Methods for Reducing the Complexity of Geometrical Structures Based on CFD Programming Time Efficient Simulations Based on Volume Forces Coupled with Single and Two-phase Flow Kamal Rezk Faculty of Health, Science and Technology Environmental and Energy Systems DISSERTATION | Karlstad University Studies | 2014:32 Methods for Reducing the Complexity of Geometrical Structures Based on CFD Programming Time Efficient Simulations Based on Volume Forces Coupled with Single and Two-phase Flow Kamal Rezk DISSERTATION | Karlstad University Studies | 2014:32 Methods for Reducing the Complexity of Geometrical Structures Based on CFD Programming - Time Efficient Simulations Based on Volume Forces Coupled with Single and Two-phase Flow Kamal Rezk DISSERTATION urn:nbn:se:kau:diva-31983 Karlstad University Studies | 2014:32 ISSN 1403-8099 ISBN 978-91-7063-565-6 © The author Distribution: Karlstad University Faculty of Health, Science and Technology Department of Engineering and Chemical Sciences SE-651 88 Karlstad, Sweden +46 54 700 10 00 Print: Universitetstryckeriet, Karlstad 2014 WWW.KAU.SE Abstract Throughout recent years, computer based programs have been applied to solve and analyse industrial problems encountered in global fields such as automobile design for reduction of CO2-gas and designing wind parks aimed at increasing power output. One of these developed programs is Computational Fluid Dynamics (CFD) which numerically solves complex flow behaviour based on computer power. As there is an ongoing expansion of CFD usage in industry, certain issues need to be addressed as they are becoming more frequently encountered. The general demand for simulations of larger control volumes and more advanced flow processes result in an extensive requirement of computer resources. Moreover, the implementation of commercial CFD codes in small-scaled industrial companies seems to generally be utilised as a black box based on the knowledge of fluid mechanic theory. Increased partnerships between industry and the academic world involving various CFD based design processes generally yield to a verbal communication interface, which is a crucial step in the process given the level of dependency between both sides. Based on these notions, a method for establishing time efficient CFD-models with implementation of volume forces as sink terms in the momentum equation is presented. The internal structure, or parts of the structure, in the simulation domain is removed which reduces the geometrical complexity and along with it, computational demand. These models are the basis of assessing the benefits of utilizing a numerical based design process in industry in which the CFD code is used as a communication tool for knowledge sharing with counterparts in different fields. 1 Summary in Swedish Genom åren har datorbaserade program applicerats för att lösa och analysera industriella problem som exempelvis design av bilar för att minska CO2-utsläpp samt utformning av vindkraftsparker som syftar till att öka producerad effekt osv. Ett av dessa program heter Computational Fluid Dynamics (CFD) som numeriskt löser komplexa flödesprocesser baserat på datorkraft. I och med den konstanta utbredningen av CFD i industrin, måste flera punkter adresseras av den orsaken att dem förekommer ofta. Generella krav för simulering av stora kontrollvolymer med avancerade strömningsförhållanden kräver omfattande datorresurser. En till viktig punkt är att användningen av CFD i småskaliga industrier har en tendens att användas som en ”black box” modell utifrån vetenskapen av strömningsmekanik. Ett ökat samarbete mellan företag och universitet i CFD-baserade designprojekt leder till problematiken runt kommunikationen mellan olika parter med olika vetenskapliga profiler. Utifrån dessa iakttagelser presenteras en metod för att fastställa tidseffektiva CFD-modeller genom implementering av volymkrafter som källtermer i rörelsemängdsekvationen. Den inre strukturen, eller delar av strukturen i simuleringsdomänen tas bort som minskar den geometriska komplexiteteten och i sin tur, behov av datorprestanda. Dessa modeller är grunden för att bedöma fördelarna med att använda en numeriskbaserad designprocess inom industrin där CFD koden används som ett kommunikationsverktyg för kunskapsdelning mellan professioner i olika områden. 2 Summary of Papers In paper I, a heuristically determined design process of the geometry near the front trap door of an internal duct system of a heat pump tumble dryer was achieved by implementing the CFD code COMSOL MultiPhysics as a communication tool. The design process was established by two counterparts in the project, in which CFD calculations and geometry modifications were conducted separately. Two design criteria presenting the pressure drop in the duct and the outflow uniformity were used to assess geometry modifications conducted by a mechanical engineer. The geometry modifications were based on visual results of the flow patterns. The modifications confirmed an improvement in the geometry as the pressure drop was reduced by 23%, and the uniformity was increased by 3%. In paper II, single-phase simulations were conducted on a tube-fin heat exchanger in order to simulate airflow. The heat exchanger is regarded as a porous medium in which the arrangement of fins and tube bundles are replaced with volume forces such as sink terms in the momentum equation. Hence, the computational time was reduced significantly for the structureless model. The focus was on achieving a correct volume flow rate and pressure drop relation. Moreover, experimental results of the flow rate and pressure drop relation showed good agreement with the volume force model. In paper III, a numerical model of a vacuum dewatering process was established by using a two-phase flow model. In this study, a 2-dimensional model of the paper sheet was analysed numerically using a Level-Set method, coupled with the continuity and momentum equations. The purpose of analysing vacuum dewatering numerically is to gain a physical understanding of the process and eventually to obtain better methods of dimensioning industrial equipment. The aim was to calculate the dry content and the flow rates of water and air as a function of dwell time and vacuum levels. Simulations of vacuum dewatering of paper with a basis weight of 50 g/m2 showed somewhat satisfactory results compared to experimental data, in terms of relating the dry content of the paper to the dwell time. In paper IV, the numerical model presented in paper III was further developed by considering 3-dimensional flow resistance in a 2-dimensional two-phase flow model. A MATLAB algorithm linked to the CFD program COMSOL was 3 created in order to execute numerous steps in the simulation process in a selfgoing manner. The algorithm generated fibers with randomly distributed coordinate positions and orientations in a sample domain representing a paper sheet. Single-phase simulations of air and water were conducted for various Renumbers. A numerical data set was established and used to estimate flow resistance as volume forces for the sample domains representing the paper sheet and the forming wire. These volume forces were implemented in a 2dimensional two-phase flow model, in which the dry content and dewatering rate were related to the dwell time. The porosity variation in the paper sheet and the influence of the fibers packing on the forming wire were tested numerically. The numerical simulations showed that the packing evolution of the fibers on the forming wire presented the most pleasing results compared to the experimental data. In paper V, a MATLAB algorithm linked to the CFD program COMSOL was created, in order to produce sample domains with various arrangements of cylindrical objects representing fibrous materials. The flow resistance in 3dimensional fibrous structures was investigated in a wide range of Re-numbers, in which inertia was included. Resistance coefficients were established based on steady state simulations of single-phase processes of water numerically. The numerical data set were used to estimate Forchheimer coefficients, which are used to correlate a dimensionless friction factor to a modified Reynolds expression for porous media. The friction factor and dimensionless permeability were calculated for fibrous arrangements in a 3-dimensional space. The numerical simulations agreed well to classical empirical formulations, regardless of their fiber orientation and porosity. The estimation of the friction factor is, however, sensitive depending on the range of Re-numbers over which the resistance coefficients are estimated over. The viscous resistance was estimated with better accuracy, as the dimensionless permeability was in good agreement with the current simulation data from other literature. 4 Preface I received my Master of Science degree in Environmental and Energy Systems at Karlstad University in 2007. I began my PhD studies at Karlstad University in the autumn of 2008. This thesis originates from the cooperation between Karlstad University and the household appliance company Asko Appliance AB, which also financed the project that resulted in a licentiate degree. The aim of the cooperation was to establish a good relation between the industry and the university. By doing so, several projects have been created with the purpose of sharing knowledge and creating a foundation for applied academic research in the department. I acquired my licentiate in February 2011, with the thesis labelled as “CFD as a tool for analysis of complex geometry”. The thesis addressed the beneficial aspects of implementing CFD analysis in a design cycle consisting of counterparts with different specialist fields. The focus was to point out the CFD code as a communication tool and a foundation for knowledge sharing. Also, the thesis introduced the concept of implementing volume forces as sink terms in the momentum transport equation in order to reduce computational demand. The concept was introduced by predicting the flow rate and pressure drop relation of a fin-tube heat exchanger. My PhD research continued at Karlstad University through exploration of the volume force concept by characterising flow resistance in a vacuum dewatering process of paper sheets. The method was further explored by comparing the flow resistance of the volume force terms to classically empirical formulations, other numerical simulations as well as experimental data. Parts of the text in my licentiate thesis are reused in this work which consists of background facts, theory and historical inputs in the subject of fluid mechanics. Karlstad, May 2014 5 Acknowledgement I would like to start by thanking Asko Appliance AB for financing the projects concluding paper I and II. Thanks to Peder Bengtsson and Anders Sahlén for having faith in me. Thanks to Mechanical Design Engineer Johan Olsson from Asko Appliance AB for conducting geometry modifications in paper I. I would like express my gratitude Assoc Prof Ola Holby for supervising me during the first half of this project which concluded paper I and II. I want to thank Lars Pettersson, Laboratory Assistant, for assembling laboratory setups for paper II. I want to thank my supervisor Jonas Berghel who has been a big support for from the start of my studies. I am very grateful and thank you for your valuable comments. I want to thank my supervisor Lars Nilsson who has been with me since my licentiate. I have learnt a great deal from you and your enthusiasm has been a great inspiration for me. I want to express my gratitude to Jan Forsberg who has been with me from the start of my studies. Thank you for the time you have put in. Our conversations have been a great inspiration for me personally and your encouragement has been immeasurable. I wish to thank my colleagues at the Department of Energy, Environmental and Building Technology at Karlstad University for creating an inspiring environment for learning. Finally, I would like to thank my family: my loving girlfriend for being a big support even though during a short time span of the project; my big brother for showing enthusiasm and challenging me with smart questions; my little sister for being my biggest fan and my biggest support in terms of enduring my frustration periods. And to my two greatest inspirations: my parents. My father for being proud of me and not letting a day go by without asking me about my studies. My mother who is the strongest person I know and the greatest source for me to find encouragement and strength. 6 إلى والدي هما مين و إحنا مين by Ahmed Fouad Negm هما مين و إ حنا مين ..؟ هما األمرا و السالطين .. هما المال و الحكم معاهم .. و إحنا الفقرا المحرومين حزر فزر ش ّغـل مخك شوف مين فينا بيحكم مين ؟ .............. إحنا مين و هما مين .. إحنا الفعال البنايين إحنا السنة و إحنا الفرض إحنا الناس بالطول و العرض من عافيتنا تقوم االرض .. و عرقنا يخضر بساتين حزر فزر شغل مخك شوف مين فينا بيخدم مين ..؟ ................ هما مين و إحنا مين ؟ هما االمرا و السالطين هما الفيال و العربية و النساوين المتنقية حيوانات استهالكية شغلتهم حشو المصارين حزر فزر شغل مخك شوف مين فينا بياكل مين ..؟ .................... إحنا مين و هما مين إحنا قرنفل على ياسمين إحنا الحرب :حطبها و نارها و إحنا الجيش اللي يحررها و إحنا الشهدا بكل مدارها منكسرين أو منتصرين حزر فزر شغل مخك شوف مين فينا بيقتل مين ..؟ ................ هما مين و إحنا مين هما االمرا و السالطين هما مناظر بالمزيكا و الزفة و شغل البوليتيكا و دماغهم طبعا استيكا بس البركة بالميازين حزر ..فزر شغل مخك 7 شوف مين فينا بيخدع مين ..؟ ..................... هما مين و إحنا مين هما االمرا و السالطين هما بيلبسوا أخر موضة و إحنا بنسكن سبعة في أوضة هما بياكلوا حمام و فراخ و إحنا الفول دوخنا و داخ هما بيمشوا بطيارات و إحنا نموت باالوتوبيسات هما حياتهم بامب جميلة هما فصيلة و إحنا فصيلة حزر فزر ..شغل مخك 1 شوف مين فينا حيغلب مين 1 A poem by late Egyptian vernacular poet Ahmed Fouad Negm. The translation of the title is “Who are they and who are we”. 