is performed on wires subjected
Shape memory alloys
Heat flow
170
2% strain
4% strain
6% strain
8% strain
10% strain
40
50
60
70
80
Temperature (°C)
90
100
90
100
Heat flow
(a) Martensite specimens
2% strain
4% strain
6% strain
8% strain
10% strain
40
50
60
70
80
Temperature (°C)
(b) R-phase specimens
Figure 3.22: First heating curve for specimens strained at room temperature
3.4 Experimental characterisation and modified phase diagram
171
90
85
Temperature (°C)
80
75
70
65
60
martensite
55
50
R-phase
0
2
4
6
Strain (%)
8
10
12
Heat flow
Figure 3.23: Strain versus peak temperature values for test 1
martensite
R-phase
40
50
60
70
80
Temperature (°C)
90
100
Figure 3.24: First heating curve for specimens strained at room temperature detail of 2% strain
Shape memory alloys
Heat flow
172
room T
°
T
= 90 C
T
= 110 C
T
= 130 C
max
°
max
°
max
40
50
60
70
80
Temperature (°C)
90
100
90
100
Heat flow
(a) Martensite specimens
room T
= 90 C
T
= 110 C
T
= 130 C
max
max
max
40
°
T
50
°
°
60
70
80
Temperature (°C)
(b) R-phase specimens
Figure 3.25: First heating curve for specimens thermally cycled under 200 MPa
constant stress
3.4 Experimental characterisation and modified phase diagram
173
80
Temperature (°C)
75
70
martensite
65
R-phase
room T
°
Tmax= 90 C
60
Tmax= 110 ° C
°
Tmax= 130 C
55
0
50
100
150
200
250
Stress (MPa)
300
350
400
Figure 3.26: Applied stress versus peak temperature values for test 2
to a constant load. During heating, the detwinned martensite initially obtained
with the application of the load transforms into austenite and the wire recovers
the initial deformation, working against the applied load. During cooling under
constant load, austenite transforms back into detwinned martensite because
of the temperature change. Upon load removal, therefore, the wire is in a
detwinned martensite state and the DSC test allows to measure the zero-stress
phase transition temperatures of martensite formed under constant loading.
Four wires were considered for each load value. The first was loaded at
room temperature and no thermal cycle was carried out; the second, third and
four wires were heated to a maximum temperature of 90◦ C, 110◦ C and 130◦ C,
respectively. This amounts to a significant number of tests, thus only some of
the DSC results obtained are reported here. In particular, Figures 3.25(a) and
3.25(b) report the results for 200 MPa constant stress and different maximum
temperatures for the first and second sets of wires (martensite and R-phase),
respectively. It is interesting to notice that the DSC curves are nearly superimposed and they are similar for martensite and R-phase specimens. It should
be noted that the peak temperature observed in Figures 3.25(a) and 3.25(b) is
174
Shape memory alloys
similar to the one observed in test 1 for a 4% strain. These two conditions (200
MPa stress and 4% strain) both correspond to the end of the loading plateau,
demonstrating a coherence between the results observed in tests 1 and 2.
To further investigate the effect of the applied stress on the reverse phase
transition temperatures, Figure 3.26 reports the applied stress against peak
temperature values for all considered specimens. The graph clearly shows that
the peak temperature tends to increase with increasing applied stress, regardless
of the thermal cycling. The data dispersion could be decreased by performing
more controlled experiments, recording also the deformation associated to the
thermal cycling.
Test 3: thermal cycling at constant strain
The third set of experiments concerns wires that have been subjected to a constrained recovery cycle prior to DSC testing. As already discussed in Section
3.4.1, heating and cooling at constant strain constitutes a complex loading history for the shape memory alloy wires. Indeed, the two driving forces of the
martensitic phase transition, stress and temperature, vary contemporarily during constrained recovery, thus making it difficult to separate their effects. On the
other hand, an accurate interpretation of the constrained recovery behaviour of
shape memory alloys in terms of phase transition temperatures is crucial when
dealing with shape memory alloy hybrid composites, as discussed in Chapter 4.
In order to evaluate the effect of a constrained recovery cycle on the shape
memory alloy reverse phase transition temperatures, the DSC measurements
obtained in test 3 are compared to those obtained in test 1 at the same level
of prestrain. Figures 3.27(a) and 3.27(b) show such comparison for martensite
and R-phase wires mechanically strained by 4%. The third heating curve is also
reported in the graphs as representative of the virgin specimen behaviour. As
it can be seen, the reverse phase transition temperatures for the constrained
recovery specimen fall between those for the virgin specimen and those for the
specimen strained at room temperature. Based on this observation, a thermal cycle at constant deformation appears to partially remove the martensite
stabilisation related to the cold deformation which was observed in test 1.
In relation with the complex loading history given to the wires, the DSC
results for test 3 showed some inconsistencies which were not observed in tests 1
and 2. In particular, in some cases DSC measurements carried out on specimens
175
Heat flow
3.4 Experimental characterisation and modified phase diagram
room T
T
max
°
= 120 C
third heating
40
50
60
70
80
Temperature (°C)
90
100
90
100
Heat flow
(a) Martensite specimens
room T
T
max
°
= 120 C
third heating
40
50
60
70
80
Temperature (°C)
(b) R-phase specimens
Figure 3.27: First heating curve for specimens thermally cycled at 4% constant
strain
176
Shape memory alloys
taken from the same wire yielded different results on the first DSC heating in
terms of position and depth of the phase transition peak, thus making it difficult
to univocally characterise the wires transition temperatures. A possible interpretation is related to the dominance of local effects due to the absence of global
deformation. Indeed, both tests 1 and 2 involved macroscopic deformation of
the shape memory alloy wires, thus the phase transition, initially occurring by
nucleation in restricted zones of the wires, ended up involving the whole wire. In
test 3, on the other hand, no global deformation was allowed because of the constraint, thus local effects tend to be more evident. As the present investigation
is aimed at characterising the shape memory alloy behaviour at the macroscale,
the significance of local effects could be diminished by evaluating the reverse
phase transition temperatures using techniques which involve the whole wire,
such as the resistivity measurements investigated by Antonucci et al. (2007).
Further results on the calorimetric characterisation of constrained recovery
shape memory alloys can be found in the work by Li et al. (2003), who performed DSC and recovery stress measurements on Nitinol wires which had been
mechanically strained and cycled in constrained recovery to different maximum
temperatures. Their DSC graphs display two separate peaks during the first
heating. The lower temperature peak occurred at temperature similar to those
observed for the virgin specimen, while the higher temperature peak occurred
at temperatures above those observed for the wire which had been mechanically
strained at room temperature. As the maximum temperature reached during
constrained recovery increases, the lower peak becomes more evident and the
higher peak tends to disappear. Thus, the lower peak appears to be related
to the martensite fraction which is involved in the phase transitions occurring
during constrained recovery. The higher peak, on the other hand, is related
to a residual amount of martensite which formed during the room temperature
prestrain prior to the constrained recovery cycle. The occurrence of two different phase transitions at different temperatures is confirmed by recovery stress
measurements, showing a two step increase of the recovery stress during the
second heating of a constrained recovery wire.
Interpretations and modelling considerations
The present experimental investigation shows that the zero stress reverse phase
transition temperatures for untrained shape memory alloys are not constant but
3.4 Experimental characterisation and modified phase diagram
177
significantly depend upon the loading history of the material. Summarising the
results of the three tests:
• in test 1, the reverse phase transition temperatures are shown to increase
gradually with increasing strain;
• in test 2, the reverse phase transition temperatures are shown to increase
gradually with increasing applied stress, and the thermal cycling at constant applied stress does not seem to significantly affect the observed shift;
• in test 3, the reverse phase transition temperatures are lower than those
observed in test 1 for the same strain; moreover, literature results document the presence of two different phase transition peaks, with the lowest
peak becoming more evident as the amount of martensite involved in the
constrained recovery phase transitions increased.
From the results summarised above, it is clear that the definition of two
sets of phase transition temperatures for twinned and detwinned martensite, as
proposed by He et al. (2006) and Popov and Lagoudas (2007) in their modified
phase diagrams, is not sufficient to account for the extreme variability of the
phase transition temperatures. Indeed, the twinned and detwinned martensite
fraction cannot be clearly distinguished via DSC measurements of partially detwinned specimens, such as the one reported in 3.24. In that case, the R-phase
and detwinned martensite are perfectly distinguishable (light grey line), whereas
a single peak is observed for specimen containing partially detwinned martensite
(black line).
The change in the reverse phase transition temperatures appears to be continuous and governed by the global deformation in the shape memory alloy.
Indeed, the peak temperature shift is clearly observed in both tests 1 and 2, involving formation of detwinned martensite under constant temperature or constant stress conditions. In test 3, involving formation of detwinned martensite
under constant strain conditions, the stabilisation effect related to the mechanical straining appears to have been partially removed. The literature results
for test 3, displaying two separate peaks in the DSC measurements and a two
step actuation in the recovery stress measurements, suggest the presence of two
different phase transitions involving the martensite formed during mechanical
straining at room temperature and the martensite formed during cooling in constrained recovery, respectively. An increase of the recovery stress was observed
178
Shape memory alloys
during both phase transitions, thus both martensite fractions are at least partially detwinned. The increase of recovery stress is indeed a global indicator of
detwinned martensite phase transition, as such it can be used to characterise
the macroscopic behaviour of shape memory alloys in constrained recovery. A
similar two step actuation was observed in the experimental testing of a shape
memory alloy hybrid composite beam, discussed in Section 4.2.1. In that section, the behaviour of the composite is interpreted in light of the observations
discussed here and the proposed interpretation is tested by simulating the composite behaviour and introducing a simple modification to the Brinson’s model.
The simulation is found to capture the main features of the composite behaviour.
A number of interpretations were suggested for the mechanical stabilisation
of martensite related to deformation, based on the development of plastic deformation and internal stresses (see, for example, Liu and Favier, 2000). To
the author’s knowledge, however, the effects of the different loading histories
considered here have not been fully addressed. The present experimental investigation involves mostly macroscopic measurements, thus it is not sufficient to
develop and support a complete micromechanical interpretation of the observed
phenomena. However, a preliminary explanation is proposed based on the comparison between the different tests. If the formation of detwinned martensite is
accompanied by a global deformation of the wire, a more ordered microstructure is obtained, thus a higher temperature is required to achieve the reverse
phase transition. Viceversa, detwinning associated to the martensite formed
during constrained recovery cooling occurs to accommodate a preexistent wire
deformation, thus the resulting microstructure is similar in some sense to the
twinned martensite formation under zero load and the reverse phase transition
temperatures vary accordingly. In order to further investigate this interpretation, microscopic observations or similar experimental techniques could be
employed to observe the microstructures associated to each loading history.
As regards the modified phase diagram, the present results suggest to account for the change in the reverse phase transition temperatures by redefining
the parameters As and Af as functions of the material loading history. The
definition of such functional relations, requiring the support of further experimental investigation and an accurate micromechanical interpretation, would
involve the introduction of new internal variables keeping track of the history
in terms of global deformation of the shape memory alloy material.
Capitolo 4 – Sommario
179
Compositi attivi con fili a memoria di forma
I compositi combinano materiali diversi per ottenere strutture con proprietà
uniche. Grazie al loro comportamento accoppiato, i materiali attivi sono in
grado di reagire a cambiamenti ambientali modificando le proprie caratteristiche.
L’integrazione di materiali attivi all’interno di compositi permette di creare
strutture adattive, in cui le capacità di sensore ed attuatore sono trasferite dal
materiale attivo alla struttura stessa.
A partire dai primi anni ’90 è stata proposta la realizzazione di strutture
composite con integrati elementi a memoria di forma. Come nei compositi tradizionali, la matrice ed il rinforzo possono variare significativamente a seconda
dell’applicazione. Nel seguito, l’attenzione viene posta soprattutto su compositi
fibro-rinforzati con fili o nastri a memoria di forma integrati.
I materiali a memoria di forma possiedono una significativa capacità di smorzamento, utilizzabile per il controllo passivo delle vibrazioni o l’aumento della
resistenza all’impatto. La maggior parte delle applicazioni dei compositi attivi
si basa però sul recupero attivo della forma, ottenuto in seguito ad un riscaldamento. In questo caso, elementi a memoria di forma sono predeformati ed
inseriti nella struttura in fase martensitica. Al riscaldamento, si attiva la transizione di fase ma il recupero della forma è impedito totalmente o parzialmente
da vincoli esterni o dal composito stesso. Gli sforzi che si sviluppano possono
essere utilizzati per controllare la rigidezza o la forma della struttura.
Il controllo della rigidezza avviene tramite elementi vincolati esternamente,
la cui tensione varia al variare della temperatura. Tale meccanismo permette
di influenzare la stabilità, il comportamento dinamico e la risposta all’impatto
della struttura. Variando la rigidezza a seconda della necessità è possibile,
ad esempio, modificare le frequenze proprie della struttura per allontanare il
rischio di risonanza al variare dell’eccitazione esterna. Un’altra applicazione
particolarmente promettente è il controllo dell’instabilità termica: in questo
caso il riscaldamento costituisce la fonte di instabilità e allo stesso tempo genera
sforzi di recupero nei materiali a memoria di forma, che fungono sia da sensori
che da attuatori.
Il controllo della forma è una possibilità unica fornita dai materiali attivi.
In questo caso, gli elementi a memoria di forma sono vincolati al composito,
che impedisce in parte il recupero della forma attivato durante il riscaldamento.
180
Capitolo 4 – Sommario
Le tensioni generate negli elementi attivi sono trasferite alla struttura, dando
origine ad una deformazione che dipende dalla rigidezza della struttura stessa e
dalla temperatura e posizione degli elementi a memoria di forma. Al raffreddamento, la struttura fornisce la forza di ritorno al materiale a memoria di forma
e la configurazione originale è totalmente o parzialmente recuperata.
In questo capitolo viene descritto, con un approccio comune, il controllo della
forma e della rigidezza. L’attenzione viene concentrata su elementi strutturali
tipo piastra, modificando la posizione e il vincolamento degli elementi a memoria
di forma per ottenere diverse configurazioni. Nella sezione 4.1 si propone un
modello ad elementi finiti per i compositi ibridi, mentre in 4.2 viene presentato
uno studio sperimentale di piastre ibride in controllo di forma. Lo scopo è di
definire uno strumento semplice e flessibile per la progettazione dei compositi
ibridi in grado di prevedere il comportamento strutturale in controllo di forma
o di rigidezza. Il modello è validato considerando il controllo della forma perché
costituisce la configurazione più sensibile alla scelta dei parametri del modello.
Chapter 4
Active composites with
shape memory alloy wires
Composites combine different materials to achieve structures with unique properties. Due to their coupled behaviour, active materials are able to sense changes
in the environment and actively respond to them by modifying their properties.
Embedding active materials within a composite allows to create adaptive structures, in which the sensing and actuating capabilities are transferred from the
active material to the structure itself. Some state of the art examples of active
composites are morphing structures, self-healing structures and power harvesting devices.
Since the early 1990s, the concept of embedding shape memory alloy elements in a composite has been investigated for various uses. Composites with
embedded shape memory alloys are usually termed as shape memory alloy hybrid composites, or SMAHC. As in traditional composites, both matrix and reinforcement can considerably vary according to the sought application. Metal,
polymer and concrete matrix composites with embedded shape memory alloy
particles (Lòpez et al., 2007), short fibres (Zhang et al., 2007) or long fibres
all fall into the broad definition of SMAHC (Wei et al., 1998). In the following, we concentrate mainly on fibre-reinforced composites with embedded shape
memory alloy long wires or ribbons.
Besides displaying an active behaviour related to thermomechanical cou-
182
Active composites with shape memory alloy wires
pling, shape memory alloys have a significant damping capacity in both superelastic and shape memory range, which can be exploited to enhance the composite’s energy dissipation in vibration control (Birman, 2008) or impact resistance
(Meo et al., 2005). However, most SMAHC applications make active use of the
phase transition which lies behind the shape memory alloy functional properties.
The martensite to austenite transition allows to change the SMAHC properties
according to two different mechanisms, namely active property tuning and active
strain energy tuning.
Active property tuning exploits the different properties, mainly elastic modulus, of the two phases to change the composite stiffness. In this case, no shape
recovery or macroscopic deformation is associated to the phase transition. This
concept requires a large volume fraction of shape memory alloy material within
the composite to achieve a significant stiffness change, as the ratio of the elastic
moduli of the two phases is not large (usually EA /EM ' 3 for Nitinol). Active strain energy tuning, on the other hand, makes use of constrained shape
recovery to develop large recovery stresses in the shape memory alloy, which
influence the overall behaviour of the structure. In this case, shape memory
alloy elements are prestrained and embedded in the composite in the martensitic state. Upon heating, the phase transition is triggered but shape recovery
is totally or partially hindered either by external constraints or the composite
itself. The recovery stresses which develop can be used to control the stiffness
or the shape of the structure.
Controlling structural stiffness allows to modify properties such as the buckling load, dynamic behaviour (Ostachowicz et al., 2000) and impact behaviour
(Khalili et al., 2007b) of a structure. Indeed, buckling and vibration control
applications are among the earliest studies on SMAHC (Baz and Chen, 1993;
Baz et al., 1992; Ro and Baz, 1995a,b,c). In stiffness control, shape memory
alloy wires are pre-elongated and externally constrained, thus tensile stress is
generated during actuation. Due to the external constraints, the stress is not
transferred to the composite but the taut wires contribute to increase the overall
structural stiffness. Using the terminology introduced by Khalili et al. (2007a),
this contribution can be classified as acquired stiffness, as opposed to the essential stiffness of the composite, which is due to the material characteristics only
and is not influenced by external events. The shape memory alloy wires can be
directly bonded to the composite or set within sleeves, either inside the com-
183
posite or on its surface. The sleeves, in turn, could be bonded to the composite
continuously or at discrete points (Birman, 2007, 2008). The use of sleeves removes a number of technological issues, such as surface adhesion between the
shape memory alloy wires and composite, deterioration of the shape memory
effect and heating of the polymer matrix.
Notice that taut wires made of traditional materials, such as steel wires,
could also be used to increase structural stiffness. However, the advantage of
shape memory alloys in this context is two-fold. As the wires recovery stress
varies with temperature, the stiffness can be tuned according to need, whereas
the increase in stiffness related to traditional reinforcement is fixed. This allows
for example to shift the structure’s natural frequencies away from resonance as
the external excitation changes, while also taking advantage of the additional
damping introduced by shape memory alloys elements (Aoki and Shimamoto,
2003; Zhang et al., 2006). The controlled shifting of resonant frequencies is
particularly advantageous in the design of rotating machines (see, for example, Żak and Cartmell, 2002, who used shape memory alloy wires to change
the stiffness of a rotor bearing). Moreover, in some applications environmental
changes could constitute the source of actuation. An example is thermal buckling, where a temperature increase is the source of the instability and at the
same time it triggers the shape memory alloy actuation, engendering a stiffness
increase only when needed without the aid of external actuation mechanisms.
In such cases, shape memory alloy stiffness control could be advantageous with
respect to traditional stiffeners (Birman, 1997).
Active shape control of a structure by embedded actuators is a unique possibility offered by active materials. Shape control with embedded shape memory
alloys requires transfer of the stress between the wires and the host structure.
As such, shape memory alloy wires are usually embedded directly in the composite. Upon actuation, shape recovery of the wires is partially hindered by the
composite itself, acting as a constraint. The recovery stress is transferred from
the shape memory alloy wires to the host structure, causing a shape change
which depends upon the composite stiffness, amount of recovery stress and wire
positioning within the composite. During cooling, the original shape is totally
or partially recovered as the host structure acts as a bias element to the shape
memory alloy wires. With appropriate design of the SMAHC, different shape
changes can be achieved. Out-of-plane bending of a plate can be simply obtained
184
Active composites with shape memory alloy wires
by embedding shape memory alloy wires at a distance from the plate midplane,
whereas composite lay-up can be designed to engender coupled bending-torsion
with a shape memory alloy bender element (Chandra, 2001). Obviously, the
amount of shape change that can be achieved strongly depends upon the stiffness of the host structure and its deformation before fracture. As such, a quite
different bending actuation behaviour can be obtained using elastomer (Lind
and Doumanidis, 2003), fibre-reinforced polymer (Turner et al., 2006) or concrete (Deng et al., 2006) as host structure.
The compressive recovery stresses that shape memory alloy wires can transmit to the host structure can be used for damage repair. SMAHC patches can
be used to reduce stress concentration in the adhesive layer in composite joints
(Chen et al., 2007) and to provide closure stress for composite repair (Wang,
2002), whereas in self-healing composites the cracks in the host structure are
repaired by the reverse transformation and shape recovery of embedded shape
memory alloy wires (Araki et al., 2002; Burton et al., 2006; Li et al., 2007).
Recently, this idea has been applied to crack closure in shape memory alloy
reinforced concrete beams (Pascale et al., 2008).
In this chapter, stiffness and shape control applications are discussed within
a unified framework. The attention is focused on SMAHC plate structures,
in which stiffness and shape control can be achieved by changing the boundary
conditions and the position of the shape memory alloy wires relative to the plate
midplane. In particular, wires embedded at a certain distance from the midplane
force the plate to bend. A finite element model for SMAHC is presented in
Section 4.1 and an experimental investigation of the bending control of SMAHC
plates is carried out in Section 4.2. The aim is to provide a simple and flexible
design tool which is able to predict the structural behaviour in both stiffness
and shape control. The model validation, however, is carried out by focusing
on the shape control configuration, as it is more sensitive to the choice of the
model parameters.
4.1
Modelling SMAHC
A number of approaches are described in the previous chapters regarding the
modelling of traditional fibre-reinforced composites and shape memory alloy
materials. As it should be clear from the discussion, both of them are still open
4.1 Modelling SMAHC
185
research problems and even more so is the modelling of hybrid composites incorporating active materials. In this case, indeed, the modelling issues related
to the interaction of different materials, typical of composites, are further complicated by the coupling in the active material constitutive laws. On the other
hand, a simple and flexible model is required to allow the design and control of
a hybrid composite.
As regards SMAHC, a wide variety of modelling approaches can be found
in the literature. As embedded shape memory alloys are generally in a nearly
constrained recovery configuration, the simplest idea is to model the active material simply as a force acting on the structure (Chaudhry and Rogers, 1991; Sun
et al., 2002a,b). The amount of recovery stress can be obtained from constrained
recovery experiments on shape memory alloy wires or from simple calculations
on basic constitutive models such as Liang and Rogers’ or Brinson’s. In the stiffness control configuration, externally constrained shape memory alloy wires are
sometimes modelled as an equivalent elastic foundation acting on the composite
(Birman, 2007, 2008; Epps and Chandra, 1997; Tsai and Chen, 2002), where
the elastic foundation stiffness is related to the tension in the shape memory
alloy wire. Both these approaches lead to simple models as the complex shape
memory alloy constitutive behaviour is only partially taken into account.
As pointed out by Zhang and Zhao (2007a), however, all these approaches
amount to neglecting the change in length of the shape memory alloy elements
and the interaction between the composite and shape memory alloy deformations. A second approach, allowing to take this interaction into account, is
to model the hybrid composite as an equivalent continuum by making use
of homogenisation techniques. The simplest homogenisation is analogous to
the derivation of equivalent single layer theories, described in Chapter 2, and
amounts to assuming perfect bonding and integrating over the beam or plate section (Ghomshei et al., 2001; Turner and Patel, 2007; Zhang and Zhao, 2007b).
More complex homogenisation techniques have also been proposed, involving
micro-macro analysis (Herzog and Jacquet, 2007; Marfia, 2005).
Homogenisation techniques are an accurate way of modelling the interaction
between the different composite constituents, however they have a number of
drawbacks if a simple and flexible model for SMAHC design is sought. First
of all, the material non linearity of shape memory alloys generates a non linear equivalent continuum, increasing the problem complexity. Furthermore,
186
Active composites with shape memory alloy wires
depending upon the type of homogenisation, a three-dimensional shape memory alloy material model might be needed. As discussed in Chapter 3, threedimensional constitutive models for shape memory alloys are still an open problem and require increased computational effort, whereas the shape memory alloy
elements usually introduced in SMAHC are generally described well considering a one-dimensional behaviour. Finally, changing SMAHC design parameters
such as the number, size or placement of the shape memory alloy reinforcement
elements requires to recalculate the homogenised constitutive laws for the composite. Thus, while homogenised SMAHC models can be useful to investigate
in detail the behaviour of the hybrid composite, a simpler model is needed for
its preliminary design.
A third modelling approach discussed in the literature is to consider the
shape memory alloy wires and the host structure as separate elements and model
their interaction. This approach has been proposed by a number of researchers
(Gao et al., 2006, 2005; Lagoudas and Tadjbakhsh, 1993; Lee and Lee, 2000;
Lee et al., 1999) and it is particularly convenient for preliminary design, as
it overcomes some of the drawbacks of homogenisation techniques. Indeed, if
the shape memory alloy and host structure are described separately, the host
structure could be modelled with a linear constitutive law if appropriate. Moreover, depending upon the modelling choices, a one-dimensional constitutive law
could be adopted for the shape memory alloy elements and changes in the design
parameters previously mentioned could be easily implemented into the model.
A finite element model for the SMAHC plate is developed here, based on
the third described approach (Daghia et al., 2007b). The plate is idealised by
geometrically non linear two dimensional thin plate elements, while one dimensional truss elements are implemented separately to describe the shape memory
alloy wires and then properly connected to the plate nodes. Geometric non
linearity is introduced in order to model the effect of embedded shape memory
alloy wires on the plate buckling response. By modelling the host structure and
the shape memory alloy wires separately, material non linearity affects only the
shape memory alloy elements. In addition, the choice of truss elements allows
to use a one dimensional shape memory alloy constitutive law. Differently from
the recent works by Gao et al. (2006, 2005), the present finite element model
is actually two dimensional. Indeed, it does not require any mesh refinement
along the plate thickness in order to model shape memory alloy wires embedded
4.1 Modelling SMAHC
187
off the plate midplane, so reducing the number of nodal degrees of freedom and
improving simplicity and computational efficiency.
The finite element model is described in the following and numerically validated by comparing its predictions to those previously published by Gao et al.
(2005). In Section 4.2, on the other hand, experimental work on the shape
control configuration is presented and the results obtained are used to further
validate the present model. The shape control configuration is chosen because
of its greater sensitivity to model parameters.
4.1.1
Finite element formulation
The variational formulation is based on the principle of virtual displacements.
The procedure for deriving the non linear equilibrium equations is here described
in compact notation, then the terms are specified for the host structure plate
element and the shape memory alloy wire element.
Let us consider a virtual displacement field δu, obtained as a variation of
the real deformed configuration u. The principle of virtual displacements states
that
Z
δeT sdΩ − δLext = 0
(∀δu) ,
(4.1)
Ω
where δe is the virtual generalised strain vector compatible with the virtual
displacements, s contains the generalised stresses and δLext is the virtual work
of external forces.