8 Nomenclature ̇ Force field Source term (COMSOL) Surface tension Unit tensor Curve element Normal vector Velocity field Area Endpoints Drag coefficient Dimensionless friction factor Diffusion coefficient (COMSOL) Force Surface tension Piecewise interpolant Gravitational acceleration Neumann condition Mesh cell length Weight coefficient (COMSOL) Characteristic length of pore size Characteristic length scale of porous media Mass flux Number of points Normal Pressure Heat flux radii of tube Dirichlet condition Surface Temperature Time Specific surface area Intrinsic velocity Velocity magnitude Superficial velocity Local velocity coefficient Transport quantity (COMSOL) Propagation velocity of a curve 9 [N/m3] [N/m3] [N] [-] [-] [-] [m/s] [m2] [-] [-] [-] [m2/s] [N] [N/m3] [-] [m/s2] [-] [m] [-] [m] [m] [kg/m2s] [-] [-] [Pa] [W/m2] [m] [-] [m2] [K] [s] [1/m] [m/s] [m/s] [m/s] [m/s] [-] [m/s] Volume Volume flow rate Space coordinate Space coordinate at the interface boundary Cartesian coordinates Surface function ̇ ⃑ ⃑ z [m3] [m3/h] [m] [m] [-] [-] Greek symbols and mathematical operators ∑ ( ) Ω 〈〉 Convective transport term (COMSOL) Viscous resistance coefficient Volume fraction fluid phase Convective velocity vector Inertial resistance coefficient Source/sink term (COMSOL) Re-initialisation parameter Difference term Dirac delta function Porosity Interface thickness Dynamic viscosity (COMSOL) Brinkman screening length Permeability Curvature of interface Dynamic molecular viscosity Slip length Density Summation Surface tension coefficient Level-Set function Basis function Domain Partial derivative Boundary Gradient operator Volume-averaged property Subscript Tensor indices Cartesian coordinates Critical Drag 10 [-] [kg/m3s] [-] [m/s] [kg/m4] [-] [m/s] [-] [1/m] [-] [m] [Pa∙s] [m] [m2] [1/m] [Pa∙s] [m] [kg/m3] [-] [N/m] [-] [-] [-] [-] [-] [-] [-] Interface Particle Constant properties (COMSOL) Friction Artificial Surface tension Acronyms Ca Bo BK CAD CFD DOF FDM FEM FRV FVM GFEM IEA LS NS PDE Capillary Bond Blake-Kozeny Computer aided design Computational fluid dynamics Degrees of freedom Finite difference method Finite element method Flow representative volume Finite volume method Galerkin finite element method International Energy Agency Level-set Navier-Stokes Partial differential equation Pe Re REV VAT VF Péclet Reynolds Representative elementary volume Volume-averaging theory Volume force 11 List of Publications This dissertation is based on the following papers: Paper I: Rezk K. and Forsberg J. (2010). Geometry development of the internal duct system of a heat pump tumble dryer based on fluid mechanic parameters from a CFD software. Applied Energy. 88(5): 1596-1605 Paper II: Rezk K. and Forsberg J. A fast running numerical model based on the implementation of volume forces for prediction of pressure drop in a fin tube heat exchanger. Accepted for publication in Applied Mathematical Modelling, April 2014 Paper III: Rezk K. Nilsson L. Forsberg J. Berghel J. (2013). Modelling of water removal during a paper vacuum dewatering process using a Level-Set method. Chemical Engineering Science. 101(0): 543-553 Paper IV: Rezk K. Nilsson L. Forsberg J. Berghel J. Simulation of water removal in paper based on a 2D Level-Set model coupled with volume forces representing fluid resistance in 3D fiber distribution. To be resubmitted with a minor revision to Drying Technology. Paper V: Rezk K. Nilsson L. Forsberg J. Berghel J. Characterizing flow resistance in 3-dimensional disordered fibrous structures based on Forchheimer coefficients for moderate Reynolds numbers. Manuscript ready for submission. The following related publication is not included in this thesis: Rezk K. Nilsson L. Forsberg J. Berghel J. Using a Level-Set model to estimate dwell time in a vacuum dewatering process for paper. Oral presentation at the COMSOL conference in Milan 2012. 12 The Author’s Contribution Paper I: Planning was done on equal parts. All writing was performed by me. Conducted simulations were performed by me with Jan Forsberg as a discussion partner. Ola Holby and Jonas Berghel contributed with valuable comments. Paper II: Planning was done on equal parts. All writing was performed by me, except the experimental procedure, which was written by Jan Forsberg. Conducted simulations were performed by me with Jan Forsberg as a discussion partner. Measurements were performed along with Jan Forsberg. The sensitivity study of the experimental process was conducted by Jan Forsberg. Ola Holby, Jonas Berghel and Lars Nilsson contributed with valuable comments. Paper III: Planning was done on equal parts. Most of the writing was performed by me with the help of Lars Nilsson in the introduction and paper sheet characterisation. The conducted simulations were performed by me with Jan Forsberg as a discussion partner. Measurements and compilation of experimental data from the literature were performed by Lars Nilsson. Lars Nilsson, Jan Forsberg and Jonas Berghel contributed with valuable comments. Paper IV: Planning was done on equal parts. All writing was performed by me. Conducted simulations were performed by me with Jan Forsberg as a discussion partner. Lars Nilsson, Jan Forsberg and Jonas Berghel contributed with valuable comments. Paper V: Planning was done on equal parts. All writing was performed by me. Conducted simulations were performed by me with Jan Forsberg as a discussion partner. Lars Nilsson, Jan Forsberg and Jonas Berghel contributed with valuable comments. 13 Table of Contents ABSTRACT .................................................................................................................1 SUMMARY IN SWEDISH .....................................................................................2 SUMMARY PAPERS ................................................................................................3 PREFACE ....................................................................................................................5 ACKNOWLEDGEMENTS ....................................................................................6 NOMENCLATURE .................................................................................................9 LIST OF PUBLICATIONS ................................................................................. 12 THE AUTHOR’S CONTRIBUTIONS .......................................................... 13 TABLE OF CONTENTS .................................................................................... 14 1. INTRODUCTION .......................................................................................... 16 1.1. Objectives ...................................................................................................... 18 2. BACKGROUND ............................................................................................... 20 2.1. Commercial CFD codes .............................................................................. 20 2.2. Communication interface ................................................................................ 22 2.3. Volume-averaging theory ............................................................................... 24 3. BACKGROUND, INDUSTRIAL APPLICATION FIELDS .............. 26 3.1. Heat pump tumble dryer ................................................................................ 26 3.2. Papermaking ...................................................................................................... 27 3.2.1. Paper machine.......................................................................................... 28 3.2.2. Vacuum dewatering ................................................................................. 29 3.2.3. Forming wire ........................................................................................... 30 4. THEORY OF FLUID MECHANICS........................................................ 32 4.1. Pressure in fluid flow .................................................................................. 33 4.2. Navier-Stokes equations .............................................................................. 37 4.3. Accounting for turbulence ........................................................................ 38 4.4. Two-phase flow ......................................................................................... 39 4.5. Flow in porous media .................................................................................. 41 4.5.1. Fluids and porous media as continua ....................................................... 41 4.5.2. Flow representative volume....................................................................... 42 4.5.3. Reynolds number .................................................................................. 43 4.5.4. Viscous and inertial contributions in porous media .................................. 46 5. METHOD .......................................................................................................... 49 5.1. Modelling in COMSOL ............................................................................... 49 5.1.1. Finite Element Method .......................................................................... 49 5.1.2. Numerical instabilities ........................................................................... 51 5.1.3. Boundary conditions............................................................................... 53 14 6. 7. 8. 9. 5.1.4. Level-Set method ................................................................................... 54 5.2. Volume forces coupled to single and two-phase flow ............................ 57 5.2.1. Pore-scale simulations ............................................................................... 57 5.2.2. Modelling surface tension .......................................................................... 59 5.2.3. Closure model .......................................................................................... 60 5.2.4. Sensitivity studies ..................................................................................... 62 5.2.5. Volume forces in structureless models ........................................................ 65 5.3. Characterisation of flow resistance in various fibrous structures .......... 68 5.3.1. Algorithm for production runs in COMSOL ........................................... 68 5.3.2. Assessing flow resistance based on Forchheimer coefficients ......................... 71 RESULTS AND DISCUSSION ................................................................... 73 6.1. Fitting procedure of resistance coefficients: Paper II, III, IV, V .......... 73 6.1.1. Concluding remarks ................................................................................. 75 6.2. Implementation of volume forces as sink terms: Paper II, III, IV ....... 75 6.2.1. Concluding remarks ................................................................................. 81 6.3. Sensitivity study of pore-scale simulations: Paper II, III, IV, V ............ 82 6.3.1. Concluding remarks ................................................................................. 84 6.4. Experimental validation of simulation models: Paper II, III, IV .......... 85 6.4.1. Concluding remarks ................................................................................. 86 6.5. Assessment of flow resistance coefficients: Paper V .............................. 87 6.5.1. Concluding remarks ................................................................................. 90 6.6. Assessment of numerical based design process: Paper I ........................ 91 6.6.1. Concluding remarks ................................................................................. 93 CONCLUSSIONS ............................................................................................ 94 FUTURE RESEARCH ................................................................................... 96 REFERENCES ................................................................................................. 97 15 1 Introduction In recent years, computer based programs, for instance various commercially available CFD tools, have been applied to solve and analyse industrial problems. CFD is the abbreviation for Computational Fluid Dynamics, which is one of the branches of fluid dynamics. Increased developments of highperformance computer hardware, as well as the introduction of user-friendly interfaces, have paved the way for the extensive application of CFD in various fields. Such fields include the aerodynamics of aircrafts and vehicles (reducing the drag coefficient of a car for lower fuel consumption), chemical process engineering (improving the mixing capability of a static mixer, which is applicable in wastewater treatment or refinery in the oil and gas market), biomedical engineering and drying technology. Drying technology is a typical multidisciplinary topic that covers fields such as the pulp and paper industry as well as the agricultural industry to name a few. By way of a recent progression of computer efficiency, several aspects of CFD usage have been compiled into an advantageous approach over experimentalbased approaches in fluid systems’ design that results in: the possibility to execute detailed analyses of a system; the capability to simulate large control volumes, which present difficulties in physical laboratories; a possible tool for communication of results, based on detailed visualisation images; a time and cost efficient tool in various fluid design processes. Even though a numerical design process is not capable of replacing pilot-scale testing completely, it reduces the expense due to a reduced need of manufacturing prototypes for measurements, and fewer rentals of large experimental facilities and measurement equipment. Large industrial companies generally encompass an experienced CFD department with vast full-scale experimental laboratories. In addition, they are in possession of computer cluster networks. Numerous small-sized companies, however, lack experience in the subject, computer resources and extensive experimental facilities. User-interfaces in CFD programs have developed rapidly. Yet, the history of research concerning the theory of fluid dynamics is extensive. Consequently, 16 CFD software in small-sized industrial companies is utilised as a black box, based on the knowledge of fluid mechanic theory. With the availability of userfriendly interfaces, CFD calculation can be conducted on a practical basis in various industries. Yet again, a crucial part in the simulation process is to evaluate data by using visual analysis of flow patterns, analysis on the sensitivity of the mesh grid and investigation of quantitative parameters such as pressure loss, velocity, turbulence intensity, etc. There is, however, a need for collaboration involving the industry and the academic world in order to carry out applied research towards the objectives of the industry. A direct partnership involving researchers and industry generally yields to a verbal communication interface, which is a crucial step in the process given the level of mutual dependency. In this study, the aim is focused on the methods of reducing the geometrical complexity of systems involving the transportation of mechanical energy. In an engineering design cycle, there is a demand for quick analysis in order to evaluate the modified design and performance of the product or process. Even though high computer power is accessible these days, there is a need for time efficient simulations. The most common approach, which basically every CFD engineer is utilising, is to divide the computational domain into several subdomains in order to reduce the total amount of Degrees of Freedom (DOF). A crucial aspect in this process is determining where the boundaries of these subdomains are most representable. A continually evolving method in reducing DOF in simulation models is volume-averaging constitutional variables such as velocity, pressure and temperature, to name a few. With this method, volume forces that represent flow resistance can replace complex geometrical configurations, which will reduce the total amount of DOF while solving the flow problem. This enables time efficient simulations of systems so that various geometrical configurations and process parameters could be analysed numerically. This study originates from the fact that there is a need to utilise the communication process between the different professions in a numerical based design process, and establish fast running numerical models based on volume forces that represent flow resistance in geometrically complex structures. The fields of application in this thesis are: 17 i. ii. Improving the aerodynamic capabilities of the internal duct system in a heat pump tumble dryer. Analysis of process parameters in various running conditions in a vacuum dewatering process in a paper machine. The volume-averaging method is not limited to these application fields, however, as the prospect of reducing geometrical complexity is achievable in many industrial applications. Moreover, the demand for fast running numerical models is crucial in various design processes in both small and large scale companies. 1.1 Objectives The objectives of this thesis is to investigate possible ways of reducing the complexity of geometrical structures in terms of minimising the computational time. The essence of the investigation is to present possible methods based on the capacity of standard personal computer hardware. For this, two methods are utilised: i. ii. Analysis of the early implementation of CFD analysis in a numerical lab design process, in which systems’ complexity is reduced by appropriately subdividing the system into trivial components. Implementation of volume forces as sink terms representing flow resistance in the momentum equation for a single- and two-phase flow. In (i), CFD software is utilised as a communication tool during a numerical based design process, in cooperation with a household appliance company. The design process is conducted through a communication interface between the researcher and a mechanical engineer, which is a representative of the company. The focus in the first point, (i), is: 1. Assess the contribution of early implementation of CFD simulations in a numerical based lab design cycle process. The second point, (ii), investigates the validity of implementing volume forces in systems thereby regarding them as porous media. The first application area is simulating pressure drop and flow rate relations in a staggered pin-fin heat 18 exchanger as an integral component of a heat pump tumble dryer. The second application area is simulating a vacuum dewatering process based on representing the flow resistance of the fibrous structure of paper and the woven forming wire with volume forces. The focus in the second point, (ii), is: 1. Validating system parameters based on volume force implementation in structureless models. Moreover, analysing the sensitivity of these models based on mesh configuration and numerical sample points used to estimate flow resistance. 2. Validating volume force implementation and flow resistance characterisation to experimental and literature data. 3. Assessing the contribution of volume force implementation in analysed systems in terms of evaluating computational time. 19 2 Background This chapter will provide a brief survey of commercial CFD usage, the importance of CFD based design processes and a view on the utilisation of CFD implementation in companies with a lack of experience in the field of fluid mechanics. Moreover, a review of the volume-averaging method is presented. 2.1 Commercial CFD codes A CFD program uses numerical methods to implement a discretisation on Partial Differential Equations (PDE) and to solve them. The purpose of using CFD is to solve complex fluid flow relations, which is not feasible with standard calculus. With the help of a CFD program, it is possible to simulate and visualise; e.g. the airflow around the wing of an aircraft to determine the drag and lift force on the wing, or the flow in a fuel cell stack to determine whether the flow and the design of the channels are proper for a certain product. Most commercial CFD codes involve three main elements: Pre-processor Solver Post-processing • Geometry design • Mesh grid generation • Physical phenomena • FEM • FVM • FDM • 2D and 3D surface plots • Vector and streamline plots • Display of domain geometry Fig.(1). Representative elements for a general CFD code. The pre-processor constitutes a computational domain, which can be designed by the user in the CFD code or imported from an external Computer Aided 20 Design (CAD) program. Mesh grid generation is an essential part of the simulation process. It represents a grid that divides the computational domain into subdomains, in which the PDEs are discretised and solved numerically at the nodes of the grid intersections. In general, large numbers of cells wield better solution accuracy. Numerous commercial codes comprise their own mesh grid generation features and major CFD codes are compatible with external CAD and mesh generator programs. Hence, more advanced mesh grid structures can be imported to the CFD codes. In recent CFD codes, predefined PDEs, along with material property libraries are accessible to the user. Such PDEs include various physical processes such as energy, mass and momentum transfer. The Finite Difference Method (FDM) is one of the methods used to discretise the equations of flow for a computational solution. The method approximates the solution to algebraic equations using the Taylor-expansion series through a forward, backward and/or central difference scheme. The Finite Volume Method (FVM) discretises a small volume surrounding a node point on a mesh. This method generates conservative PDEs, since the flux that enters a given control volume is identical to the one that leaves the adjacent volume, due to the divergence theorem. In the Finite Element Method (FEM), the PDEs over the elements are approximated by local functions such as polynomials, resulting in a linear system of algebraic equations. The boundary and initial value conditions are formulated in either a weak or integral form. The subdivided elements are unique, as the grid could be shaped as tetrahedrons or hexahedrons for a 3-dimensional domain. This is a result of the fact that the FEM becomes more flexible as more complicated geometries can be treated. One of the most crucial steps when utilising a commercial CFD code is evaluating the data. Due to increased implementation of commercial codes in the industry, the graphic capabilities of such programs have vast and versatile data visualisation toolboxes. Output data can be presented as 2-dimenisonal and 3-dimensional with colour, vector, contour lines and streamline plots. Typical output data includes velocity patterns, heat flux, temperature and pressure distribution. 21 2.2 Communication interface The communication process of various fields of knowledge is an issue involving the integration of CFD calculation and structural geometry design. The complexity involving CFD usage leads researchers such as Koikekoi [1], to study strategies of practical engineering software applications in order to amplify the implementation of software for the next generation. A concept of an interface instrument for different professions was proposed by Stenzel and Pourroy [2]. The implementation focuses on product knowledge as a tool to initiate communication between professions and to support their learning about the product. Petridis and Knight [3] acknowledged the difficulties of implementing a CFD software package into a so-called Intelligent Knowledge-Based System. They suggested a blackboard model to facilitate the use of CFD data by various engineering groups. A numerical based design process can be utilised in different ways. Numerical based design of various components is one of many study fields that has evolved, along with the development of CFD codes and improved computer power. There are several ways to carry out geometrical improvements through CFD calculations. El-Sayed et al. [4] integrated an optimisation code into a CFD code to create a shape optimisation tool to study the design of an aerofoil and an S-shaped duct. The aerodynamic performance of a wing has been optimised by combining the CFD software Fluent and the iSIGHT design platform [5]. Another method for designing a geometrical structure is to visually analyse the airflow patterns in the simulation model. In an article Yakinthos et al. [6], a 2dimensional approach was used to solve the Launder-Sharma low-Reynolds k-ε model in order to capture recirculation regions close to the walls of heat exchangers installed in a recuperative aero engine. Several configurations were investigated to achieve the lowest value of the pressure drop. Lakshmiraju and Cui [7] implemented a k-ε model in the CFD software Fluent. Detailed data was established on the flow field in order to analyse the mechanisms of pressure loss in a power plant stack model. A typical numerical based design process involving several professions is described in Fig (2). It is viewed as a deterministic algorithm, considering the fact that each procedure is operated under the same considerations. Geometry modifications are heuristically determined based on visual images of flow patterns and design criteria for the model. 22 CAD Simulation process Geometry modification by the company’s mechanical engineer Evaluation of flow parameters Design criteria Fig. (2). Schematic description of the design process. Due to its complex geometric structure, considerations of how fluid mechanic parameters are visually presented are critical in order to amplify the communication. It is essential to present a high spatial resolution in the domain that provides the mechanical engineer with detailed information on where unfavourable flow behaviour is generated. The evaluation of the flow parameters is the criterion that determines the modification to the geometry. The modifications are confirmed at the end of the design loop by comparing the design criteria with the previous model. The design criteria are usually coupled to integrating constitutional variables such as temperature, pressure and velocity. The criteria can vary depending on the system at hand, e.g. heat flow depends on the geometrical configurations of a heat exchanger, and the dewatering rate depends on the arrangement of the fibers constituting the paper sheet. The design criteria and visual data are the foundation of the communication process. It is preferable to communicate the CFD results verbally. Otherwise, visual data of flow patterns can be misinterpreted. The communication is a continuous process and it is looped until further modifications are no longer possible. 23 2.3 Volume-averaging theory The Volume-Averaging Technique (VAT) is used to derive continuum equations for multiphase systems. The purpose of the VAT theory is to reduce the complexity of geometrical structures such as heat exchangers, paper sheets and other dense structured systems. The method focuses on moving the attention from establishing valid equations in the pore-space, to using the porescale characteristics to derive volume-averaged equations which are spatially smoothed over in the entire domain. Upon volume-averaging on the pore-scale level, integrals involving micro-scale variations of the averaged variable arise. These integral equations present a closure problem in which a scheme of how to obtain volume-averaged expressions of the local dependant variable of the integral equations is essential. Much of the early work in developing the VAT theory was conducted during the 1960s [8-10]. These papers present a rigorous mathematical foundation for the VAT. The development of the theory of VAT and its application in various cases is presented by Whitaker [11]. The implementation of the VAT has been extended to analysing highly porous and heterogeneous structures in which turbulent effects need to be considered [12-14]. The principles of volume-averaging are applicable on single and multiphase systems consisting of different fluids. In the case of modelling multiphase systems, an important issue is establishing the correct mathematical descriptions of boundary surfaces that separate the phases [15-18]. Volume-averaging theories have been recently applied to heat exchanger analysis. Based on the fact that overall heat exchangers comprise a dense structure of an array of fins and tubes, the heat exchangers are viewed as porous media in which the specific geometry of the original structure is replaced with averaged properties of the flow process. By using the VAT theory, the momentum and energy equations can be solved for several types of heat exchanger configurations. Major contributions in VAT theory applied on heat exchangers have been established [19-26]. These papers have, in general, contributed to establishing fast running computational algorithms based on the VAT to simulate airflow through mostly an aluminium chip heat sink with a staggered array of pin-fin configuration and fin-tube heat exchangers. They handle the closure problem with two essentially different approaches. Much of these studies used empirical relations based on the experimental work of drag and heat transfer coefficients in order to handle closure problems from upscaling at the pore-scale level to the macro-level [19-21]. The other approach is 24 that CFD simulations are used in their Representative Elementary Volumes (REV) in order to obtain solutions for their integral terms, which represented local heat transfer and local pressure drag [22-26]. The method enabled them to investigate possible geometric improvements to achieve higher thermal effectiveness. Based on the fact that these days, empirical relations from experimental work and CFD calculations are used to handle closure problems; it is essential that the applied theory is compared to experimental data regarding the flow process. 25 3 Background, Industrial Application Fields This chapter presents the application fields encountered in this thesis, which are the heat pump tumble dryer and the vacuum dewatering process in a paper machine. Certainly, CFD implementation in design processes and establishing time efficient simulations in systems regarded as porous media are not limited to these fields, as these aspects have become more crucial to companies in general. 