The displacement approximation is introduced as follows:
u = Uq,
(4.2)
where q are the nodal displacement parameters and U is the matrix of the shape
functions. Based on this assumption, the virtual displacements and strains can
be put in the form
δu = Uδq,
δe = Bδq,
(4.3)
where B = B (q) is the strain-displacement matrix, which depends upon the
nodal displacement parameters q due to geometric non linearity.
Substituting in Equation (4.1) yields
µZ
¶
δqT
BT sdΩ − fext = 0
(∀δq) ,
(4.4)
Ω
188
Active composites with shape memory alloy wires
being fext the external nodal forces vector. Setting the bracketed term equal to
zero yields the classical form of the nodal equilibrium equation:
R (q) = fint (q) − fext = 0.
(4.5)
The Newton-Raphson solution of the non linear equilibrium equation requires
the computation of the tangent stiffness matrix:
Z
Z
∂BT
∂R
T ∂s
=
B
dΩ +
sdΩ = KT m + KT g .
(4.6)
KT =
∂q
∂q
Ω
Ω ∂q
Here, KT m is the material stiffness, while KT g is the geometric stiffness. Introducing a general constitutive relation s = C (e), the material stiffness matrix
becomes
Z
Z
∂C ∂e
∂C
KT m =
BT
dΩ =
BT
BdΩ,
(4.7)
∂e ∂q
∂e
Ω
Ω
allowing for the presence of material non linearity.
The Newton-Raphson strategy is briefly outlined here. Let qi be the solution
at the previous load step. A load increment is applied and the residual forces
vector is approximated in Taylor series as follows:
¯
∂R ¯¯
R (q) ' R (qi ) +
(q − qi ) = 0.
(4.8)
∂q ¯qi
An iteration scheme can be established:
KkT ∆qk = −Rk ,
(4.9)
where the superscript k refers to quantities evaluated at the k-th iteration.
4.1.2
Host structure plate element
The host structure finite element is based on the CLPT. The linear continuum
model introduced in Chapter 2 is enriched by introducing the non linear Von
Kármán strain assumption. The compatibility relations become:
(s)
e = ∇(s)
x u − z∇x (∇x w) +
1
(∇x w ⊗ ∇x w) = µ + zχ = µl + µnl + zχ. (4.10)
2
The membranal generalised strains are composed of a linear part, depending
on the in-plane displacements u, and of a non linear part, depending on the
4.1 Modelling SMAHC
189
z
SMA hybrid composite plate
z
xy2
x'
xx1
plate coordinate system
wire coordinate system
Figure 4.1: SMAHC plate and reference systems
transverse displacement w. Besides the introduced geometric non linearity, the
continuum model is analogous to the one previously described and it is not
reported here.
A four node finite element is developed by introducing the following displacement field assumptions:
u = Uu qu , w = UH
w qw ,
(4.11)
where Uu collects the standard Lagrangian shape functions, while UH
w contains
Hermite polynomials, ensuring C 1 continuity to the transverse displacement w
as required by the CLPT displacement-based finite element formulation. Notice
that the vector of nodal w parameters, qw , contains the nodal displacements
wi and its derivatives, namely ∂wi /∂x1 , ∂wi /∂x2 and ∂ 2 wi /∂x1 ∂x2 , thus the
element has six degrees of freedom per node.
The discretised compatibility equations, Equation (4.10), are written in operator form as follows:
1
H
µl = Dm Uu qu , χ = Db UH
(4.12)
w qw , µnl = Dmw Dnl Uw qw ,
2
where

∂/∂x1
Dm=
0
∂/∂x2



0
−∂ 2 /∂x21
,
∂/∂x2  , Db=
−∂ 2 /∂x22
2
∂/∂x1
−2∂ /∂x1 ∂x2
(4.13)
190
Active composites with shape memory alloy wires
 ¡
¢
∂UH
w /∂x1 qw
Dmw=
0
¢
¡ H
∂Uw /∂x2 qw

·
¸
0
¢
¡ H
∂/∂x1

.
∂U /∂x2 qw , Dnl =
¢
¡ w
∂/∂x2
/∂x
q
∂UH
1
w
w
(4.14)
Variation of the generalised strains yields:
δµl
=
δχ =
δµnl
=
Dm Uu δqu ,
(4.15)
Db UH
w δqw ,
Dmw Dnl UH
w δqw ,
(4.16)
(4.17)
thus the matrix B for the host structure finite element is expressed as follows:
·
¸
Dm Uu Dmw Dnl UH
w
B=
,
(4.18)
0
Db UH
w
where the dependence on the nodal displacement parameters is contained in the
matrix Dmw .
The host structure constitutive relations are linear:
·
¸ ·
¸·
¸
N
Cm Cmb
µ − µ̄
=
(4.19)
,
M
Cmb Cb
χ − χ̄
where µ̄ and χ̄ are the strains due to thermal loads and the constitutive matrices
are defined by Equations (2.20) to (2.22).
4.1.3
Shape memory alloy wire element
The shape memory alloy wires are modelled using non linear truss elements. The
local reference system is shown in Figure 4.1. The displacement components in
the x̂ and z directions are denoted by û and w, respectively.
The uniaxial strain is defined as:
µ
¶2
dû 1 dw
ε = εl + εnl =
+
.
(4.20)
dx̂ 2 dx̂
A two-node finite element is developed by introducing the following displacement field assumptions:
û = Uû qû , w = Uŵ qŵ ,
(4.21)
4.1 Modelling SMAHC
z
w
191
z
wire nodes
1
2
w
w
1
rigid body connections
d
w*
w*
1
2
w*
1
x'
nodes on midplane
x'
Figure 4.2: Rigid body connections
where Uû and Uŵ contain Lagrange shape functions. Note that the nodal
parameters qû can be transformed to qu by a reference system rotation, and
that qŵ contain only the nodal displacements wi and not its derivatives.
The discretised compatibility equations read:
εl =
d
1
Uû qû , εnl =
dx̂
2
µ
d
Uŵ qŵ
dx̂
¶2
and their variation yields the matrix B
h
³
´
i
dU
dU q
dU
B=
.
dx̂ û
dx̂ ŵ ŵ dx̂ ŵ
(4.22)
(4.23)
The strain is related to the axial stress through shape memory alloy one
dimensional constitutive relations. To this purpose, the model proposed by
Brinson (1993) is adopted here because of its simplicity, but at this stage any
other shape memory alloy one dimensional constitutive model could be chosen,
such as the ones discussed in Chapter 3. Knowing the strain-temperature state
of the shape memory alloy element it is possible through the Brinson’s model to
evaluate the stress. The generalised stress of the one dimensional model is the
axial force, computed as the product of the stress and the wire cross-sectional
area.
Given the two dimensional nature of the model, assembly requires that all the
nodes should belong to the plate midplane. To this purpose, shape memory alloy
wires embedded off the plate midplane are connected to plate nodes through
rigid body links. The idea is shown in Figure 4.2. The wire nodes, labelled
by w and positioned at z = d, are projected onto the plate midplane (nodes
labelled by w∗ ) and rigid body connections are introduced. The displacement
192
Active composites with shape memory alloy wires
DT
SMA
wtip
DT
SMA
wmid
cooling
nd
2 heating
st
1 heating
Stiffness control
Shape control
Figure 4.3: Numerical validation: stiffness and shape control
parameters q and the corresponding generalised forces f are transferred from
the w to the w∗ nodes before assembly. If d is set to zero the w and w∗ nodes
coincide, while if the shape memory alloy is embedded off the plate midplane
the wire elements contribute to the rotational stiffness of the midplane nodes.
4.1.4
Numerical validation
The model is numerically validated by comparing it with the results previously
published by Gao et al. (2005) on a laminated composite plate strip in which
shape memory alloy ribbons are replacing part of the laminate layers. The
SMAHC was subjected to a thermal load which is uniform along the beam
length. By changing the laminate boundary conditions and the position of the
shape memory alloy ribbons with respect to the neutral axis, a stiffness and a
shape control configurations are obtained.
The results predicted by the present model are shown in Figure 4.3. The
temperature is reported in Fahrenheit degrees and the displacement in inches
to facilitate comparison with the graphs in Gao et al. (2005). It can be immediately observed that the predictions of the two models are in very good
agreement. As already mentioned earlier, however, the present model is simpler
and computationally more efficient being fully two dimensional.
4.2 Building and testing shape control SMAHC
193
Figure 4.3 (left) represents the stiffness control case. The SMAHC is clamped
at both ends and the shape memory alloy ribbons are embedded symmetrically
with respect to the plate midplane. The temperature is gradually increased
from 75 F to 240 F. The mid-span displacement is plotted versus the temperature in two different cases: the dotted line represent the composite without
shape memory alloy reinforcement, while the solid line shows the behaviour of
the SMAHC. In the first case, thermal buckling occurs because of the thermal
load and the displacement increases with temperature. When the shape memory alloy ribbon is present, however, the shape memory alloy actuation causes
the mid-span displacement to decrease, thus preventing thermal buckling and
controlling the structure’s stiffness.
Figure 4.3 (right) represents the shape control case. The SMAHC is in a
cantilevered configuration and the shape memory alloy ribbons are embedded at
a certain distance from the neutral axis. The temperature is gradually increased
from 75 F to 240 F, then gradually decreased to 70 F, then increased again up to
240 F. The plot shows the variation of the tip displacement as the temperature
changes. Upon heating above the shape memory alloy transformation temperature, the SMAHC starts to bend and the tip displacement gradually increases.
During cooling, the inverse transformation takes place and the tip displacement
decreases. A residual displacement is present at the end of the cooling branch
as the recovery stress was not completely released.
In the stiffness control configuration, the actuation temperature of the shape
memory alloy elements is the main design parameter. Once the ribbons are in
tension, the plate quickly reaches the configuration with zero mid-span displacement, a further temperature increase does not cause significant changes in the
plate configuration. On the other hand, in shape control the plate configuration
continues to change as the temperature increases. The amount of displacement
that can be achieved depends upon the geometric and material characteristics
of the SMAHC components. Thus, in shape control the correct choice of the
model parameters plays a crucial role in the prediction of the SMAHC response.
4.2
Building and testing shape control SMAHC
The present experimental investigation on shape memory alloy hybrid composites focuses on shape control applications. Indeed, the shape control configura-
194
Active composites with shape memory alloy wires
tion is highly sensitive to the material parameters of both shape memory alloy
wires and host structure, thus it requires careful design and provides a more
severe test for the model proposed in the previous section. Similarly to the numerical simulation in Figure 4.3, the investigation is centred on the controlled
bending of a cantilevered plate by actuation of prestrained shape memory alloy
wires embedded at a distance from the plate neutral axis.
The experimental work is divided in two parts. The aim of the first part is
to experimentally demonstrate the capability of embedded shape memory alloy
wires to influence structural properties. In this preliminary investigation, an
SMAHC beam is built by gluing some prestrained shape memory alloy wires
on top of a carbon fibre laminate strip. The simple proof-of-concept specimen
allows to remove all technological issues related to the hybrid composite manufacturing while still obtaining experimental data to preliminary validate the
SMAHC concept and the proposed model. In order to achieve controlled heating and cooling of the shape memory alloy wires, the SMAHC beam is tested in
an environmental chamber, recording temperature and displacement data and
comparing them to the model predictions.
In the second part of the experimental investigation, hybrid composite making is investigated. Different materials and manufacturing techniques are tested
and their advantages and drawbacks are discussed. As a result, an SMAHC plate
made of glass fibre and epoxy resin pre-preg with embedded shape memory alloy
wires is built and tested. In this case, the shape memory alloy wires are heated
via Joule heating, as it would be the case in most applications. Temperature
and displacement data are recorded and compared to the model predictions.
4.2.1
SMAHC beam
The experimental work on the SMAHC beam, discussed in this section, was
carried out at the Center for Intelligent Material Systems and Structures of the
Virginia Polytechnic Institute and State University. The aim is to demonstrate
the shape control capability of embedded shape memory alloy wires by a simple
experiment. In order to remove all technological difficulties related to the hybrid composite manufacturing, a simple proof-of-concept specimen is built by
gluing prestrained shape memory alloy wires on top of a carbon fibre laminated
composite plate strip.
4.2 Building and testing shape control SMAHC
195
Table 4.1: Axial and flexural rigidities of laminate strips
Direction 1
Direction 2
Axial rigidity
kN
981
1045
Bending rigidity
kNmm2
211
224
Material characterisation
The host structure and shape memory alloy wires are tested separately in order
to obtain their material properties, which are used later in this section for the
numerical simulations.
The host structure is a symmetric cross-ply laminate plate of thickness 1.5875
mm from McMaster-Carr1 . No data concerning the laminae properties and the
lamination scheme was available, therefore only the overall plate properties could
be obtained from laminate testing. Plate strips of 203.2 × 25.4 mm (8 × 1 in, the
same dimensions of the SMAHC beam) were cut in the two principal directions
and tensile and bending tests were carried out. Tensile rigidity, relating the axial force to the axial strain, and bending rigidity, relating the bending moment
to the curvature, were thus evaluated and they are reported in Table 4.1. The
laminate stiffness matrices required by the finite element model (see Equation
(4.19)) are reported in Table 4.2. They were obtained by calculating the tensile
and flexural rigidities per unit width and assuming a typical Poisson’s coefficient
ν = 0.23 and an in-plane shear modulus G12 = 6000 MPa. These two material coefficients are indeed not simply determined via laminate static testing,
however this assumption is not expected to significantly modify the beam-like
response of the specimen under consideration.
A spool of 0.38 mm diameter Nitinol wire was provided by Fort Wayne
Metals Research Products Corp. The material properties of the Nitinol were
determined via differential scanning calorimeter (DSC) tests and tensile tests
at different temperatures. The zero-stress phase transition temperatures determined via DSC were Ms = 0◦ C, Mf = 27◦ C, As = 40◦ C and Af = 55◦ C.
According to Abel et al. (2004), who discussed various experimental techniques
1 www.mcmaster.com
196
Active composites with shape memory alloy wires
Table 4.2: Constitutive matrices for the finite element model, units N and mm
Cm
40642 9348
 9348 43303
0
0


0
0 
9500
Cmb

all zero
3 × 3 matrix
8733
 1963
0
Cb

1963
0
9242
0 
0
2000
Table 4.3: Nitinol wires: material properties
Mf
◦
C
-10
CA
MPa/◦ C
8
CM
MPa/◦ C
8
σs
MPa
90
Ms
◦
C
17
σf
MPa
140
As
C
30
◦
Af
◦
C
45
EM
GPa
7
EA
GPa
28
Θ
MPa/◦ C
0.55
εL
0.065
to determine the transformation temperatures, the DSC test tends to yield
higher values than those observed in the applications. For this reason, in light
of the experimental results on the SMAHC beam presented further in this section, lower temperature values were used in the simulation of the composite
behaviour. The tensile tests, carried out inside a Tenney vacuum-temperature
chamber (model 36ST), were used to characterise the Brinson’s model phase
diagram (see Figure 3.11(b)). The material properties obtained are reported in
Table 4.3. Even though different elastic moduli are reported in Table 4.3 for
austenite and martensite, a mean elastic modulus was considered in the finite
element model for simplicity. Finally, the value of the material modulus Θ was
taken from the technical literature and not measured, since it does not modify
significantly the SMAHC response.
Various factors influenced the choice of the adhesive used to bond the shape
memory alloy wires to the host structure. Indeed, the adhesive should be compliant in order to modify as little as possible the stiffness properties of the original
laminate. Moreover, it should cure at room temperature in order to avoid any
4.2 Building and testing shape control SMAHC
197
Figure 4.4: SMAHC specimen
actuation of the shape memory alloy wires during the specimen building. This
way, the wires do not need to be restrained during adhesive cure. All the stated
reasons let to the choice of 3M Scotch-Weld 2216 B/A Epoxy to create the
bonding layer.
Specimen preparation
The SMAHC beam preparation requires two steps:
• the shape memory alloy wires are prestrained;
• the shape memory alloy wires are bonded to the host structure.
The Nitinol wires were prestrained by 4% using an Instron 4204 machine.
The amount of prestrain influences the amount of recovery stress that can be
achieved, thus the behaviour of the SMAHC specimen strongly depends upon
the prestrain. For this reason, it is important that all the wires are prestrained
by the same amount. In order to eliminate undesired internal stresses, the wires
were heated with a Craftsman industrial heat gun before prestraining. Two 550
mm long shape memory alloy wires were prestrained contemporarily to ensure
the same amount of deformation; each wire was cut into half to obtain four
equally prestrained wires.
The wires were laid equally spaced on the surface of the laminated composite beam. Two Omega type J thermocouples (model 5TC-TT-J-36-36) were
198
Active composites with shape memory alloy wires
thermocouples
sensor
head
vibrometer
controller
laptop with
pDaqView
data acquisition
system
USB
port
Figure 4.5: Experimental setup for the SMAHC beam
attached to two of the wires at approximately half the beam length with conductive tape. The wires were bonded to the beam using 3M Scotch-Weld 2216
B/A Epoxy adhesive. Two different layers of epoxy were applied, each layer
was left to cure 24 hours at room temperature. Figure 4.4 shows a picture of
the specimen, the thermocouples’ positions are marked with circles. The shape
memory alloy wires positioning is a critical phase of the specimen preparation:
the shape memory alloy wires must be evenly spaced and laid out symmetrically
with respect to the beam axis of symmetry, in order avoid undesired torsional
deformation during the actuation. Indeed, some imperfections are inevitably
present and a small amount of torsion was observed during the tests.
Experimental setup
The SMAHC was tested in a Tenney vacuum-temperature chamber (model
36ST). Starting from room temperature, the air in the chamber was slowly
heated to 125◦ C and then slowly cooled to 20◦ C, since a slow temperature
change allows to neglect the effects of the loading rate. The sequence was
4.2 Building and testing shape control SMAHC
199
repeated several times in order to obtain various actuation cycles. In the last
actuation, the temperature in the chamber was further increased to 137◦ C in order to investigate the effect of a change in the actuation conditions. A schematic
representation of the experimental setup is reported in Figure 4.5.
In real life applications, shape memory alloy actuators are usually heated via
resistive heating. This technique allows quick heating, achieving rapidly the final
configuration, while little control is possible on the cooling. For experimental
purposes, a relatively slow and controlled heating and cooling was required, thus
it was considered best to conduct the experiments in a temperature controlled
chamber.
The SMAHC was clamped at one end, leaving a free length of 168 mm.
Reflective tape was attached at the free end of the beam and a Polytec laser vibrometer (OFV 303 Sensor head and OFV 3001 Vibrometer controller) was used
to measure the tip displacement. Temperature readings were obtained from the
two embedded thermocouples and from another thermocouple of the same type,
attached to the wire just outside the epoxy layer. The external thermocouple
was necessary to capture the temperature gradient generated along the wires:
the wire ends, in direct contact with the air inside the chamber, heat up and
cool down more quickly than the parts embedded in the epoxy.
The thermocouples and the laser vibrometer were connected to a USB data
acquisition system by Omega (model OMB-DAQ-56). The data was recorded
on a laptop equipped with the pDaqView software.
Experimental results
The tip displacement is plotted against the external thermocouple reading for
different loading cycles. Figure 4.6 depicts the first thermal cycle. The graph
starts from point A and follows the direction of the arrows. Upon heating the
displacement increases, first slowly due to a different coefficient of thermal expansion between the wires and the laminate, then more quickly as the M → A
phase transformation starts. At 125◦ C the tip displacement is 12.4 mm. Upon
cooling, the displacement decreases again thanks to the A → M phase transformation. However, a residual tip displacement is present when the specimen
reaches room temperature.
Figure 4.7 shows again the first actuation (black line) and two representative
examples of the following cycles (starting from points B and C, respectively).
200
Active composites with shape memory alloy wires
16
Tip displacement (mm)
cycle 1
14
12
cooling
10
8
6
heating
4
2
0
0
A
20
40
60
80
100
120
140
Temperature ( °C)
Figure 4.6: SMAHC beam response - cycle 1
Tip displacement (mm)
16
cycle 1
cycle 3
cycle 5
14
12
cooling
10
8
6
heating
B
4
C
2
0
0
A
20
40
60
80
100
120
140
Temperature ( °C)
Figure 4.7: SMAHC beam response - cycles 1, 3 and 5
4.2 Building and testing shape control SMAHC
201
Tip displacement (mm)
16
cycle 1
cycle 3
cycle 5
cycle 10
14
12
cooling
10
8
6
4
heating
B
D C
2
0
0
A
20
40
60
80
100
120
140
Temperature ( °C)
Figure 4.8: SMAHC beam response - cycles 1, 3, 5 and 10
Between the different loading cycles, the environmental chamber was opened
in order to check the equipment and the specimen, therefore there is a little
temperature gap when data was not recorded. As it can be noted in the graphs,
the heating branch of the first cycle is quite different from that of the following
cycles, which are stable. In particular, the first actuation occurs at higher temperatures. In order to further investigate this peculiarity, Figure 4.8 reports the
previous plots together with the last one, in which the maximum temperature of
137◦ C was reached (light grey line, starting from point D). Two different steps
in the actuation can be observed in the last cycle. The first follows the path of
the stabilised cycles, whereas the second, starting at 125◦ C, continues along the
path marked by the first cycle.
Comparison with the model predictions
The experimental behaviour depicted by Figures 4.6 to 4.8 is rather different
from the expected response as predicted by models, such as the one discussed
in Section 4.1. Indeed, according to Figure 4.3 (right), the heating curves for
202
Active composites with shape memory alloy wires
the different cycles should overlap and no change in slope is expected during the
actuation. However, the different actuation temperatures between the first and
subsequent heatings and the change in slope observed in the last cycle could
be interpreted in light of the experimental observations on shape memory alloy
wires reported in Section 3.4.2.
Let us consider the loading history of the wires in the SMAHC beam. As the
wires are prestrained at room temperature, the reverse phase transition temperatures are expected to increase with respect to those of the virgin specimen.
During the first heating cycle, the M → A transformation occurs and part of the
martensite is transformed into austenite. Upon cooling, the A → M transformation occurs at nearly constant strain due to the composite beam constraint,
thus the newly formed martensite has lower reverse phase transformation temperatures. At this point, two different types of detwinned martensite are present
in the wires. The first one, denoted with ξ d∗ , was formed during the wires prestrain and its phase transition temperatures are denoted with A∗s > As and
A∗f > Af . The second one, denoted with ξ d , is the martensite formed during
cooling at constant strain, its phase transition temperatures are the same as
those of the virgin specimen (As and Af ). During the first thermal cycle, only
ξ d∗ is present and actuation occurs at temperature A∗s . In the following cycles,
on the other hand, ξ d is the cause of the first actuation step, starting at temperature As . If the temperature is further increased, the transformation of ξ d∗
starts, explaining the behaviour observed during the last thermal cycle.
The interpretation proposed is tested by simulating the SMAHC beam behaviour using the model discussed in Section 4.1 together with a slightly modified Brinson’s constitutive model. The modified model contains two separate
internal variables to depict ξ d and ξ d∗ , whose kinetic laws are defined in two
different phase transition strips. The increase in the reverse phase transition
temperatures associated to ξ d∗ is taken to be 30◦ C. The modification introduced in the shape memory alloy constitutive law is rather crude and proposed
here only as a way to preliminarily test this interpretation of the observed phenomenon. A modified model accounting for the shape memory behaviour for
complex loading histories as discussed in Section 3.4.2 is the subject of ongoing
research.
The geometrical and mechanical characteristics of the laminate and shape
memory alloy wires, discussed earlier in this section, were used as input data.
4.2 Building and testing shape control SMAHC
203
Tip displacement (mm)
16
experiment
simulation
14
12
10
8
6
4
2
0
0
20
40
60
80
100
120
140
Temperature ( °C)
Figure 4.9: Simulation of SMAHC beam response - cycle 1
Tip displacement (mm)
16
experiment
simulation
14
12
10
8
6
4
2
0
0
20
40
60
80
100
120
140
Temperature ( °C)
Figure 4.10: Simulation of SMAHC beam response - cycles 1, 3 and 5
204
Active composites with shape memory alloy wires
Tip displacement (mm)
16
experiment
simulation
14
12
10
8
6
4
2
0
0
20
40
60
80
100
120
140
Temperature ( °C)
Figure 4.11: Simulation of SMAHC beam response - cycles 1, 3, 5 and 10
The stiffness of the epoxy was neglected since the chosen adhesive is extremely
compliant with respect to the composite plate. A mesh of 96 × 2 thin plate
elements was considered. The shape memory alloy wires were modelled using 96
truss elements, the cross-sectional area of each truss element corresponds to the
area of all four wires. This FEM model was considered sufficient to capture the
beam-like bending behaviour of the SMAHC. Since the FEM structural model
requires the temperature of the wire as an input, the readings of thermocouples
1 and 3 were used as input data and a linear temperature gradient was assumed
between them.
The SMAHC beam response as predicted by the finite element model is
depicted in Figures 4.9 to 4.11. As it can be seen, the model is able to accurately
capture the slope of the experimental displacement-temperature graph, while
the modified constitutive law for the shape memory alloy wires allows to describe
the change in response in the different heating cycles.
4.2 Building and testing shape control SMAHC
4.2.2
205
SMAHC plate
The second part of the experimental work concerns the manufacturing of a
hybrid composite plate. The work was carried out in collaboration with the Institute for Composite and Biomedical Materials of the Italian National Research
Council (CNR). Two different manufacturing techniques, namely VARTM and
pre-preg lay-up, were investigated.
Initially, VARTM was chosen as a simple and cost-effective technique for
SMAHC plate manufacturing. Testing of a VARTM plate, however, revealed
some problems which turned out to be related to the manufacturing technique
itself. For this reason, a second attempt was made using the pre-preg layup
technique. The pre-preg SMAHC plate was tested and its behaviour was compared with the finite element model predictions. The two fabrication techniques
are documented in the following, then results concerning the pre-preg SMAHC
plate are presented and discussed.
VARTM plate
The VARTM technique consists in impregnating layers of dry fibres by injecting
low viscosity resin with the aid of vacuum (see Section 1.3). The described
process can be carried out with a variety of resins and fibres. Particularly
attractive for the SMAHC manufacturing is the possibility to choose resins
whose cure temperature is lower than the shape memory alloy wires actuation
temperature, so that the wires do not need to be restrained during composite
cure. This is the same concern which governed the choice of the adhesive used
for the SMAHC beam in the first part of the experimental investigation.