3.1 Heat pump tumble dryer Household appliance products constitute one of the many energy demanding markets. They consume almost one third of all of the electricity produced by IEA (International Energy Agency) member countries [27]. The tumble dryer is one of several energy demanding household appliance products in, among other places, residential homes. A report by ETSAP [28] presents some interesting data. They state that in the year 2009, approximately 79% of US households owned a laundry dryer, and that in the EU there is an annual sale of approximately 30,000 units. These units mostly consist of electrical dryers. As there is a general growth in population, the demand for household appliance products is also increasing. Hence, it is crucial to reduce the energy demand for such products and to make continuous efforts to increase their efficiency. The heat pump tumble dryer is a new concept in the household appliance industry. The working medium of the tumble dryer is transported in a closed internal duct system, see Fig (3). The humid air is transported from the drum into a heat exchanger unit, where the evaporated water is condensed in the first heat exchanger it meets with. In the second exchanger, the air is heated again before it is transported back to the drum. The energy of condensation is utilised as a low-temperature energy source, which the working medium in the heat pump absorbs. Compared to its predecessors, the heat pump dryer is extensively more energy efficient. As a result of it, the dryer only uses half of the electricity of a traditional condensing dryer [29]. Moreover, the life cost cycle of the heat pump dryer is more profitable than the other drying concepts, making it a sustainable approach towards the drying of clothes [30]. 26 Fig. (3). Components of the internal duct system of a heat pump tumble dryer. The analysis of the airflow behaviour has yet to be studied or published to a further extent. Since the production and the manufacturing of tumble dryers are continuously increasing, analysis of the airflow could provide some motivating results in terms of reducing pressure drop and noise pollution. 3.2 Papermaking Papermaking is the process of producing paper, which today is mostly used for printing, household usage and packaging. The concept of papermaking dates back to 2000 years ago when the Chinese collected various plants such as mulberry bark and hemps and separated the fibers onto old rags which they laid in the sun to dry the web of the fibers, which resulted in a paper sheet. During the timeline of papermaking, the method evolved and spread to the Islamic world where the process of papermaking was refined, as machinery was designed for bulk manufacturing. Papermaking reached Europe in the late 11th century where paper mills were introduced in countries such as Spain and France, to name a few. However, in conjunction with the industrial revolution, modern papermaking took place, which catapulted the production of paper worldwide [31]. 27 3.2.1 Paper machine Modern papermaking took place in the early 19th century with the development of the Fourdinier machine. The concept of producing paper in this machine differs from previous methods because the paper is produced in continuous sheets rather than being produced in individual sheets. Fig. (4). Water removal in a Fourdinier paper machine [32] . The modern paper machine consists of three main operational sections. The first stage is the forming section; water is removed by gravity and vacuum filtration is applied at various pressure levels. The second stage is the press section; water is pressed out of the web through mechanical pressing in several press nips. The drying section is the final stage where the majority of the remaining water is removed in a series of steam heated cylinders. The fiber suspension enters through the head box in the forming section, where the dryness is approximately 1%. Most of the water in the fiber suspension is removed in the forming section, and the dryness of the paper is generally at a level of 20%. At the end of the press section, the dryness is roughly at 40% and the final dryness is reached in the drying section, which is approximately 95%. There is a high demand of energy for paper machines during the production of paper. In the year of 2012, 8,525 paper machines were in operation and the average annual electrical use for one paper machine was 140 TJ [33]. The ratio of this demand is approximately 20% for electrical energy input to the vacuum system. The thermal drying section considerably exceeds the other dewatering stages in terms of energy use. It was reported that 80% of the total energy used in a paper machine is steam used for the dryers [34]. Hence, increasing the incoming dryness to the thermal stage could lead to a major reduction of the production cost. 28 Many attempts on further developing the process of papermaking have involved optimising control systems for various components and sections of a paper machine [34-36]. These studies implemented control systems which improved wet end capabilities in terms of improving the consistency of the formation of the paper sheet and reduction of the unpredictability of the wire retention. Moreover, they were able to reduce specific steam consumption on the dryers by 10%. Bhutani et al. [37] developed a solution for a paper machine steam energy fingerprint. They investigated a dryer section of a paper machine with a potential steam savings of 10-15%, which was achieved by optimising the control set points for a steam dryer group. 3.2.2 Vacuum dewatering The process of applying vacuum in several suction boxes occurs in the forming section. The forming section is generally divided into two zones. In the first zone, the diluted fiber suspension is sprayed onto the forming wire from a head box, whereafter water is removed by gravity and low vacuum levels are initiated by dewatering elements such as foils. The second zone comprises suction boxes, generally placed under the forming wire. These boxes apply higher vacuum pulses at levels ranging from 15 to 65 kPa. Most of the research conducted on vacuum dewatering has either been done by using laboratory equipment or by using pilot paper machines. The process parameters that are usually investigated are the influence on the dryness by the applied vacuum, the dwell time and the basis weight [38-44]. The mechanisms involved in vacuum dewatering have been described as web compression, displacement of water by air and rewetting from the forming wire [42,45,46]. Even though there are established design data based on experimental work done on vacuum dewatering, there is still a lack of validated CFD-models predicting dewatering rates in relation to dwell times for various basis weights of paper sheets. Capturing all dewatering mechanisms is challenging, considering the movement of fibers and the inclusion of capillary forces to capture the rewetting phenomena. However, conducting experimental analysis on large scale paper machines requires costly resources. Establishing numerical models would yield a more cost efficient process for analysis of process parameters such as the basis weight of the paper sheet, testing various vacuum levels and 29 duration of pulses, analysing the influence of different forming wire configurations, etc. 3.2.3 Forming wire The forming wire is an essential component in the process of paper production. The wire is a 3-dimensional woven matrix which consists of orthogonal crossdirection filaments in several layers, see Fig. (5). During the process, the wet pulp is dewatered while on the forming wire, which in turn yields the formed paper sheet. Hence, the capability of the forming wire is significant in terms of achieving a more energy efficient dewatering process; more economic sustainability, by reducing the tear of the wire material; improvement in the end quality of the paper sheet. 2 Fig. (5) . A CAD-model illustrating the arrangement of the forming wire. There has been research done regarding the influence of the forming wire on the dewatering process. Granevald et al. [47] studied the influence of ten different forming wires on the sheets’ solid content and the dewatering rate during vacuum dewatering of low grammage sheets. They concluded that caliper, void volume and air permeability are three important parameters. The forming wire consists of yarns with a diameter typically around 100-200 µm on the paper side, and a pitch of 200-400 µm. The geometric configuration is far 2 Albany International Corporation 30 coarser than the paper, which should not contribute much to the total flow resistance. Still, there is a reason to investigate the influence of the contact surface between the paper and the forming wire as a result of the forming and packing of the fibers on the pore space during the high pressure vacuum process. In a review article by Hubbe et al. [48], they addressed the issue involving the interaction between fibers and the forming wire. Aside from improving the performance of the forming wire through analysis of flow properties, surface treatment was conducted in terms of a heating compaction process that flattened out the top surface yarns on the paper side of the forming wire [49]. This reduced the fabric caliper and internal void volume by 515% and, in turn, improved vacuum dewatering efficiency. 31 4 Theory of Fluid Mechanics The mathematics that describes flow behaviour is challenging, yet we encounter complex flow patterns in everyday life. Water, for instance, flows out of a tap, smoke rises up a chimney, the irregular motion of a leaf falling from the tree and the flow distribution in a lecture room are only a few of an innumerable amount of events that involve complex flow patterns. It is difficult to establish when the study of fluid motion first took place, due to the art of documentation being relatively new to mankind. The development of hydraulics was purely empirical, as our ancestors built farms where the utilisation and transportation of water were crucial. The discovery of constructed irrigation canals dating from 4000 years B.C. was made in ancient Egypt and Mesopotamia. More advanced water systems for transportation and storage were constructed in old civilisations such as Jerusalem, Greece and, especially, the Roman Empire. The evolution of the ship is a result of empirical studies of fluid mechanics. Mankind first developed simple boats out of logs and moved towards advanced ships with sails, which allowed them to navigate the oceans. The Phoenicians and Egyptians are known for having built excellent ships. During the second half of the 15th century, a polymath named Leonardo Da Vinci was one of the first scientists to visualise and document fluid flow in detail. He wrote extensive descriptions of the movement of water in the form of eddies, water waves, free jets and falling water, to name a few. His work paved the way for future research in the fields of hydraulics, as well as hydrodynamics. In the late 17th century, Isaac Newton tried to quantify fluid flow through the elementary Newtonian physical equations. This includes the concepts of the Newtonian viscosity and reciprocity principle. His famous second law: , is the foundation for the deduction of the Navier-Stokes 3 (NS) equations. 3 Worlds of flow – A history of hydrodynamics from the Bernoullis to Prandtl. New York, 2005 32 4 5 Fig. (6) and (7) : Studies of water passing a solid object and water falling into still water by Da Vinci. In the mid-18th century, Daniel Bernoulli contributed significantly to the field of hydrodynamics by trying to mathematically describe the motion of fluids. In his treatise, Hydrodynamica, he acknowledged the properties of basic importance with regard to fluid flow such as pressure, density and velocity. His famous Bernoulli’s principle was crucial in the early studies of aerodynamics. During the time of Bernoulli, Leonard Euler proposed the so-called Euler’s equation, which describes conservation of momentum and mass for an inviscid fluid. George Gabriel Stokes, who introduced viscous transport to the equations, and together with Claude Louis Marie Henry Navier, conducted further developments on Euler’s equations, These equations form the basis of modern day CFD usage.6 4.1 Pressure in fluid flow The concept of pressure is an integral part of the field of fluid dynamics. The unit for pressure is Pascal (Pa), which equals force, (N), divided by unit area (m2). Hence, the definition of the term pressure is that of a force that is perpendicularly applied to the surface of an object. The term is a scalar quantity, as it relates the normal vector of the surface and to the force vector acting in a normal direction to it. The concept of pressure was introduced in ancient Greece when Archimedes presented the principle of buoyancy, stating that force acting on an immersed body is equal to the weight of the water http://www.siggraph.org/publications/newsletter/volume-42-number-2/interacting-with-threedimensional-flow-fields, 2011-01-10, 00:57 5 http://www.portlandart.net/archives/2009/01/art_and_nature.html, 2011-01-10, 01:35 6 Worlds of flow – A history of hydrodynamics from the Bernoullis to Prandtl. New York, 2005 4 33 volume it displaces. The principle is based on a static system in which the pressure acts on a non-moving object. Pressure can be identified at every point in a body of fluid, regardless of the fluid being in motion or not. The first equations coupling velocity to pressure in fluid motion equations were introduced by Daniel Bernoulli in his treatise Hydrodynamica. They are described in the so-called Bernoulli's law. He used the concept of water decanting in a vertical vessel as a result of the efflux of water in an attached tube in the lower section. He stated that the acceleration of the water in the intersection of the vessel and the tube is equal to the loss of the fluid's potential energy. He put it in words that the exerted pressure by the fluid on the walls of the vessel is equal to the difference between the hydrostatic pressure and dynamic pressure. This statement led to what is known today as Bernoulli's equation: (1) Later on, Daniel's father, Johann Bernoulli, stated that the pressure of the fluid in the vessel is a result of adjacent fluid parts exerted on one another, which he called internal pressure. This pressure is today better known as the static pressure of the fluid, and it is represented by the first term on the left hand side of Eq. (1). The second term to the left of Eq. (1) is often called the datum of a pipe or channel and the third is the dynamic pressure, which is called kinetic energy, if Eq. (1) is interpreted in energy units. Bernoulli's statement is that the energy density is constant following a streamline in the flow, which is the sum of all terms in Eq. (1). Moreover, friction losses due to viscous stress are disregarded. When studying fluid flow through a pipe or channel, a similar approach to determine the relation between the velocity and pressure is to apply the conservation of energy. If the frictional losses were disregarded, it would result in Bernoulli's equation. With a presented control volume of a fluid channel with a steady state flow at hand, however, the conservation theorem can be set up in different forms of conservative units such as mass, momentum or mechanical energy. The continuum principle is one of the central assumptions in this field. Here, the discrete nature of matter is overlooked, given the fact that the length scale of the macroscopic matter is large, compared to the atomic discrete length scales. Thus, the matter is treated as a continuum, which indicates that the 34 quantities of the matter are continuously differentiable. In 1750, Leonhard Euler claimed that Newton’s second law of motion applied to infinitesimal bodies was the true basis of continuum mechanics. He stated that the acceleration of a fluid element depends on the combined effect of the pressure gradient and external forces, such as gravity. Based on Newton’s second law, Euler obtained what is known as one of the first PDEs for fluid motion. 