A spool of 0.5 mm diameter shape memory alloy wire was provided by Memory Metalle GmbH2 . A bicomponents polyester resin (Arotran Q6530, produced
by Ashland) with 1% mekp as the initiator was used. The resin is usually cured
in vacuum at a temperature of 60◦ C for one hour. However, an alternative cure
cycle of 50◦ C for one hour and a half was used in this case in order not to exceed the shape memory alloy wires actuation temperature. The reinforcement
is made up of unidirectional glass fibre fabric (weight 600 gr/m2 ), produced by
Chomarat. Glass fibres are chosen both for their relatively low stiffness as compared to carbon fibres and for the electrical insulation they provide. Indeed,
2 www.memory-metalle.de
206
Active composites with shape memory alloy wires
flow distributor
breether/bleeder
vacuum tube
vacuum bag
resin injection tube
tacky tape
lower mould release film glass fibres layers
prestrained SMA wires
Figure 4.12: VARTM lay-up
(a)
(b)
Figure 4.13: VARTM composite making
carbon fibres conduce electricity, thus Joule heating of the embedded shape
memory alloy wires would be difficult to achieve. The laminate is composed
of three layers with the fibres perpendicular to the shape memory alloy wires
direction. This way, the bending stiffness in the wires direction is minimum and
the maximum actuation should be achieved. Eight wires are placed between the
first and second fibres layer, thus giving the maximum possible offset from the
plate midplane.
The fibre layers and the shape memory alloy wires were placed on top of a
heated plate, which had been covered with a release agent to facilitate removal
of the composite after cure. On top of the fibre layers, a number of other
layers were placed: a release film, which facilitates composite detachment, a
4.2 Building and testing shape control SMAHC
(a) Room temperature
207
(b) 110◦ C
Figure 4.14: VARTM plate actuation
Compoflex layer used as both breather and bleeder, a flow distributor and finally
the vacuum bag. An outline of the final lay-up is shown in Figure 4.12. Figure
4.13 shows two different phases of the VARTM process: in Figure 4.13(a), the
first fibres layer and the shape memory alloy wires are placed inside the yellow
tacky tape frame and between the air outlet and resin inlet tubes; Figure 4.13(b)
shows the whole system after vacuum was pulled.
The SMAHC plate was clamped on one side and actuated using Joule heating. Figure 4.14 shows a picture of the plate at room temperature and during
actuation. As it can be seen, significant bending was observed. During cooling,
however, the displacement reached during actuation was only partially recovered, and after a few cycles the composite remained in the deformed configuration even at room temperature.
The permanent deformation in the VARTM plate was explained by considering the incomplete resin polymerisation at low temperatures. Indeed, the
estimated reticulation of the resin cured at 50◦ C for 90 minutes is about 0.77.
For most applications, this is not an issue as the partially polymerised matrix has already developed the required structural properties. In the SMAHC,
however, the uncured fraction of the resin could reach cure temperature during actuation, thus the matrix could complete its cure cycle in the deformed
configuration.
In light of the interpretation proposed, the SMAHC manufacturing process
needs to be modified. One possibility is to use a resin which completely polymerises at low temperatures, thus significantly limiting the choice of materials
for SMAHC making. Alternatively, higher cure temperatures could be applied
208
Active composites with shape memory alloy wires
provided that the shape memory alloy wires are restrained during the cure cycle. The last approach was chosen in the following. Removing the limitation on
the cure temperature allowed a broader choice of manufacturing materials and
processes, thus the highly controllable and repeatable pre-preg lay-up process
is considered in the following.
Pre-preg lay-up plate
In pre-preg lay-up, fabrics which are already impregnated with semi-solid resin
are placed on a lower mould and cured by applying pressure and temperature,
usually in autoclave or press. The process is highly controllable and repeatable,
allowing to obtain a material with better properties by controlling the amount
of resin present in the composite.
As the resin is semi-solid at room temperature, temperatures as high as
120◦ C are usually required in the cure cycle. During SMAHC cure, therefore,
the embedded shape memory alloy wires reach the actuation temperature and
need to be prevented from losing their prestrain. A steel frame was build to
restrain the shape memory alloy wires during composite making. The frame
has an embedded system allowing to prestrain the wires directly before blocking
them.
Material characterisation A unidirectional S glass fibre epoxy pre-preg was
used for the host structure manufacturing. A plate with five layers placed in
the 0◦ direction and no embedded shape memory alloy wires was made and
tested in order to obtain the host structure mechanical properties. The cure
cycle used is the same described in the next section for the SMAHC plate.
The flexural modulus in the 0◦ and 90◦ directions, Ea and Eb , were obtained
through mechanical testing, while the Poisson’s coefficient νab and in-plane shear
modulus Gab were assumed. The material properties are reported in Table 4.4.
A spool of 0.5 mm shape memory alloy wire was provided by Saes-Getters3 .
Differently from the shape memory alloy wires used in the experimental work
described up to now, these wires were subjected to training by the producers
before purchase. Training consists in cyclically loading the shape memory alloy wires multiple times in order to stabilise the material properties. Different
3 www.saesgetters.com
4.2 Building and testing shape control SMAHC
209
Table 4.4: Pre-preg glass fibres composite lamina: material properties
Ea
GPa
34.587
Eb
GPa
7.537
νab
Gab
GPa
3
0.3
Table 4.5: Trained Nitinol wires: material properties
As
C
70
◦
Af
◦
C
78
CA
MPa/◦ C
5
Θ
MPa/◦ C
0
εL
0.06
A®R
Heat flow
R®M
E
GPa
18
R,M®A
0
20
40
60
Temperature (°C)
80
100
Figure 4.15: DSC measurement on trained wires
210
Active composites with shape memory alloy wires
training cycles can be designed, involving for example superelastic loading and
unloading or heating and cooling under a constant load. In this case, the procedure used for training was not disclosed by the producers. The overall effect
of training is to stabilise the material properties of the shape memory alloys
and induce a two-way memory effect. The phase transition temperatures for
the shape memory alloy wires were obtained via DSC. The calorimetric measurement is reported in Figure 4.15. As it can be seen, austenite to R-phase
transition is present during cooling with a very small temperature hysteresis.
This feature makes the shape memory alloy wires particularly suitable for actuation by limiting the hysteretic response. The R-phase transition, however,
is not described by any of the constitutive models examined in this work. For
this reason, the behaviour of the shape memory alloy wires at cooling cannot
be described by the numerical simulation reported in the following, which includes only the heating branch. The remaining material properties of the wires
which are needed in the simulation of the heating branch were characterised via
mechanical testing and are reported in Table 4.5.
Composite design and fabrication The composite lamination scheme is
designed to have low stiffness in the direction of the shape memory alloy wires.
Indeed, the stacking sequence is (90/90/0/90/90), where the 0◦ direction coincides with the wires direction. The external layers are all placed at 90◦ , resulting
in a strong bending anisotropy, while the middle lamina with 0◦ fibres slightly
decreases in-plane anisotropy. Ten wires are placed between the first and second
layer, thus achieving the maximum possible offset from the plate midplane. The
in-plane distance between the wires is 15 mm.
The composite was laid up directly on the steel frame. A release agent was
placed on the frame in order to ensure easy detachment of the final composite
from the mold. Five layers of pre-preg were cut with dimensions 150×240 mm
and laid up on the frame. The shape memory alloy wires were placed between
the first and second pre-preg layer. In order to improve adhesion between the
shape memory alloy wires and pre-preg layers, the surface of the wires was
mechanically treated by abrading it with grit paper (Rossi et al., 2007). The
wires were then prestrained by 3.5% and allowed to recover the elastic part of
the deformation, resulting in a transformation prestrain of 2.75%. A release
film was placed on top of the pre-preg layers, finally the vacuum bag was put
4.2 Building and testing shape control SMAHC
211
(a)
(b)
(c)
(d)
Figure 4.16: Pre-preg SMAHC plate making
into place. The various step of composite making are documented in Figure
4.16. Figure 4.16(a) shows the steel frame with the first pre-preg layer, the
prestrained shape memory alloy wires and the tacky tape frame necessary to
seal the vacuum bag on top of the laminate. Figure 4.16(b) shows the complete
composite lay-up, while in Figure 4.16(c) the vacuum tube and the top release
film are put into place. Finally, in Figure 4.16(d), the vacuum bag is placed.
The composite was cured with the help of vacuum and a press with controllable plate pressure and temperature. The cure cycle is the following:
• apply vacuum at 0.7 bar for one hour;
• apply a pressure of 3 bar;
• increase the temperature to 128◦ C at a rate of 2◦ C/min;
• maintain a constant temperature of 128◦ C for 90 minutes;
• cool to room temperature at a rate of 2◦ C/min;
• release the applied pressure.
212
Active composites with shape memory alloy wires
bonding defects between SMA and composite
vacuum pump direction
Figure 4.17: Pre-preg SMAHC plate
Figure 4.17 shows a picture of the composite plate as removed from the
steel frame. The shape memory alloy wires are perfectly visible under the first
glass fibres layer. A visual inspection reveals non perfect bonding between the
shape memory alloy wires and the composite layers: the light grey areas around
the wires, which are evident in particular in the zones marked by the arrows,
denote the presence of some air between the composite layers. Note that the
defects are more significant on the side opposite to where the vacuum tube was
placed. The presence of bonding defects is mainly related to the size of the
wires as compared to the laminae thickness: indeed, 0.5 mm wires were used
and the overall laminate thickness is 1.3 mm (mean ply thickness 0.26 mm).
The bonding could be improved by using wires with smaller diameter and a less
viscous resin which may better penetrate the gaps between the shape memory
alloy wires and the composite fibres.
After cure, a 150×150 mm square containing all the shape memory alloy
wires was cut from the plate and used for testing. The remaining strips, consisting only of glass fibres composite, were saved in case a calorimetric charac-
4.2 Building and testing shape control SMAHC
acquisition
system
213
LVDT
A
thermocouple
temperature
controller
+
control thermocouple
Figure 4.18: Experimental setup for the pre-preg SMAHC plate
terisation of the composite was needed.
Experimental setup The shape memory alloy wires were actuated using
Joule heating. An outline of the experimental setup is shown in Figure 4.18.
Five couples of wires, each couple connected in parallel, were connected in series and attached to a Eurotherm 2704 temperature controller, using a K type
thermocouple as feedback. The plate was clamped on one side, leaving a free
length of 125 mm. The displacement of point A in Figure 4.18 was monitored
using an LVDT (DFg5 unguided min LVDT, ±5mm stroke). The LVDT and a
second K type thermocouple attached to one of the shape memory alloy wires
were connected to a DAQ acquisition system, displacement and temperature
data were recorded using Labview.
Experimental results Figure 4.19 shows a profile picture of the SMAHC
specimen at room temperature and at a temperature of 160◦ C, where a significant deformation is observed. As it can be seen from the picture, at room
temperature the plate is not flat but it shows an initial curvature, mainly related
214
Active composites with shape memory alloy wires
(a) Room temperature
(b) 160◦ C
Figure 4.19: Pre-preg SMAHC plate actuation
to the presence of an initial stress in the shape memory alloy wires. Indeed, during composite cure at 128◦ C the shape memory alloy wires developed a recovery
stress which was not completely released upon cooling. This is the same phenomenon occurring in the SMAHC beam after the first actuation cycle (see Figure
4.6). In the present case, however, the first actuation of the shape memory alloy
wires occurs during composite cure, thus the residual curvature is present right
at the beginning of composite testing. The initial displacement of point A with
respect to the undeformed flat plate configuration was approximately evaluated
to be 3 mm.
Figure 4.20 shows the point A displacement versus temperature graph for a
number of loading cycles, reaching different maximum temperatures. Differently
from what observed in the SMAHC beam experiment, the actuation is rather
repeatable and the hysteresis between the heating and cooling branches is nearly
absent. These differences are related partly to the type of shape memory alloy
4.2 Building and testing shape control SMAHC
215
Point A displacement (mm)
8
7
6
cooling
5
heating
4
3
2
1
0
20
35
50
65
80
95
110 125 140 155 170
Temperature ( °C)
Figure 4.20: Pre-preg SMAHC plate response for different cycles
wires used in the two experiments and partly to the different specimen manufacturing process and experimental setup. Indeed, trained wires tend to have
a more stable behaviour than untrained ones, with a smaller hysteresis related
to the presence of R-phase. Moreover, the shape memory alloy wires heating
occurring during pre-preg SMAHC manufacturing is expected to partially remove the effect of the loading history on the shape memory alloy wires, which
constituted the main cause of the different responses observed in the SMAHC
beam. Finally, the different experimental setup (Joule heating versus environmental chamber) made it more difficult to achieve controlled shape memory
alloys cooling, thus the thermocouple response time may have influenced data
acquisition, especially in the cooling branch.
The temperature control and acquisition were indeed a critical point in the
experimental design. No thermocouples were embedded in the SMAHC plate
in order to prevent the creation of interlaminar defects. For this reason, only
the temperature in the external part of the shape memory alloy wires could be
recorded, where the significant movements occurring during actuation made it
difficult to achieve perfect contact between the thermocouple and wires. Besides
216
Active composites with shape memory alloy wires
providing useful information for the evaluation of the SMAHC plate response
and the validation of the numerical model, good temperature measurements
are a crucial feedback for the control system. An idea to improve temperature
acquisition for control purposes is to use micrometric thermocouples or optical fibres sensors, which could be directly embedded in the hybrid composite
without causing significant interlaminar defects.
Comparison with the model predictions The numerical model discussed
in Section 4.1 is used to simulate the behaviour of the SMAHC plate. In Section 4.2.1, a complete simulation was carried out to predict the behaviour of
the SMAHC beam during the different loading cycles. In this case, on the
other hand, a number of difficulties arise, regarding the definition of the plate
geometry and initial conditions and the modelling of the cooling branch.
As already anticipated, the presence of R-phase does not allow to model the
shape memory alloy behaviour in cooling. Thus, the numerical simulation is
not able to describe a complete actuation cycle and only the heating branch
is reported in the following. Due to the actuation occurring during composite
cure, the initial conditions of the SMAHC are difficult to evaluate, as is the
initial stress in the shape memory alloy wires. On the other hand, a simulation
considering different initial stress conditions in the shape memory alloy wires
is still expected to capture the slope of the displacement-temperature curve
during actuation. For this reason, the simulation was carried out considering
an initially flat plate with untensioned shape memory alloy wires. Finally, the
composite manufacturing makes it difficult to accurately evaluate the offset of
the shape memory alloy wires with respect to the plate midplane, which is a
crucial information greatly influencing the predicted plate behaviour. In the
following simulation, the offset d was assumed to be d = h/2 − φ/2 − hl = 0.3
mm, where h = 1.3 mm is the plate thickness, φ = 0.5 mm is the wire diameter
and the topmost ply thickness was assumed to be hl = 0.1 mm.
The geometrical and mechanical characteristics of the composite material
and shape memory alloy wires were used as input data for the simulation. A
20×20 two-dimensional plate elements mesh was considered, with shape memory alloy truss elements connected at the nodes in the appropriate places. The
thermocouple reading was used as input data, assuming a uniform temperature along the shape memory alloy wires. As the effects of the loading history
4.2 Building and testing shape control SMAHC
217
Point A displacement (mm)
8
experiment
simulation
7
6
5
4
3
2
1
0
20
35
50
65
80
95
110 125 140 155 170
Temperature ( °C)
Figure 4.21: Simulation of pre-preg SMAHC plate response
are partially removed by the constrained recovery thermal cycle occurring during pre-preg SMAHC manufacturing, the modification to the Brinson’s model
discussed in Section 4.2.1 is not considered here.
The SMAHC plate response as predicted by the numerical model is depicted in Figure 4.21. As expected, notwithstanding the different initial conditions, the numerical simulation is able to capture the slope of the displacementtemperature curve in the actuation region.
218
Active composites with shape memory alloy wires
Conclusions and
perspectives
This dissertation discusses a number of issues related to active composites with
embedded shape memory alloys. Their different constituents, namely fibrereinforced composites and shape memory alloy elements, are first examined
separately and some original contributions are developed whose validity extends
beyond active composites applications. Then, active composites are considered
by developing and experimentally validating a simple model to describe both
shape and stiffness control configurations.
While the proposed modelling tools are simple and useful for preliminary
design and the experimental work demonstrates the feasibility of the shape
control concept, a number of problems still need to be faced in order to apply
such concept to real structures.
The critical analysis of the shape memory alloy constitutive models has
emphasised the need for further research work, related mainly to the threedimensional behaviour. A particular attention should be paid to the description
of complex loading histories, such as those involving a simultaneous change of
the stress and temperature conditions. Indeed, the experimental characterisation of Nitinol wires has shown that some traits of the shape memory alloys
material behaviour are generally not included in the current constitutive modelling, while the results on the hybrid composite beam have underlined the
importance of such traits in the composite modelling. Furthermore, features
such as the presence of the R-phase phase transition should not be neglected in
the modelling, as they might be significant when dealing with shape memory
220
Conclusions and perspectives
alloy actuators.
In order to improve the modelling of both shape memory alloys and their
interaction with the host structure, the macroscopic approach adopted for composites and shape memory alloy modelling could be replaced with multiscale
techniques, allowing to observe each phenomenon occurring in these complex
structures at the appropriate level. This approach should allow to thoroughly
describe the active composite behaviour, from the ordinary working conditions
all the way to damage and failure mechanisms.
From the technological point of view, the choice of the composite constituents, shape memory materials and manufacturing process should be optimised according to the sought application, addressing the problems which
were encountered in the present experimental work, such as the adherence of
the shape memory elements to the composite matrix.
Bibliography
Abel, E., Luo, H., Pridham, M., and Slade, A. (2004). Issues concerning the
measurement of transformation temperatures of NiTi alloys. Smart Materials
and Structures, 13:1110–1117.
ADINA (2006). ADINA 8.4 Theory and Modeling Guide - Volume 1. ADINA
R & D, Inc.
Alfano, G., Auricchio, F., Rosati, L., and Sacco, E. (2001). MITC finite elements
for laminated composite plates. International Journal for Numerical Methods
in Engineering, 50:707–738.
Antonucci, V., Faiella, G., Giordano, M., Mennella, F., and Nicolais, L. (2007).
Electrical resistivity study and characterization during NiTi phase transformations. Thermochimica Acta, 462:64–69.
Aoki, T. and Shimamoto, A. (2003). Active vibration control of epoxy matrix
composite beams with embedded shape memory alloys TiNi fibers. International Journal of Modern Physics B, 17:1744–1749.
Araki, S., Ono, H., and Saito, K. (2002). Micromechanical analysis of crack
closure mechanism for intelligent materials containing TiNi fibers. JSME
International Journal, 45(2):208–216.
Araujo, A. L., Mota Soares, C. M., Herskovits, J., and Pedersen, P. (2002).
Development of a finite element model for the identification of mechanical
and piezoelectric properties through gradient optimization and experimental
vibration data. Composite Structures, 58:307–318.
222
Bibliography
Araujo, A. L., Mota Soares, C. M., Moreira di Freitas, M. J., Pedersen, P.,
and Herskovits, J. (2000). Combined numerical-experimental model for the
identification of mechanical properties of laminated structures. Composite
Structures, 50:363–372.
Auricchio, F., Lovadina, C., and Sacco, E. (2001). Analysis of mixed finite
elements for laminated composite plates. Computer Methods in Applied Mechanics and Engineering, 190:4767–4783.
Auricchio, F., Marfia, S., and Sacco, E. (2003). Modelling of SMA materials:
training and two way memory effects. Computers and Structures, 81:2301–
2317.
Auricchio, F. and Petrini, L. (2002). Improvements and algorithmical considerations on a recent three-dimensional model describing stress-induced solid
phase transformations. International Journal for Numerical Methods in Engineering, 55:1255–1248.
Auricchio, F. and Petrini, L. (2004a). A three-dimensional model describing
stress-temperature induced solid phase transformations: solution algorithm
and boundary value problems. International Journal for Numerical Methods
in Engineering, 61:807–836.
Auricchio, F. and Petrini, L. (2004b). A three-dimensional model describing
stress-temperature induced solid phase transformations: thermomechanical
coupling and hybrid composite applications. International Journal for Numerical Methods in Engineering, 61:716–737.
Auricchio, F., Reali, A., and Stefanelli, U. (2007). A three-dimensional model
describing stress-induced solid phase transformation with permanent inelasticity. International Journal of Plasticity, 23:207–226.
Auricchio, F. and Sacco, E. (1997). A one-dimensional model for superelastic
shape-memory alloys with different elastic properties between austenite and
martensite. International Journal of Non-Linear Mechanics, 32(6):1101–1114.
Auricchio, F. and Sacco, E. (1999a). A mixed-enhanced finite-element for the
analysis of laminated composite plates. International Journal for Numerical
Methods in Engineering, 44:1481–1504.
223
Auricchio, F. and Sacco, E. (1999b). A temperature-dependent beam for shapememory alloys: constitutive modelling, finite-element implementation and
numerical simulations. Computer Methods in Applied Mechanics and Engineering, 174:171–190.
Auricchio, F. and Sacco, E. (2001). Thermo-mechanical modelling of a superelastic shape-memory wire under cyclic stretching-bending loadings. International Journal of Solids and Structures, 38:6123–6145.
Auricchio, F. and Sacco, E. (2003). Refined first-order shear deformation theory
models for composite laminates. Journal of Applied Mechanics, 70:381–390.
Auricchio, F., Sacco, E., and Vairo, G. (1999). Partial-mixed formulation and
refined models for the analysis of composite laminates within an FSDT. Composite Structures, 46:103–113.
Auricchio, F., Sacco, E., and Vairo, G. (2006). A mixed FSDT finite element
for monoclinic laminated plates. Computers and Structures, 84:624–639.
Auricchio, F. and Taylor, R. L. (1994). A shear deformable plate element with an
exact thin limit. Computer Methods in Applied Mechanics and Engineering,
118:393–412.
Ayorinde, E. O. and Yu, L. (2005). On the elastic characterization of composite
plates with vibration data. Journal of Sound and Vibration, 283:243–262.
Bartoli, I., Di Leo, A., and Viola, E. (2003). Parameter estimation of fibre reinforced composite materials using Bayesian sensitivity analysis. In Proceedings
of Composites in Construction International Conference, pages 595–600.
Bastianini, F., Corradi, M., Borri, A., and di Tommaso, A. (2005). Retrofit and
monitoring of an historical building using ”Smart” CFRP with embedded fibre
optic Brillouin sensors. Construction and Building Materials, 19(7):525–535.
Bathe, K.-J. (1996). Finite Element Procedures. Prentice-Hall.
Baz, A. and Chen, T. (1993). Performance of nitinol-reinforced drive shafts. In
Smart Structures and Intelligent Systems - SPIE Vol. 1917, pages 791–808.
224
Bibliography
Baz, A., Poh, S., Ro, J., Mutua, M., and Gilheany, J. (1992). Intelligent structural systems, chapter Active control of nitinol-reinforced composite beam.
Kluwer Academic Publishers.
Beck, J. V. and Arnold, K. J. (1977). Parameter Estimation in Engineering and
Science. John Wiley and Sons, New York.
Benedetti, A., de Miranda, S., and Ubertini, F. (2006). A posteriori error estimation based on the superconvergent Recovery by Compatibility in Patches.
International Journal for Numerical Methods in Engineering, 67:108–131.
Birman, V. (1997). Theory and comparison of the effect of composite and
shape memory alloy stiffeners on stability of composite shells and plates.
International Journal of Mechanical Science, 39(10):1139–1149.
Birman, V. (2007). Enhancement of stability of composite plates using shape
memory alloy supports. AIAA Journal, 45:2584–2588.
Birman, V. (2008). Shape memory elastic foundation and supports for passive
vibration control of composite plates. International Journal of Solids and
Structures, 45:320–335.
Bledzki, A. K., Kessler, A., Rikards, R., and Chate, A. (1999). Determination
of elastic constants of glass/epoxy unidirectional laminates by the vibration
testing of plates. Composites Science and Technology, 59(13):2015–2024.
Bo, Z. and Lagoudas, D. C. (1999a). Thermomechanical modeling of polycrystalline SMAs under cyclic loading, Part I: theoretical derivations. International Journal of Engineering Science, 37:1089–1140.
Bo, Z. and Lagoudas, D. C. (1999b). Thermomechanical modeling of polycrystalline SMAs under cyclic loading, Part III: evolution of plastic strains and
two-way shape memory effect. International Journal of Engineering Science,
37:1175–1203.
Bo, Z. and Lagoudas, D. C. (1999c). Thermomechanical modeling of polycrystalline SMAs under cyclic loading, Part IV: modeling of minor hysteresis
loops. International Journal of Engineering Science, 37:1205–1249.
225
Bo, Z., Lagoudas, D. C., and Miller, D. A. (1999). Material characterization of
SMA actuators under non-proportional thermomechanical loading. Journal
of Engineering Materials and Technology, 121(1):75–85.
Boyd, J. G. and Lagoudas, D. C. (1996a). A thermodynamical constitutive
model for shape memory materials. Part I. the monolithic shape memory
alloy. International Journal of Plasticity, 12(6):805–842.
Boyd, J. G. and Lagoudas, D. C. (1996b). A thermodynamical constitutive
model for shape memory materials. Part II. the SMA composite material.
International Journal of Plasticity, 12(7):843–873.
Brinson, L. C. (1993). One-dimensional constitutive behavior of shape memory alloys: thermomechanical derivation with non-constant material functions
and redefined martensite internal variable. Journal of Intelligent Material
Systems and Structures, 4:229–242.
Buehler, W. J., Gilfrich, J. V., and Wiley, R. C. (1963). Effect of lowtemperature phase changes on the mechanical properties of alloys near composition TiNi. Journal of Applied Physics, 34(5):1475–1477.
Buravalla, V. R. and Khandelwal, A. (2007). Differential and integrated form
consistency in 1-D phenomenological models for shape memory alloy constitutive behavior. International Journal of Solids and Structures, 44:4369–4381.
Burton, D. S., Gao, X., and Brinson, L. C. (2006). Finite element simulation of
a self-healing shape memory alloy composite. Mechanics of Materials, 38:525–
537.
Carrera, E. (2000). A priori vs. a posteriori evaluation of transverse stresses in
multilayered orthotropic plates. Composite Structures, 48:245–260.
Carrera, E. (2002). Theories and finite elements for multilayered, anisotropic,
composite plates and shells. Archives of Computational Methods in Engineering, 9:87–140.
Carrera, E. (2007). On the use of transverse shear stress homogeneous and nonhomogeneous conditions in third-order orthotropic plate theory. Composite
Structures, 77:341–352.
226
Bibliography
Carvelli, V. and Savoia, M. (1997). Assessment of plate theories for multilayered
angle-ply plates. Composite Structures, 39:197–207.
Casciati, F. and Faravelli, L. (2004). Experimental characterisation of a Cubased shape memory alloy toward its exploitation in passive control devices.
Journal de Physique IV, 115:299–306.