7 (2) Eq. (2) is written for a steady state process, as the time derivative is disregarded. To study the conservation of an arbitrary control volume, the divergence theorem by Johann Carl Friedrich Gauss is used to clarify the balance of the equation. The divergence theorem states that the outward flux of a vector field through the surfaces of a closed control volume is equal to the volume integral of the divergence of the vector field in the region.[50] ∫ ∫ (3) The term on the left hand side of Eq. (3) represents the divergence of the vector field inside the control volume, which represents the sum of all sources and sinks. The term on the right hand side of Eq. (3) represents the vector field acting in the normal direction at the surface of the control volume. Hence, this theorem is a conservation law, and it is applicable in other fields such as electromagnetism and quantum mechanics. Euler’s momentum equation may be applied to an arbitrary control volume in which the sum of all forces on the flux boundaries is determined. The force that represents the vector field described in Eq. (3) is the parameter of interest in the control volume. 7 Worlds of flow – A history of hydrodynamics from the Bernoullis to Prandtl. New York, 2005 35 Fig. (8). Momentum flow balance over an arbitrary control volume. Applying force balance in terms of momentum fluxes and pressure forces on the boundaries of a control volume indicate the state at each end point, see Eq. (4). ∑ (∫ ∫ ) (4) According to the divergence theorem, the sum of all forces that are exerted on the fluid may act like sinks in the region. 4.2 Navier-Stokes equations The most common PDE in fluid flow analysis are the NS equations. The equations are nonlinear PDE that state that the momentum change rate per unit volume of a fluid element (term on the left hand side of Eq. (6)) corresponds to pressure (first term on the right hand side) and viscous forces (second term on the right hand side) acting on the element boundaries. The continuity equation states that mass flux entering and leaving the control volume is equal to the change in mass, see Eq. (5). 36 (5) (6) In some cases, gravitational forces need to be introduced. When dealing with air, this term is often negligible. When dealing with the incompressible form of the NS equations, the term is constant. In the view of flows with significant temperature variation, buoyancy effects need to be considered. Moreover, the compressibility in the NS equations needs to be considered when dealing with high velocities (Mach number > 0.3). The NS equations are similar to Euler’s momentum equation in Eq. (2), except for the added term of viscous forces. The indices and in Eq. (2), (4),and (6) represent three coordinate equations in which and may take on the values of 1, 2 or 3. If we want to write the NS equations in the x-coordinate, we set to 1. Since the index appears twice in the convective and diffusive term, we have to apply the summation rule due to the fact that the index is regarded as a so-called dummy index. Further contributions to the boundary layer theory and turbulence were made by Ludwig Prandtl during the first half of the 20th century. While Prandtl was working in die Maschinenfabrik Augsburg-Nürnberg, he attempted to improve a suction device for the removal of shavings. During the course of the project, he realised that the pressure rise expected in a divergent tube failed to occur because the lines of the flow tended to separate from the walls. On the basis of this observation, Prandtl founded his well-known boundary-layer approach to resistance in viscous fluids. Prandtl stated that a fluid with a relatively low viscosity, such as water or air, is inviscid in a flow domain; except near the wall where the fluid adheres to the solid wall. This region is called the boundary layer and it is based on the assumption of the no-slip condition. Furthermore, he put forth several approximations for the boundary layer through scale analysis, in which he determined that for a 2-dimensional flow, the streamwise velocity is larger than the spanwise. Yet, the derivative (change) in the spanwise direction is larger. Moreover, he determined that the change in pressure over the boundary layer thickness is negligible.8 8 Worlds of flow – A history of hydrodynamics from the Bernoullis to Prandtl. New York, 2005 37 4.3 Accounting for turbulence Already in the late 18th century, Daniel Bernoulli and his father shared a preconceived notion about mechanical losses that was later put forth by Leibnitz. The notation states that losses of mechanical energy in fluid motion occur in the interaction of small-scale motion. Bernoulli's equations exclude fluid resistance, which was judged by Bernoulli to be beyond the grasp of mathematics at the time.9 Turbulence is a common occurrence in practical fluid flow. Turbulence is characterised by randomness or chaotic behaviour of the flow in which turbulent eddies occur at different velocity, length and time scales. High values of momentum diffusion occur in turbulent flow, due to effective mixing caused by the eddy motion. The length scale of the largest eddies in a flow process is of the same order as the flow geometry. These eddies are created in presence of a mean velocity gradient, hence kinetic energy in the mean flow is transferred to the eddy motion of the largest scales. The transfer of kinetic energy continues to a smaller eddy and so on. At the smallest scale, the eddy dissipates to thermal energy, as the kinetic energy in the eddy motion is not able to overcome the viscous forces in the fluid. The process in which the kinetic energy is transferred from the largest eddies to the dissipative ones is called the cascade process. The dissipative scales were put forth by Russian mathematician Andrey Nikolaevich Kolmogorov, they are famously known as Kolmogorov scales. In theory, the NS equations describe laminar as well as turbulent flow conditions. With increasing Reynolds number however, disturbances in the flow are created due to the development of fluctuations in the velocity and pressure fields. Hence, in the presence of turbulence, it would require a great amount of mesh elements to resolve all the turbulent scales in spatial coordinates and a fine resolution in time, given the fact that turbulent flow is unsteady. Direct numerical simulation is equivalent to applying the NS equations without an additional turbulence model to resolve a flow process. Such simulations are generally not feasible for standard computer usage, as the mesh grid easily exceeds millions of elements for relatively small Reynolds numbers. This issue has paved the way for extensive research on developing turbulence models. 9 Worlds of flow – A history of hydrodynamics from the Bernoullis to Prandtl. New York, 2005 38 4.4 Two-phase flow Systems involving simultaneous fluids, e.g. a gas and liquid phase that are immiscible, are referred to as a two-phase flow system. Other classifications of two-phase flow are liquid-solid and gas-solid. In practical systems, the twophase flow phenomena appear in a wide variety of engineering conditions, e.g. dewatering during papermaking. Key areas have benefited from two-phase flow studies such as pump cavitation, in which vapour pressure near regions of the pump or propeller can be detected. In industrial systems such as power plants, two-phase flow phenomena occur in combustion and boiling processes in which the behaviour of the flow, heat transfer and pressure evolution are significantly different from single-phase systems. In two-phase flow systems with an interface boundary, phenomena are encountered which are caused by the existence of surface tension on the interface boundary of the fluids. Cohesive and adhesive forces arise from intermolecular action close to the surface. These forces arise at the surface of the fluid due to unsymmetrical net force balance, thus causing the molecules to contract to a minimal area of the liquid surface, e.g. formation of droplets. In other words, surface tension affects the dynamics of the flow process as there is a change in the boundary condition of the interface. A classical description of surface tension forces is presented in Fig. (9), which depicts a capillary tube action. Fig. (9). Schematic description of surface tension forces in a capillary tube. The equilibrium of forces is the surface tension forces due to curvature balanced by the weight of the liquid. The surface tension is described in differential form according to Eq. (7) [51]. 39 (7) In Eq. (7), the left-hand side is the force acting on a surface at a location, based on the cross product of the curve element and the surface normal determining the position and direction of the force. In the case of the capillary tube, the body force is cancelled out by surface tension, according to Eq. (8). (8) The significance of the capillary force could roughly be determined by the dimensionless Capillary (Ca) number or the Bond (Bo) number; see Eqs. (9) and (10) [52,53]. (9) (10) The Ca-number describes the ratio of viscous forces to capillary forces, whereas the Bo-number presents the ratio of gravity forces to capillary forces. The term presents the filtration velocity, is the density difference in the two fluids, is the fluid viscosity of the liquid, is the surface tension coefficient, is the gravitational acceleration and is the characteristic dimension of the pore size. Historically, the two-phase flow phenomena have extensively been studied in various fields through experimental and empirical analysis, due to the complex nature of the flow process. These studies are also limited as a result of their complexity, which has led to the implementation of numerical analysis as a tool for understanding these processes. 4.5 Flow in porous media The concept of porous media flow is widely encountered in various fields such as filtration theory, petroleum engineering, soil mechanics, etc. A porous medium is characterised by subdividing the total volume of the system into a solid matrix and a pore space. The solid matrix is constituted by an 40 interconnected pattern of solid materials, creating a dense structure. Such structures include e.g. a packed bed of solid grains, an entanglement of fibers and a connected pin-fin arrangement from a heat exchanger. Typically porous materials are, e.g. rocks, wood, paper, bones and ceramics, to name a few. These days, the flow in porous media has emerged as a separate field of study, due to its wide interest and application field. 4.5.1 Fluids and porous media as continua The concept of continuum mechanics is to consider a modelled material as continuous mass rather than discrete particles. By definition, when considering a matter as a continuum, the body can be continually subdivided into infinitesimal elements in which the properties of the body are the same as the bulk material. All materials are, however, composed of molecules which discretely connect atoms and, in turn, fail to meet the criteria of being continuous in space. Modelling fluid mechanics based on the Lagrangian perspective would result in describing fluid behaviour on a molecular scale in order to track the behaviour of each molecule, which would not be feasible due to the complex nature of the fluid. The fundamental assumption in continuum mechanics is to model materials on length scales far greater than the interatomic distances in order for the statement to be valid. Thus, we must diverge from the Lagrangian concept and involve microscopic scale effects in terms of averaging. The same concept is applicable for porous structures, as the aim is to make a transition from modelling constitutive variables in the pore space to a continuum of macroscopic field variables. In turn, the detailed structure of the pore geometry is eliminated. 4.5.2 Flow representative volume Numerical analysis of flow in porous materials in practical environmental systems is not feasible when a detailed description of the pore geometry in the scale of micrometres is the objective. Thus, when choosing a macroscopic scale, in which the details of the flow process are ignored, the average behaviour of the fluid in the porous medium is modelled. The macroscopic behaviour of the flow is determined by volume-averaging the constitutional variables of the mass and momentum equations over a so-called Flow Representative Volume (FRV). 41 The FRV is viewed as a sample of a porous medium in which the size of the FRV determines the level of variation in the solid matrix structure of the porous medium, upon which the volume-averaging is executed. In other literature regarding random particle generation with, e.g. statistical models on samples, they refer to these volumes as a representative elementary volume. A porous medium could be categorised as an ordered or disordered medium, see Fig. (10). Fig. (10) Examples of FRV-models representing a characteristic structure of the porous medium. In (a), the FRV is chosen for an ordered structure of a pin-fin arrangement in a heat exchanger; and in (b) the FRV represents a disordered arrangement of a fibrous structure representing a paper sheet. The structure of a classical pin-fin arrangement could be regarded as an ordered porous media due to a repeated identical structure in the streamwise and spanwise direction, see Fig. (10a). Choosing an appropriate FRV is to find the minimum repetitive volume in the sample. Depending on the size of the heat exchanger, the FRV should be extended in order to include the effect of entrance regions and re-circulation effects near the outlet. In Fig (10b), the FRV consists of a disordered porous medium where cylindrical elements have been randomly distributed over a sample volume, which in this case represents a paper sheet. The choice of FRV is based on different criteria than for an ordered structure such as the heat exchanger. Due to the irregular shape and displacement of solid particles in the sample volume, there is a distribution of porosities within the sample. Selecting a small size for 42 the FRV would increase the heterogeneity of volume fractions of each phase. Increasing the size of the FRV would, however, smooth out local heterogeneities as the scale of the sample far exceeds the particle scale. 4.5.3 Reynolds number In the theory of fluid mechanics, a key dimensionless quantity, which describes the unstable nature of the flow is called Reynolds (Re) number, named after the Irish engineer Osborne Reynolds. The Re-number represents the ratio of inertial forces to viscous forces, see Eq. (11). (11) The quantity is used to characterise different flow regimes within the same fluid and system. With dominating viscous forces, the flow process is laminar and with increasing inertial forces, the flow is trending towards a turbulent nature. In porous media, a modified Re-number is used to determine different flow regimes in which the velocity term and the characteristic length scale are modified to the particle size characteristics of the porous medium. In a system containing solid particles, the so-called interstitial velocity, which is the velocity in the pore-space, is the appropriate term. This term is related to the superficial velocity with the porosity of the medium, in order to preserve fluid continuity, see Eq. (12). (12) The term denotes superficial velocity, which is viewed as the velocity in the medium removed from its solid particles, and is the porosity of the medium which is the ratio of pore space volume (void) to the total volume, see Eq. (13) and Fig. (11). (13) 43 Fig. (11). Representation of flow through a column of unidirectional circular fibers. The characteristic length scale of a porous material with a packed bed of solid particles was deduced by the Austrian physicist Josef Kozeny, and it is the volume of the pore-space divided by the specific open area over which the fluid must flow, see Eq. (14). ( (14) ) The term represents the specific surface area, which in the case of circular fibers is the ratio 4/D. The characteristic length scale along with Eq. (12) is implemented in Eq. (11), in order to achieve the so-called particle Re-number. ( (15) ) The particle Re-number represents the ratio of inertial to viscous forces, just as in the original expression. This expression is viewed as a bed average quantity and will accurately represent a flow regime with fairly uniform flow patterns and geometric properties. Due to the complex configuration of the solid matrix, locally the flow will exhibit various conditions such as laminar with low inertia and high inertia and turbulence with unsteady flow patterns. However, with a large bed geometry compared to the particle size of the porous medium, the effect of these local heterogeneities should be reduced. This is further evaluated in chapter 5.2.3. 44 The threshold for inertial effects for the particle Re-number is extensively lower (Re > 10) in porous media than regular pipe flow systems [54-57]. There are other studies that claim that a weak inertia term exists in Re-numbers close to unity [58,59]. Strong inertia, which can be viewed as a transitional flow regime, exists in the range of Re-numbers at 10 to 200, which are then followed by an unstable turbulent flow. However, these critical values are still not clearly defined, due to difficulties in observing fluid behaviour. 4.5.4 Viscous and inertial contributions in porous media One of the first scientists to study fluid flow in a porous medium was the French engineer Henry Darcy. His famous Darcy’s law was used to solve problems such as water transport in an aquifer, or oil migration to a well. These processes of fluid transport are characterised by low flow velocity and a dense structure of pores in the material, i.e. low porosity. Darcy’s law is established based on experimental analysis at very low Re-numbers and resembles the NS equations excluded from the advective term, see Eq. (16). [ ( )] (16) The advective term representing inertial forces is disregarded, which yields a linearisation of the NS equations to Stokes flow, which is presented in steady state without body forces. Assuming that viscous resistance is linear to the velocity, the forces are characterised with the porous properties of the medium, i.e. the porosity and permeability. A homogenisation of the pore structure and the fluid into one single medium is a common approach when using Darcy’s law, in which the pressure gradient is the major driving force, Eq. (17). 〈 〉 (17) In this equation, is the permeability of the porous medium and it represents the resistance to flow of the medium. Depending on the isotropic nature of the porous material, the permeability could be defined as a second-order tensor, indicating anisotropic resistance or a scalar term indicating identical resistance spatially in the material. 45 [ ] (18) In the second case the indices are identical, which results in the summation of the diagonal term in Eq. (18), see Eq. (19). [ ] (19) Along with the pressure gradient, gravitation is a driving force of the flow.. A rigorous mathematical derivation of the Darcy’s law is presented in a theoretical description of how the permeability tensor is determined [60]. An extension of Darcy’s law was conducted in which the transitional flow between boundaries separating an open area to a porous area is taken into consideration [61]. Darcy’s law is extended with a term associated with drag forces experienced by the fluid flowing in the permeable region of the porous medium, see Eq. (20). 〈 〉 [ ( )] (20) In porous media, where Re-numbers exceed the value of creeping flow conditions (Re > 1), weak inertial effects should be considered as momentum losses are influenced by non-linear effects. The deviation from Darcy’s law has been observed, based on experimental analysis and through empiricism; researchers have extended Darcy’s law with correction terms including inertia in the flow process. The most notable formulation is the one presented by the Austrian engineer Philip Forchheimer in 1901, which today is known as the Forchheimer equation and is presented in Eq. (21). √ and (21) are vectors that denote the volume force (N/m3) and velocity field (m/s), respectively. The and coefficients represent viscous and inertial resistance, respectively. The high velocity inertial effects are included by the non-linear term representing the kinetic energy of the fluid. The viscous 46 resistance coefficient is related to a permeability second-order tensor in Eq. (18), for anisotropic porous structures. The inertial resistance coefficient has been the foundation of a research area, which is based on extensive experimental studies in terms of empirically determining the validity of the Forchheimer equation. In recent times, however, various theoretical approaches have been used to mathematically derive the equation from its first principles. A thorough presentation is given of how Darcy’s law, along with the Forchheimer correction terms are derived through a volume-averaging method [62]. Along with the VAT theory, homogenisation theories are used to derive Forchheimer’s equation [63,64]. Non-Darcy flow has been the target of many numerical studies as well in which the main focus was set on determining the inertial resistance coefficient in both isotropic and non-isotropic porous media in 2- and 3-dimensional systems [55,6570]. According to Wang et al. [71], the inertial term has a tensorial form if the porous structure is anisotropic. The Forchheimer equation is valid for low as well as high Re-numbers as was stated by Andrade et al. [66]. They observe, however that at a deduced friction factor from the Forchheimer equation overestimates flow resistance compared to experimental data in range of Renumbers 0.06 to 11. In terms of characterising inertia in porous media, researchers have proposed Forchheimer’s equation to empirically account for non-linear effects based on experimental and literature data [72-79]. In practice, the inertial resistance coefficient is usually determined by fitting the data in which the aim is to achieve correct correlations of experimental data for various systems. The most widely used empirical expression of the inertial term was put forth by Ergun [80] and is expressed as: (22) √ Here, is the dimensionless friction coefficient deduced empirically by Ergun and could be expressed as: ( (23) ) 47 The model expression established by Ergun satisfies linear and nonlinear flow regimes at a wide range of Re-numbers with reasonable accuracy. The Ergun equation has been the foundation of extensive experimental research on flow through packed beds, both as a validation tool and as a reference model for which modifications have been conducted in order to assess the validity of the model and, moreover, to improve the accuracy of the model. 48 5 Method The method described in this thesis is divided into three main categories: i. ii. iii. Modelling in COMSOL Volume forces coupled to single and two-phase flow Characterisation of flow resistance in various fibrous structures The first category, (i), is a brief introduction of the simulation program COMSOL. The second category, (ii), is a description of the methodology of establishing volume forces as sink terms in the momentum equation. The third category, (iii), is characterisation and assessment of the reliability of these volume forces through comparison of our numerical data to classical empirical formulations of the dimensionless friction coefficient, as well as current numerical results. The presented results in this thesis will involve categories (ii) and (iii), as the focus has been to present the reliability of the volume force terms and assess the benefit of implementing them in structureless domains regarded as porous media. 5.1 Modelling in COMSOL This chapter presents a short COMSOL MultiPhysics, in which handling numerical instabilities Additionally, the two-phase flow briefly introduced. 5.1.1 introduction to the simulation program the discretisation method, the generic PDE, and boundary conditions are addressed. model called the Level-Set (LS) method is Finite Element Method The discretisation method used in COMSOL is the FEM. The development of the FEM began in the middle of the 20th century, mostly for structural analysis in civil engineering. The method evolved as it gathered a strong mathematical foundation and has since been generalised to several fields of applied mathematics for numerical modelling, such as fluid mechanics. 49 The method subdivides an object into small finite-size elements. Each element is described by a number of DOF in which a set of characteristic equations is solved simultaneously. The finite element approximations of PDEs are found in the space of linear functionals. This space is called the finite element space. A brief introduction of simple problems is presented as we discuss polygonal and planar domains. The concept of the method is that a known surface is represented by an approximate surface: ( ) ( ) (24) For a one-dimensional problem, the curve is approximated with a linear ( ), then a interpolation. If and are the endpoints of an interval, number of points (nodes) are selected on the interval: (25) The piecewise linear interpolant ( ) is formed by connecting each node with a neighbouring node, see Fig. (12). Fig. (12) 10. A curve and its piecewise linear interpolant. Each node is associated with a basis function ( ). In numerical analysis, a basis function provides an interpolating function of a curve or surface in the function space, see Eq. (26). 10 Introduction to the Numerical Analysis of Incompressible Viscous Flows, pp. 18 50 ( ) ∑ ( ) ( ) (26) In two dimensions, the approximation of the function is on a surface and it is ( ), in which ( ) is defined on done by introducing a triangulation, each triangle. Fig. (13) 11. A finite element space formed as a hat-function. The triangle is a finite element space. For the finite element space, is a linear or higher order polynomial on each mesh element, with the value of 1 in node and 0 in all other nodes. A few basic conditions are required for creating a triangulation: the triangles are conforming, which indicate that the vertex of one triangle cannot be positioned on the edge of another triangle; the triangles are not extensively skewed; no triangle has all three vertices on a part of the boundary.[81] 5.1.2 Numerical instabilities During numerical calculations of e.g. a flow process, approximation errors generally occur. In cases when the error is amplified, the solution is referred to numerically instable. There are techniques applicable for handling numerical instabilities without having to refine the mesh structure further. In the case of a Galerkin finite element method (GFEM), which is used in COMSOL, an artificial diffusion 11 Introduction to the Numerical Analysis of Incompressible Viscous Flows, pp. 19 51 coefficient can be applied to COMSOL’s generic scalar convection-diffusion transport equation: ( ) (27) Eq. (27) is a general PDE in COMSOL, which describes transport phenomena such as heat, mass and momentum transfer. In the case of momentum transfer, represents the advective velocity vector; represents the viscosity in this case; is the transported velocity vector and F is a source term. When discretising Eq. (27), the Péclet (Pe) number indicates instability when the value surpasses one. ‖ ‖ (28) The dimensionless quantity Pe was deduced by the French physicist Jean Claude Eugène Péclet during the first half of the 19th century. The quantity is defined as the ratio of the rate of advection of a quantity due to the influence of a fan or pump, to the rate of diffusion of the same quantity due to gradients. In this case, Eq. (28) indicates that it is preferable that the convective term has a low value and/or that the size of the mesh cell, ( ), is small in order to keep Pe < 1. Generally speaking, simulation of flow with high velocity requires a dense mesh structure. Moreover, when treating fluids with a high viscosity, such as oil, it damps the effect of oscillations due to the fact that the coefficient is in the denominator of Eq. (28). The artificial diffusion is implemented by adding a term to in Eq. (27). ‖ ‖ (29) The tuning parameter transport equation is: (( has its default value at 0.5. Hence, the modified ) ) , (30) in which the new Pe number is: ‖ ‖ (31) ‖ ‖ 52 The new Pe number will not exceed one if ‖ ‖ approaches infinity. The tuning parameter, however, should be set as low as possible, given the fact that the new modified transport equation could present solutions that deviate to far from the original problem.12 5.1.3 Boundary conditions In COMSOL, there are several predefined boundary