Casciati, F., Faravelli, L., and Petrini, L. (1998). Energy dissipation in shape
memory alloy devices. Computer-aided Civil and Infrastructure Engineering,
13:433–442.
Castellazzi, G., de Miranda, S., and Ubertini, F. (2006). A posteriori error
estimation in finite element analysis of plate structures. In Proceedings of the
7th world Congress in Computational Mechanics - WCCM VII, Los Angeles,
USA.
Cazzani, A., Garusi, E., Tralli, A., and Atluri, S. N. (2005). A four-node hybrid
assumed-strain finite element for laminated composite plates. Computers,
Materials & Continua, 2:22–38.
Cen, S., Long, Y., and Yao, Z. (2002). A new hybrid-enhanced displacementbased element for the analysis of laminated composite plates. Computers and
Structures, 80:819–833.
Chandra, R. (2001). Active shape control of composite blades using shape
memory actuation. Smart materials and structures, 10:1018–1024.
Chang, L. C. (1951). Atomic displacements and crystallographic mechanism in
diffusionless transformation of gold-cadmium single crystals containing 47.5
atomic percent cadmium. Acta Crystallographica, 4:320–324.
Chaudhry, Z. and Rogers, C. A. (1991). Reponse of composite beams to an
internal actuator force. In Proceedings 32nd AIAA/ASME/ASCE/AHS/ASC
conference, pages 186–193.
Chen, J., Wu, X., Li, G., Pang, S.-S., and Taheri, F. (2007). Analysis of an adhesively bonded single-strap joint integrated with shape memory alloy (SMA)
reinforced layers. International Journal of Solids and Structures, 44:3557–
3574.
227
Coleman, B. D. and Noll, W. (1963). The thermodynamics of elastic materials with heat conduction and viscosity. Archive for Rational Mechanics and
Analysis, 13:167–178.
Collins, J. D., Hart, G., Hasselman, T. K., and Kennedy, B. (1974). Statistical
identification of structures. AIAA Journal, 12(2):185–190.
Collins, J. D. and Thomson, W. T. (1969). The eigenvalue problem for structural
systems with statistical properties. AIAA Journal, 7(4):642–648.
Daghia, F., de Miranda, S., Ubertini, F., and Viola, E. (2007a). Estimation
of elastic constants of thick laminated plates within a Bayesian framework.
Composite Structures, 80:461–473.
Daghia, F., de Miranda, S., Ubertini, F., and Viola, E. (2008). A hybrid stress
approach for laminated composite plates within the First-order Shear Deformation Theory. International Journal of Solids and Structures, 45:1766–1787.
Daghia, F., Inman, D. J., Ubertini, F., and Viola, E. (2007b). Shape memory alloy hybrid composite plates for shape and stiffness control. Journal of
Intelligent Material Systems and Structures. In press.
Daghia, F., Viola, E., Seigler, T. M., and Inman, D. J. (2006). Constitutive
modeling of SMA in constrained recovery mode. In Proceedings of ASME International Mechanical Engineering Congress and Exposition, Chicago, USA.
Paper number IMECE206-15616.
de Miranda, S. and Ubertini, F. (2006). A simple hybrid stress element for
shear deformable plates. International Journal for Numerical Methods in
Engineering, 65:808–833.
Deng, Z., Li, Q., and Sun, H. (2006). Behavior of concrete beam with embedded
shape memory alloy wires. Engineering Structures, 28:1691–1697.
Di Paola, M. and Bilello, C. (2004). An integral equation for damage identification of Euler-Bernoulli beams under static loads. Journal of Engineering
Mechanics, 58:225–234.
DiCarlo, A., Podio-Guidugli, P., and Williams, W. O. (2001). Shells with thickness distension. International Journal of Solids and Structures, 38:1201–1225.
228
Bibliography
Dolce, M. and Nicoletti, M. (1999). Guidelines for shape memory alloy-based
seismic devices. In Proceedings of MANSIDE project final workshop, pages II
209–232.
Edwards, K. L. (1998). An overview of the technology of fibre-reinforced plastics
for design purposes. Materials and Design, 19:1–10.
Epps, J. and Chandra, R. (1997). Shape memory alloy actuation for tuning of
composite beams. Smart Materials and Structures, 20:251–256.
Epps, J. J. and Chopra, I. (1997).
Comparative evaluation of
shape memory alloy constitutive models with test data.
In 38th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference and Adaptive Structures Forum, Kissimmee, USA.
Fabrizio, M., Grandi, D., and Daghia, F. (2008). A non isothermal GinzburgLandau model for martensitic phase transitions in shape memory alloys. In
preparation.
Fox, R. L. and Kapoor, M. P. (1968). Rates of change of eigenvalues and
eigenvectors. AIAA Journal, 6(12):2426–2429.
Fragnito, M. and Vetrella, S. (2002). A space release/deployment system actuated by shape memory wires. Acta Astronautica, 51(11):761–770.
Frederiksen, P. S. (1997). Experimental procedure and results for the identification of elastic constants of thick orthotropic plates. Journal of Composite
Materials, 31(4):360–382.
Frederiksen, P. S. (1998). Parameter uncertainty and design of optimal experiments for the estimation of elastic constants. International Journal of Solids
and Structures, 35(12):1241–1268.
Gao, X., Burton, D., Turner, T. L., and Brinson, L. C. (2006). Finite element analysis of adaptive-stiffening and shape-control SMA hybrid composites. Journal of Engineering Materials and Technology, 128:285–293.
Gao, X., Turner, T. L., Burton, D., and Brinson, L. C. (2005). Finite element
analysis of adaptive-stiffening and shape-control SMA hybrid composites. In
229
Armstrong, W. D., editor, Smart Structures and Materials 2005: Active Materials: Behavior and Mechanics - SPIE Vol. 5761, pages 406–416. SPIE.
Ghomshei, M., Khajepour, A., Tabandeh, N., and Behdinan, K. (2001). Finite
element modeling of shape memory alloy composite actuators: theory and
experiment. Journal of Intelligent Material Systems and Structures, 12:761–
773.
Ghugal, Y. M. and Shimpi, R. P. (2002). A review of refined shear deformation
theories of isotropic and anisotropic laminated plates. Journal of Reinforced
Plastics and Composites, 21(9):775–813.
Hasselman, T. K. and Hart, G. C. (1972). Modal analysis of random structural
systems. Journal of the Engineering Mechanics Division - ASCE, 98:561–579.
He, Z., Gall, K. R., and Brinson, L. C. (2006). Use of electrical resistance testing
to redefine the transformation kinetics and phase diagram for shape-memory
alloys. Metallurgical and Materials Transactions A, 37A:579–587.
Herzog, H. and Jacquet, E. (2007). From a shape memory alloys model implementation to a composite behavior. Computational Materials Science, 39:365–
375.
Hongxing, H., Sol, H., and de Wilde, W. P. (2000). Identification of plate
rigidities of a circular plate with cylindrical orthotropy using vibration data.
Computers and Structures, 77:83–89.
Huang, N. N. (1994). Influence of shear correction factors in the higher order
shear deformation laminated shell theory. International Journal of Solids and
Structures, 31(9):1263–1277.
Hwang, S. F. and Chang, C. S. (2000). Determination of elastic constants of
materials via vibration testing. Composite Structures, 49:183–190.
ISTECH (2000). Proceedings of ISTECH project final workshop.
Janke, L., Czaderski, C., Motavalli, M., and Ruth, J. (2005). Applications of
shape memory alloys in civil engineering structures — overview, limits and
new ideas. Materials and Structures, 38:578–592.
230
Bibliography
Jenkins, P. P. and Landis, G. A. (1995). A rotating arm using shape-memory
alloy. In NASA. Johnson Space Center, the 29th Aerospace Mechanisms Symposium, pages 167–171.
Khalili, S. M. R., Shokuhfar, A., and Ashenai Ghasemi, F. (2007a). Effect of
smart stiffening procedure on low-velocity impact response of smart structures. Journal of Material Processing Technology, 190:142–152.
Khalili, S. M. R., Shokuhfar, A., Malekzadeh, K., and Ashenai Ghasemi, F.
(2007b). Low-velocity impact of active thin-walled hybrid composite structures embedded with SMA wires. Thin-Walled Structures, 45:799–808.
Kim, J. S. and Cho, M. (2007). Enhanced first-order theory based on mixed
formulation and transverse normal effect. International Journal of Solids and
Structures, 44:1256–1276.
Krevet, B. and Kohl, M. (2003). 3D finite element simulation of a shape memory
alloy microgripper. In Smart Structures and Materials 2003: Active Materials,
Behavior and Mechanics. Proceedings of SPIE Vol. 5053, pages 119–129.
Ladevèze, P., Lubineau, G., and Marsal, D. (2006). Towards a bridge between
the micro- and mesomechanics of delamination for laminated composites.
Composites Science and Technology, 66:698–712.
Lagoudas, D. C. and Bo, Z. (1999). Thermomechanical modeling of polycrystalline SMAs under cyclic loading, Part II: material characterization and experimental results for a stable transformation cycle. International Journal of
Engineering Science, 37:1141–1173.
Lagoudas, D. C., Bo, Z., and Qidway, M. A. (1996). A unified thermodynamic
constitutive model for SMA and finite element analysis of active metal matrix
composites. Mechanics of Advanced Materials and Structures, 3(2):153–179.
Lagoudas, D. C., Entchev, P. B., Popov, P., Patoor, E., Brinson, L. C., and Gao,
X. (2006). Shape memory alloys, Part II: Modeling of polycrystals. Mechanics
of Materials, 38:430–462.
Lagoudas, D. C. and Tadjbakhsh, I. G. (1993). Deformation of active flexible rods with embedded line actuators. In Recent developments in stability,
vibration, and control of structural systems, AMD - Vol. 167, pages 89–106.
231
Lai, T. C. and Ip, K. H. (1996). Parameter estimation of orthotropic plates by
Bayesian sensitivity analysis. Composite Structures, 34:29–42.
Laitinen, M., Lahtinen, H., and Sjölind, S. G. (1995). Transverse shear correction factors for laminates in cylindrical bending. Communications in Numerical Methods in Engineering, 11:41–47.
Lauwagie, T., Sol, H., Roebben, G., Heylen, W., Shi, Y., and Van der Biest, O.
(2003). Mixed numerical-experimental identification of elastic properties of
orthotropic metal plates. NDT&E International, 36:487–495.
Lee, H. J. and Lee, J. J. (2000). A numerical analysis of the buckling and
postbuckling behavior of laminated composite shells with embedded shape
memory alloy wire actuators. Smart Materials and Structures, 9:780–787.
Lee, H. J., Lee, J. J., and Huh, J. S. (1999). A simulation study on the thermal buckling behavior of laminated composite shells with embedded shape
memory alloy (SMA) wires. Composite Structures, 47:463–469.
Lemaitre, J. and Chaboche, J.-L. (1994). Mechanics of Solid Materials. Cambridge University Press.
Lembo, M. and Podio-Guidugli, P. (2007). How to use reactive stresses to
improve plate-theory approximations of the stress field in a linearly elastic
plate-like body. International Journal of Solids and Structures, 44(5):1337–
1369.
Lexcellent, C. and Schlömerkemper, A. (2007). Comparison of several models for
the determination of the phase transformation yield surface in shape-memory
alloys with experimental data. Acta Materialia, 55:2995–3006.
Li, H., Liu, Z.-Q., and Ou, J.-P. (2007). Experimental study of a simple reinforced concrete beam temporarily strengthened by SMA wires followed by
permanent strengthening with CFRP plates. Engineering Structures. In press.
Li, Y., Cui, L. S., Xu, X. B., and Yang, D. Z. (2003). Constrained phasetransformation of a TiNi shape-memory alloy. Metallurgical and Materials
Transactions A, 34A:219–223.
232
Bibliography
Liang, C. and Rogers, C. A. (1990). One-dimensional thermomechanical constitutive relations for shape memory materials. Journal of Intelligent Material
Systems and Structures, 1(2):207–234.
Lind, R. J. and Doumanidis, C. C. (2003). Active deformable sheets: prototype
implementation, modeling, and control. Optical engineering, 42(2):304–316.
Liu, Y. and Favier, D. (2000). Stabilisation of martensite due to shear deformation via variant reorientation in polycrystalline NiTi. Acta materialia,
48:3489–3499.
Lòpez, G. A., Barrado, M., Bocanegra, E. H., San Juan, J. M., and Nò, M. L.
(2007). Influence of the matrix and of thermal treatment on the martensitic
transformation in metal matrix composites. Materials Science and Engineering. In press.
Lubliner, J. and Auricchio, F. (1996). Generalized plasticity and shape-memory
alloys. International Journal of Solids and Structures, 33(7):991–1003.
Malvern, L. E. (1969). Introduction to the Mechanics of a Continuous Medium.
Prentice-Hall.
MANSIDE (1999). Proceedings of MANSIDE project final workshop.
Marfia, S. (2005). Micro-macro analysis of shape memory alloy composites.
International Journal of Solids and Structures, 42:3677–3699.
Mazumdar, S. K. (2002). Composites manufacturing : materials, product, and
process engineering. CRC Press.
Meo, M., Antonucci, E., Duclaux, P., and Giordano, M. (2005). Finite element
simulation of low velocity impact on shape memory alloy composite plates.
Composite Structures, 71:337–342.
Miller, D. A. and Lagoudas, D. C. (2000). Thermomechanical characterization of
NiTiCu and NiTi sma actuators: influence of plastic strains. Smart Materials
and Structures, 9:640–652.
Mohite, P. M. and Upadhyay, C. S. (2002). Local quality of smoothening based
a-posteriori error estimators for laminated plates under transverse loading.