conditions for fluid flow, heat and mass transport. These conditions are applied by the user, depending on the physical conditions surrounding the computational domain. The boundaries are characterised as exterior, if they encompass the computational domain. The interior boundary condition is a dividing interface between two subdomains for the entire computational model, see Fig. (14). Fig. (14). Representation of an exterior and interior boundary. For interior boundary conditions, COMSOL ensures continuity of fluxes across the interfaces in most cases. In CFD, there are two basic types of boundary conditions: The Dirichlet boundary condition The Neumann boundary condition 12 COMSOL Multiphysics documentation 53 Classical boundary conditions such as the no-slip and constant pressure are characterised as Dirichlet conditions, indicating that the constitutional variables are fixed at the boundary. { (32) The general heat flux condition is a typical Neumann condition, as the temperature on the wall is not fixed. ( ) (33) Even though a constant heat flux is defined at the boundary, the absolute value of the temperature may vary spatially. In COMSOL, the generic notations of the boundary types are: (34) ( ) (35) The coefficient represents a physical quantity such as temperature, concentration and/or velocity. Eq. (34) represents the Dirichlet condition, which specifies a constant value on the boundary of the domain. Hence, the value is applied at the boundary. The Neumann condition specifies the normal derivative of the quantity at the boundary, see Eq. (35). The term , of Eq. (35), represents the diffusive term, is the convective term, and represents a source term. The weight coefficient is an -matrix, where is the number of dependant variables.13 5.1.4 Level-Set method Along with solving the NS equations in flow process, a suitable tracking method is required in order to accurately determine the location of the interface separating two immiscible fluids. There are several techniques which are used to 13 COMSOL Multiphysics documentation 54 trace the interface of two-phase flows. The two most common techniques are based on the Eulerian and Lagrangian approaches. The Lagrangian method is suitable for “simple” flows, as the interface surface is explicitly tracked by marking the interface with particles used for tracking. In complex flow processes, many tracking particles are required to adequately describe the shape of the interface, which increase the probability of numerical convergence issues. Therefore, the method used in this study is based on the Eulerian approach. The Eulerian based LS method is applied, which numerically tracks the interface in a stationary Eulerian grid. The grid of the interface is, however, separately configured and moves along the interface during the simulation process. The LS function is used to locate, in this case the air–water interface, in which the function takes a positive or negative value depending on which side of the interface is viewed: (⃗ ) ⃗ ⃗ ⃗ { (36) where ( ⃗ ) is a scalar quantity named the LS function, and is the interface boundary. The LS function is assigned to every grid point in the domain, thus determining whether the scalar function is in the liquid or gas phase, see Fig. (15). Fig. (15). Schematic illustration of a curve representing the interface between two fluid domains. The interface motion is defined by an advection equation in a Eulerian representation , (37) 55 where the first term in this PDE is the local time derivative of the LS function, and the second term represents the advection of the interface. The LS function is a smooth function and is assigned as a distance function, according to Eq. (38). | ( ⃗ )| (| ⃑ ⃑ |) for all ⃑ , (38) where ( ⃗ ) when ⃑ is equals the space coordinate at the interface boundary ( ⃑ ). The signed distance function is approximated numerically, in order to achieve smoothness of the solution. This is achieved by a first-order accurate approximation using a Heaviside function: ( ( ⃗ )) { ( ⃗) ( ( ⃗) ( ⃗) ( ⃗) ( ⃗) ) (39) where is a parameter that determines the thickness of the numerically smeared interface. As Eq. (37) is solved during a simulation process, the motion of the fluid–fluid interface changes the LS function, and causes the regulated interface thickness to increase from a designated signed distance function. This leads to inaccurate solutions of local properties at the interface. The achievement of a constant interface thickness minimises the numerical error, due to smearing from diffusive effects. Consequently, Eq. (37) is extended with a re-initialisation term: ( ( ) | | ) (40) Here, is the re-initialisation parameter and equals ( ( ⃗ )). Both the density and the viscosity are functions of the LS function, which spatially smooth the variation of the material properties along the interface thickness. The interface boundary moves with a propagation velocity normal to the surface, see Fig. (15). This velocity is a function of the fluid velocity, the curvature and the normal direction of the boundary. It is therefore, important to accurately model the advancement of the interface boundary. The normal 56 vector and the curvature at the air/water interface are determined with the LS function. | | (41) Here, the spatial derivative points in the direction of an increasing value of the LS function. The curvature is calculated from the divergence of the normal unit vector. ( | | ) (42) Here, the parameter is positive for convex regions and negative for concave regions. The curvature term is important in numerical simulations of two-phase flows. When the interface boundary is propagating at a constant speed, it may yield singularities at several locations of the boundary. In these regions, the LS function is not differentiable and the evolution of the boundary is not clear. The curvature term is used to smooth and reshape the moving front of the boundary, preventing such occurrences. 5.2 Volume forces coupled to single and two-phase flow The aim of this chapter is to present a method to establish time efficient simulations. The concept is to reduce the geometrical complexity of systems by replacing the internal structure with sink terms in the momentum equation. The chapter consists of: addressing various types porous structures, e.g. a heat exchanger and a paper sheet; modelling surface tension; assessment of box size, mesh configuration and sensitivity of the amount of numerical sample points; estimation of Forchheimer coefficients from numerical data set; establishing and implementing volume forces in structureless models. 5.2.1 Pore-scale simulations Both a dense structured heat exchanger component and a paper sheet fall under the category of a porous medium. However, there are several characteristics of 57 each component, which set them apart. A classic fin-tube heat exchanger is ordered in its arrangement and its solid matrix has a repetitive appearance in the system; whereas the paper is composed by irregular shapes of fibers which are spatially disordered in the system, Figs. (16), (17) and (18). Fig. (16) and (17). A cross section of a tube-fin heat exchanger and a closer look at the arrangement of fins and tube bundles. Fig. (18). Microscopic view of aligned fibers with various shapes. 58 As it was stated in chapter 4.5.1, it is not feasible to model constitutive variables at a pore-scale level if the entire system is to be considered. Creating a mesh grid for a geometrically compact system generally results in a dense grid arrangement which requires extensive computer power. The purpose of volume force implementation in the momentum equation is to bypass a dense mesh grid arrangement by removing the internal structure. The method of implementing volume forces in COMSOL is similar to the VAT theory presented in chapter 2.3. An appropriate FRV is established in the system in which a direct numerical simulation of the NS equations is conducted at the pore-scale level. The COMSOL notation of the NS equations is symbolic and includes a body force term , see Eq. (43). ( ) [ ( ( ) )] (43) In flow processes dominated by the pressure gradient, the term is generally disregarded. However, in the case of modelling the flow process in a dense structure, the parameter is included as a sink term in the momentum equation to account for the resistance force due to the presence of the porous medium. To put it in perspective with the VAT theory, a closure model is used to determine in the momentum equation. 5.2.2 Modelling surface tension In two-phase models, a key parameter needs to be considered in the momentum equation, which is the contribution of capillary forces. In COMSOL, surface tension forces are defined according to Eq. (44). ( ( ) ) (44) Here, is the Dirac delta function, is the unit matrix, and is a surface tension coefficient. When a system involves a moving fluid-fluid interface, the conditions concerning the solid boundaries are treated by implementing artificial friction forces in order to prevent numerical difficulties when the fluidfluid interface stagnates at the solid boundary, due to the no-slip condition. The wetted wall function in COMSOL is applicable at the solid walls to allow for 59 capillary forces and also to provide greater numerical stability than the no-slip condition. , . (45) Eq. (45) indicates a slip condition at the solid walls. In addition, this condition adds a frictional force, where is the slip length. It is the length in which the extrapolated tangential velocity is set to zero at the distance from the surface, see Fig. (19). The slip length is equal to the local mesh element size. 14 Fig. (19) . Representation of the slip length in the wetted wall function. 5.2.3 Closure model As mentioned in chapter 4.5.4, macroscopically characterising flow resistance in porous media, including inertial effects, proposes Forchheimer’s equation, see Eq. (21). The constitutional variables of the momentum equation are calculated in the FRV of the system and are volume- averaged over the pore-scale. 〈 〉 ∫ (46) The intrinsic quantity of the variable is related to the superficial quantity through the porosity of the FRV. 〈 〉 〈 〉 (47) 14 COMSOL Multiphysics documentation 60 A momentum flow balance over the FRV is applied based on the volumeaveraged quantities. The balance considers the applied pressure force on the boundaries and momentum transport across the boundaries. Shear stresses at the macroscopic level are generally disregarded at the boundaries of the computational domain. This was depicted in Eq. (4) and Fig. (8), in chapter 4.1. The reaction force is volume-averaged over the pore space of the FRV, see Eq. (48). 〈 〉 (48) The term is related to the volume-averaging velocity over the pore space of the FRV through Forchheimer’s equation, according to Eq. (49). 〈 〉 〈 〉 〈 √ 〉 (49) The second-order tensor elements are determined through fitting numerical data with the least square method. A data set is established by conducting single-phase simulations in the FRV for various Re-numbers. The interval of Re-levels included in the data set considers all flow regimes encountered in the system. The volume force terms ( ) are established based on the Forchheimer coefficients including viscous and inertial flow resistance in the flow regime. The resistance coefficients are viewed as 3x3 matrices, if the porous medium has an anisotropic structure, see Eq. (50). The order of these tensors could be reduced to zero if the solid matrix of the porous medium resembles an isotropic arrangement. Moreover, the number of unknown coefficients could be reduced based on the notion of the isotropic behaviour of the fluid, which states that the non-diagonal elements with the same yet reversed index in Eq. (51) are identical. [ ] [ ][ ] [ ][ ] (50) The left index of and represents the normal direction of the plane in which the vector is located. The right index represents the direction of the vector 61 quantity. Hence, the direction and the magnitude of the velocity field will determine which resistance coefficients are active in the control volume. The least square method was used to approximate the resistance coefficients in which the residual is defined as the difference between the actual value, which is the numerical data of the volume force, and a predicted value described with the Forchheimer expression. The optimum fitting is achieved when the summation of the squared residuals are at a minimum. A system of equations is attained based on using the least square method in Eq. (51). ∑[ ] ∑[ ][ ] (51) The equation system is presented for a case in which the velocity and the force term are acting in one direction. As a result of a disordered porous medium in which velocity is acting in all directions, Eq. (51) extends as the number of unknowns in the equation system increase from 1 to 9 for each resistance coefficient in the Forchheimer equation. ∑ ∑ [ ] [ ][ ] (52) In Eq. (52), the viscous and inertial resistance coefficients have been estimated in relation to the force component in the x-direction. The same procedure has been executed on the remaining force components. Moreover, the number of unknown coefficients could be reduced based on the notion of the isotropic behaviour of the fluid. This would reduce the total amount of unknowns from 18 to 12 coefficients. 5.2.4 Sensitivity studies The sensitivity of the pore-scale simulations has to be considered based on three factors, which are the size of the sample domain, the generated mesh grid 62 and the sensitivity of the resistance coefficients based on amount of numerical sample points used. Choosing an appropriate sample volume is essential in volume-averaging processes of porous media, as the objective is to create volume-averaged quantities that are independent of the local heterogeneities [82]. The Brinkman screening length criterion is used to analyse the sample size, see Eq. (53). (53) √ The criterion is determined based on the square root of the permeability and is associated with stokes flow dynamics. In order to strengthen the choice of box size, symmetry conditions of the original sample volume are extended to the outer boundaries of a larger sample, see Fig. (20). The flow characteristics are determined in the FRV by direct numerical simulations of the NS equations. The resolution of the boundary layer is therefore critical which demands a sensitivity study of the mesh grid. It is desirable to establish a model independent of further mesh refinement. It indicates that for the current boundary conditions, velocity gradients and pressure of the fluid are calculated with a higher accuracy. An unstructured mesh grid containing tetrahedral elements was created for the 3-dimensional model presented, according to Fig. (20). 63 Fig. (20). Mesh configuration of a disordered fibrous sample domain. The grid is solved in a finite element space where a set of basis functions are used to create piecewise linear relations between the mesh elements and convert them to weak formulations. The mesh parameters that were tested are the maximum mesh element sizes, the minimum element sizes, the resolution of narrow regions and the element growth rate, see Table (1). Table (1). Grid element parameters tested of a 3-dimensional disordered fibrous structure Minimum element size [µm] 40.2 Element growth rate Resolution of narrow regions Extra coarse Maximum element size [µm] 190 1.40 0.3 Coarser (standard) 115 28.7 1.30 0.4 Coarse Normal 74.7 57.5 23.0 17.2 1.25 1.20 0.5 0.6 The mass flux of the porous structure were compared for all of the mesh cases in order to assess the sensitivity of the simulation at the pore-scale level. As was stated, the Forchheimer resistance coefficients are estimated from a numerical data set which includes flow regimes at various Re-levels. The 64 sensitivity study is conducted by varying the amount of numerical data points used to approximate these coefficients. The study provides an insight of the dependency of the resistance coefficients based on estimating over a certain range of Re-levels. 5.2.5 Volume forces in structureless models In the up-scaled structureless model, the volume forces in terms of the Forchheimer coefficients presented in the previous chapter are implemented as sink terms in the momentum equation, depicted as in Eq. (43). The sink term is expressed component wise which is presented in Eqs. (54) and (55) for a 2dimensional space. [( ) ( ) ] (54) [( ) ( ) ] (55) These expressions are coupled to the local velocity field of the porous medium. A negative sign is employed on each Forchheimer expression in order for the force component to act in the opposite direction of the local velocity field, see Fig. (21). Fig. (21). Schematic description of the sink term in a 2-dimensional domain in which the magnitude and the direction of the volume force terms are related to the product of the local velocity and the flow resistance coefficients. 65 The ratio of the inlet/outlet surface area and the volume flow rate changes due to the removal of the internal structure. The constraint in the volume force model is that the relation between the volume flow rate and the pressure force is equal to the relation in the FRV-model. Hence, the resistance coefficients are scaled with the porosity of the medium as a result of the local velocity field occurring in a structureless medium. The viscous contribution is scaled with the porosity and the inertial contribution is scaled with the second power of the porosity, due to the product of two velocity components. With the elimination of the no-slip condition, the advective and viscous stress terms in the NS equations are eliminated and replaced with the expression from Eq. (49). 〈 〉 〈 〉〈√ 〉 (56) Eq. (56) is similar to the momentum transport equation presented in Horvat and Catton [22,23], see Eq. (57). ( ) (57) They state that the pressure force is balanced by shear forces and an empirical correlation of a local drag coefficient, which represents inertial force as, in their case, flow resistance across the tubes. The volume force expression in Eq. (56) constitutes these forces, as we stated that and represent viscous and inertial forces, respectively. In the case of implementing volume forces in a two-phase flow system, spatial variation of fluid properties needs to be considered, due to the existence of two fluids in the same system. The resistance coefficients in these equations vary, depending on the type of the fluid. Eq. (58) is therefore implemented in COMSOL to regulate the coefficients at each phase over the interface thickness. { ( ( ) ) (58) 66 Here is the LS coefficient, which represents the interface thickness and varies from zero to one, see Eq. (39). As the internal solid matrix is removed in the volume force model representing the two-phase flow, capillary forces are excluded. Depending on the heterogeneities of the pore size, local capillary forces could dominate the displacement of fluids. This is illustrated in Fig. (22). Fig. (22). Contour plot of the air/water interface during a dewatering process represented by a pore-scale (a) and volume force (b) model. The purpose of the volume force model is to achieve accurate relation between bulk properties such as flow rate to the system pressure drop. Removal of the internal structure of the system will vastly reduce the computational time as the resolution of boundary layers are not the objective of such simulations. However, any conclusions of local discrepancies of the flow process should carefully be discussed. The flow rate and pressure drop relation for volume force models representing a heat exchanger and paper sheet are compared to their respective FRV-models. This comparison will validate if the transformation of pore-scale to porous simulations are possible. Moreover, the computational times for these models are compared in order to assess the benefit of such models. There are cases in which the volume force model could be implemented in systems regarded as porous media, however, with larger length scales. The contribution of the volume force model is the fact that the computational 67 domain can be extended, as a result of the reduced computational effect, to solve the subdomains with implemented volume forces. 5.3 Characterisation of flow resistance in various fibrous structures This chapter presents the procedure of: generating extensive numerical data for, in this case, 3-dimensional fibrous media; characterising the Forchheimer based flow resistance coefficients as a non-dimensional friction factor; validating the friction factor to classical empirical formulations. 5.3.1 Algorithm for production runs in COMSOL In the case of repeating simulations with similar build-ups, an approach to achieve a more efficient overall process is used to automatise the simulation process from defining the sample volume to exportation of numerical data to a post processing program, see Fig. (23). 68 Number of sample runs Control of successful CAD and mesh Fiber sampling 1 Invert internal structure and mesh Reset parameters if error occurs in CAD or mesh sequence Initial simulation of the computational domain Implementation of numerical diffusion if convergence issues arise 2 End session when minimal diffusion is reached Setup for simulation of several Re-levels Execute simulation of current Re-level Implementation of numerical diffusion if convergence issues arise 3 Calculate volume- and surface integrals for current Re-level End session when all Relevels are simulated 4 Export numerical data as text file for further post processing End procedure Fig. (23). Main procedure of the algorithm. These procedures are executed sequentially in a MATLAB algorithm coupled to the commercial CFD code COMSOL Multiphysics. In the process of characterising flow resistance numerically in various fibrous structures, an algorithm is built up by four general sections. The first section loops a sequence in which the structure of the domain, boundary conditions and mesh grid generation are applied until each step is executed successfully. The second section initiates the simulation process at the lowest Re-level. Implementation 69 of numerical diffusion is activated if convergence issues arise. The third section is a setup of a loop, executing several simulations at various Re-levels. For each Re-level, volume and surface integrals of the computational domain are calculated. After all Re-level simulations are executed, the numerical data set is exported as text files for further post processing. In the fiber sampling blocks in the first section in Fig. (23), an algorithm was used to generate and uniformly distribute pseudorandom integers for which the position and orientation for the fibers are determined in the sample volume. The spatial coordinates and orientation of the fibers are based on three uniformly distributed variables, respectively, which determine the position of each fiber and the level of orientation along their respective axis. The algorithm is able to determine the anisotropic nature of the fibrous structure based on a few settings, see Fig. (24). Fig. (24). Various fibrous arrangement in a 3-dimensional sample volume in which (a) presents an aligned arrangement, (b) presents a layered arrangement in the x-y plane, (c) presents an isotropic arrangement. The Brinkman screening length criterion is used to analyse the sample size, see Eq. (53). According to Clague and Phillips [83], each side of the sample volume should exceed Brinkman’s screening length criterion times 14 in order to smooth out local heterogeneities. The length scale is equivalent to the square root of the permeability. This criterion is vastly used in permeability studies in 3-dimensional fibrous structures [84-86]. The permeability is calculated based on Stokes’ flow simulations on a considered sample volume. Based on the fiber diameter and a porosity of approximately 0.7, a sample volume with the dimension 400x400x400 µm is large enough to meet the Brinkman criterion. 70 5.3.2 Assessing flow resistance based on Forchheimer coefficients The approach chosen to quantify flow resistance in these studies is based on the work of Ahmed and Sunada [87], in which they predict a pressure drop and flow rate relation with empirical and quasi-empirical correlations. In Macdonald et al. [78], they state that the porous medium is complex to the extent that the Forchheimer coefficients ought to be functions of the porous medium, rather than universal constants. The Forchheimer coefficients are estimated in FRV’s representing the structure of the porous medium, and hence, related numerically to the porous medium. A dimensionless friction factor is utilised to assess flow resistance in these porous mediums. The friction factor is related to the Forchheimer coefficients through a modified Reynolds expression, see Eq. (59). The friction factor is deduced by expressing the Forchheimer equation in a non-dimensional form which presents an expression in correlation with a modified Re-number. (59) This equation has been used extensively to correlate experimental data for various flow conditions and porous materials [77]. Based on this expression, the modified Re-number is related to the viscous and inertial term of the Forchheimer equation (60) These coefficients are not universal, as they have shown to vary with changes in the porous structure and flow process. The inertial term could be attained by curve fitting data sets composed by numerical experiments. In that sense, choosing an appropriate FRV for volume-averaging of velocity and pressure for various Re-numbers is essential in determining the viscous and inertial term. Many papers have contributed to establishing valid empirical correlations of the friction factor and particle Re-number for various cases of packed beds [78,80,8897]. A few expressions are used to compare the numerical simulations conducted in this work. 71 ( ) ( ) ( ( ) ( ) (Ergun) (61) ) (Montillet) (63) (Tamayol) (64) ( ) ( ) ⁄ √ (a, b and c are fitting parameters) These empirical correlations are used to compare the friction factor with Eq. (59) for an aligned disordered distribution of cylindrical fibers in a 3dimensional sample domain. 72 6 Results and Discussion The focus of this chapter is to assess the proposed methodology in establishing volume forces as sink terms in the momentum equation in order to conduct time efficient simulations in models regarded as porous media. These bullet points are the foundation of this chapter: Outcome of the fitting procedure. Comparison of FRV and volume force models in terms of relating pressure and flow rates. Sensitivity study of box size and mesh configurations. Experimental validation of simulation models. Assessing flow resistance by comparing numerical data with empirical formulations and other simulation data. Assessing the benefits of implementing time efficient simulations in a numerical based design process. 6.1 Fitting procedure of resistance coefficients: Paper II, III, IV, V The approximated resistance coefficients concluding the Forchheimer equation should be validated in order to justify the fitting procedure. The numerical data set is used to compare the volume-averaged force term to the force term constituted by the Forchheimer coefficients. Fig. (25) represents the fitting result of the force term established for a pin-fin heat exchanger with a R2-value of 0.999. 73 Fig. (25). Fitting of the numerical data set to the Forchheimer expression in Eq. (49) for the 2 pin-fin heat exchanger. The R -value is approximately 0.999. The comparison made in Fig. (25) is for an ordered structure in which open boundaries exist in one direction. Hence, the force term acting in that direction is equal to the balance of momentum, shear and pressure forces in the direction of the open boundary. In a disordered structure, see Fig. (20), there are open boundaries in all directions in space resulting in the activation of three force terms. Fig. (26). Fitting of the numerical data set (blue circles) to the Forchheimer expression 2 (green dashed line) in Eq. (49) for the paper sheet. The R -value is 0.999, 0.899 and 0.997 for the x, y- and z-direction respectively. 74 The fitting procedure is successful as presented in Fig. (26). The R2-value based on comparison of Forchheimer expression of the volume force terms to numerical data is 0.999, 0.899 and 0.997 for the x, y- and z-direction for the paper sheet. 6.1.1 Concluding remarks Overall, the fitting procedure was successfully executed based on the low values of the estimated error. Moreover, further analysis indicates that increasing numerical sample points reduces the estimated error. 6.2 Implementation of volume forces as sink terms: Paper II, III, IV The next step of analysing the validity of the Forchheimer based volume force terms is implementing them in a structureless domain in the CFD code COMSOL MultiPhysics. The coefficients are coupled to the local velocity field according to Eqs. (54) and (55), in which a sink term is established in the momentum equation. These volume force models are calculated in the same pressure drop range as their respective FRV-models. In the case of analysing the heat exchanger, two FRV-models are compared in which FRV1 considers a small segment of the heat exchanger with periodic condition at the open boundaries, Fig. (10a). FRV2 considers the actual streamwise length of the heat exchanger. The comparisons of these models are presented in Fig. (27). 75 Fig. (27). Comparison of flow rate and pressure drop relation for a structureless model with volume forces to its respective FRV-model. The mean deviation for the FRV and VF relations is approximately 4% and 1% respectively. The volume force models representing the heat exchanger presents a