Computers and Structures, 80:1477–1488.
233
Motahari, S. A. and Ghassemieh, M. (2007). Multilinear one-dimensional shape
memory material model for use in structural engineering applications. Engineering Structures, 29:904–913.
Noor, A. K., Burton, W. S., and Peters, J. M. (1990). Predictor-corrector
procedures for stress and free vibration analysis of multilayered composite
plates and shells. Computer Methods in Applied Mechanics and Engineering,
82:341–363.
Noor, A. K., Kim, Y. H., and Peters, J. M. (1994). Transverse shear stresses and
their sensitivity coefficients in multilayered composite panels. AIAA Journal,
32:1259–1269.
Ölander, A. (1932). An electrochemical investigation of solid cadmium-gold
alloys. Journal of the American Chemical Society, 54:3819–3833.
Ostachowicz, W., Krawczuk, M., and Żak, A. J. (2000). Dynamics and buckling of a multilayer composite plate with embedded SMA wires. Composite
Structures, 48:163–167.
Otsuka, K. and Wayman, C. (1999). Shape Memory Materials. Cambridge
University Press.
Park, J. W. and Kim, Y. J. (2003). Local recovery of through-the-thickness
stresses in laminated composite plates. Composite Structures, 59:291–296.
Pascale, G., Giammarruto, A., and Daghia, F. (2008). In preparation.
Patoor, E., Lagoudas, D. C., Entchev, P. B., Brinson, L. C., and Gao, X.
(2006). Shape memory alloys, Part I: General properties and modeling of
single crystals. Mechanics of Materials, 38:391–429.
Peultier, B., Ben Zineb, T., and Patoor, E. (2007). A simplified micromechanical
constitutive law adapted to the design of shape memory applications by finite
element methods. Materials Science and Engineering A. In press.
Pollard, E. L. and Jenkins, C. H. M. (2005).
Shape memory alloy
deployment of membrane mirrors for spaceborn telescopes.
In 46th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics & Materials Conference.
234
Bibliography
Popov, P. (2005). Constitutive modelling of shape memory alloys and upscaling
of deformable porous media. PhD thesis, Texas A&M University.
Popov, P. and Lagoudas, D. C. (2007). A 3-D constitutive model for
shape memory alloys incorporating pseudoelasticity and detwinning of selfaccommodated martensite. International Journal of Plasticity, 23(10–
11):1679–1720.
Prahlad, H. and Chopra, I. (2001). Comparative evaluation of shape memory alloy constitutive models with experimental data. Journal of Intelligent
Material Systems and Structures, 12:383–395.
Price, A. D., Jnifere, A., and Naguib, H. E. (2007). Design and control of a shape
memory alloy based dexterous robot hand. Smart Materials and Structures,
16:1401–1414.
Qidway, M. A. and Lagoudas, D. C. (2000). On thermomechanics and transformation surfaces of polycrystalline NiTi shape memory alloy material. International Journal of Plasticity, 16:1309–1343.
Raniecki, B. and Lexcellent, C. (1994). RL -models of pseudoelasticity and their
specification for some shape memory solids. European Journal of Mechanics
A/Solids, 13(1):21–50.
Reddy, J. N. (1984). A simple higher-order theory for laminated composite
plates. Journal of Applied Mechanics, 51:745–752.
Reddy, J. N. (1997). Mechanics of Laminated Composite Plates - Theory and
Analysis. CRC Press, Boca Raton.
Rikards, R., Chate, A., and Gailis, G. (2001). Identification of elastic properties
of laminates based on experimental design. International Journal of Solids
and Structures, 38:5097–5115.
Ro, J. and Baz, A. (1995a). Nitinol-reinforced plates: part I. Thermal characteristics. Composites Engineering, 5(1):61–75.
Ro, J. and Baz, A. (1995b). Nitinol-reinforced plates: part II. Static and buckling characteristics. Composites Engineering, 5(1):77–90.
235
Ro, J. and Baz, A. (1995c). Nitinol-reinforced plates: part III. Dynamic characteristics. Composites Engineering, 5(1):91–106.
Rolfes, E. and Rohwer, K. (1997). Improved transverse shear stresses in composite finite elements based on first order shear deformation theory. International
Journal for Numerical Methods in Engineering, 40:51–60.
Rossi, S., Deflorian, F., Pegoretti, A., D’Orazio, D., and Gialanella, S. (2007).
Chemical and mechanical treatments to improve the surface properties of
shape memory NiTi wires. Surface & Coatings Technology. In press.
Sol, H., Hua, H., De Visscher, J., Vantomme, J., and de Wilde, W. P. (1997). A
mixed numerical/experimental technique for the nondestructive identification
of the stiffness properties of fibre reinforced composite materials. NDT&E
International, 30(2):85–91.
Soutis, C. (2005). Fibre reinforced composites in aircraft construction. Progress
in Aerospace Sciences, 41:143–151.
Souza, A. C., Mamiya, E. N., and Zouain, N. (1998). Three-dimensional model
for solids undergoing stress-induced phase transformations. European Journal
of Mechanics A/Solids, 17:789–806.
Stratford, T., Pascale, G., Manfroni, O., and Bonfiglioli, B. (2004). Shear
strengthening masonry panels with sheet glass-fiber reinforced polymer. Journal of Composites for Construction, 8(5):434–443.
Sun, S., Sun, G., Han, F., and Wu, J. (2002a). Thermoviscoelastic analysis
for a polymeric composite plate with embedded shape memory alloy wires.
Composite Structures, 58:295–302.
Sun, S., Sun, G., and Wu, J. (2002b). Thermo-viscoelastic bending analysis of
a shape memory alloy hybrid epoxy beam. Smart Materials and Structures,
11:970–975.
Tanaka, K. (1986). A thermomechanical sketch of shape memory effect: one
dimensional tensile behavior. Res Mechanica, 18:251–263.
Teresi, L. and Tiero, A. (1997). On variational approaches to plate models.
Meccanica, 32:143–156.
236
Bibliography
Tsai, X.-Y. and Chen, L.-W. (2002). Dynamic stability of a shape memory alloy
wire reinforced composite beam. Composite Structures, 56:235–241.
Turner, T. L., Buehrle, R. D., Cano, R. J., and Fleming, G. A. (2006). Modeling,
fabrication, and testing of a sma hybrid composite jet engine chevron concept.
Journal of Intelligent Material Systems and Structures, 17:483–497.
Turner, T. L. and Patel, H. D. (2007). Analysis of SMA hybrid composite
structures in MSC.Nastran and ABAQUS. Journal of Intelligent Material
Systems and Structures, 18:435–447.
Ubertini, F. (2004). Patch recovery based on complementary energy. International Journal for Numerical Methods in Engineering, 59:1501–1538.
Vallet, M.-G., Manole, C.-M., Dompierre, J., Dufour, S., and Guibault, F.
(2007). Numerical comparison of some hessian recovery techniques. International Journal for Numerical Methods in Engineering, 72:987–1007.
Van Humbeeck, J. (2001). Shape memory alloys: a material and a technology.
Advanced Engineering Materials, 3(11):837–850.
Viola, E., Kay, J. P., and Bartoli, I. (2003). Dynamic identification of elastic
constants using finite element method and Ritz approach. In 16th AIMETA
Congress of Theoretical and Applied Mechanics, Ferrara, Italy.
Wada, K. and Liu, Y. (2005). Shape recovery of NiTi shape memory alloy under
various pre-strain and constraint conditions. Smart Materials and Structures,
14:S273–S286.
Wang, X. (2002). Shape memory alloy volume fraction of pre-stretched shape
memory alloy wire-reinforced composites for structural damage repair. International Journal of Solids and Structures, 11:590–595.
Wei, Z. G., Sandström, R., and Miyazaki, S. (1998). Shape memory materials and hybrid composites for smart systems: Part II shape-memory hybrid
composites. Journal of Materials Science, 33:3763–3783.
Yuan, K. Y., Huang, Y. S., and Pian, T. H. H. (1993). New strategy for assumed
stresses for 4-node hybrid stress membrane element. International Journal for
Numerical Methods in Engineering, 36:1747–1763.
237
Żak, A. J. and Cartmell, M. P. (2002). Dynamics of a rotor system with a smart
SMA-embedded sleeve-ring component. Technical Report 4, Department of
Mechanical Engineering, University of Glasgow.
Żak, A. J., Cartmell, M. P., Ostachowicz, W. M., and Wiercigroch, M. (2003).
One-dimensional shape memory alloy models for use with reinforced composite structures. Smart Materials and Structures, 12:338–346.
Zhang, R.-X., Ni, Q.-Q., Masuda, A., Yamamura, T., and Iwamoto, M. (2006).
Vibration characteristics of laminated composite plates with embedded shape
memory alloys. Composite Structures, 74:389–398.
Zhang, R.-X., Ni, Q.-Q., Natsuki, T., and Iwamoto, M. (2007). Mechanical
properties of composite filler with SMA particles and short fibers. Composite
Structures, 79:90–96.
Zhang, Y. and Zhao, Y.-P. (2007a). A discussion on modeling shape memory
alloy embedded in a composite laminate as axial force and elastic foundation.
Materials and Design, 28:1016–1020.
Zhang, Y. and Zhao, Y.-P. (2007b). A study of composite beam with shape
memory alloy arbitrarily embedded under thermal and mechanical loadings.
Materials and Design, 28:1096–1115.
Zhuiykov, S. (2008). Novel sensor-actuator device for early detection of fire.
Sensors and Actuators A, 141:89–96.
238
Bibliography
Acknowledgements
I would like to thank Prof. Erasmo Viola for giving me the opportunity of attending the Ph.D. program in Structural Mechanics and for the encouragement
and guidance given me throughout these years. I am extremely grateful to Prof.
Francesco Ubertini, who taught me many important lessons, both academic and
personal.
This work owes to the direct and indirect contribution of my colleagues at
DISTART and LAMC, who provided me with advice, support, and most of all
friendship. The cover and figures artwork would not have been the same without
the help (and patience) of Giovanni Castellazzi.
I would like to thank the research groups at CIMSS, Virginia Tech, USA,
and at the CNR Institute for Composite and Biomedical Materials, Portici, for
their help in carrying out the experimental part of this work and for allowing
me to be one of them for a while. I am particularly grateful to Prof. Daniel J.
Inman, who was a wonderful host and guidance during my stay in the USA.
I would like to thank Dr. Elena Trentini, Dr. Antonino Coglitore and Dr.
Martino Labanti from ENEA, Faenza, for their help with last minute experimental characterisation. I am grateful to Prof. Ferdinando Auricchio and Prof. Elio
Sacco for the valuable discussions on shape memory alloy constitutive models.
I dedicate this work to my parents, Silvia and Claudio, my sister Beatrice
and my boyfriend Ludovico, who were always with me from near and from afar
and who supported all my crazy ideas.
Finally, I am grateful to my two cats, Alpha and Omega, who are the best
way to chill out after a stressful day’s work, and to all my friends from Impro’,
who allowed me to escape from reality once a week and who taught me that
improvisation is crucial, in theatre and in real life.
Curriculum Vitæ
Federica Daghia was born in Imola, Italy, on the 12th November 1980. In
1997 she won a two-years scholarship to attend the United World College of
the Adriatic in Duino, Italy, and she received the International Baccalaureate
(bilingual Italian/English diploma) from the same institution in June 1999. In
July 2004, she received the Laurea degree in Civil Engineering (Structures),
100/100 cum laude, from the University of Bologna. From January 2005 to
December 2007 she attended the Ph.D. program in Structural Mechanics at the
University of Bologna on a full scholarship.
Within the Ph.D. program, she spent eight months as a Visiting Scholar
at the Center for Intelligent Material Systems and Structures of the Virginia
Polytechnic Institute and State University, USA. Furthermore, she actively collaborated with the CNR Institute for Composite and Biomedical Materials in
Portici, Italy, within the context of her doctoral thesis.
Her research interests are in the fields of Computational mechanics and Multiphase materials and structures.
Selected publications
Daghia, F., de Miranda, S., Ubertini, F., and Viola, E. (2008). A hybrid
stress approach for laminated composite plates within the First-order
Shear Deformation Theory. International Journal of Solids and Structures, 45:1766-1787.
Daghia, F., de Miranda, S., Ubertini, F., and Viola, E. (2007). Estimation of
Elastic Constants of Thick Laminated Plates within a Bayesian Framework. Composite Structures, 80:461-473.
Daghia, F., Inman, D. J., Ubertini, F., and Viola, E. (2007). Shape memory
alloy hybrid composite plates for shape and stiffness control. Journal of
Intelligent Material Systems and Structures. In press.
Daghia, F., Inman, D. J., Ubertini, F., and Viola, E. (2007). Experimental
testing of a shape memory alloy hybrid composite plate for active shape
control. In: III ECCOMAS thematic conference on smart structures and
materials, Gdansk, Poland.
Daghia, F., Viola, E., Seigler, T. M., and Inman, D. J. (2006). Constitutive
modeling of SMA in constrained recovery mode. In: Proceedings of ASME
International Mechanical Engineering Congress and Exposition, Chicago,
USA. Paper number IMECE2006-15616.
Contact
LAMC – Laboratory of Computational Mechanics
DISTART Department
Alma Mater Studiorum • Università di Bologna
viale Risorgimento 2
40136 Bologna (Italy)
Tel. +39 051 2093373
Fax. +39 051 2093496
Email. [email protected]
This thesis is available in pdf format on the LAMC website:
www.lamc.ing.unibo.it
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement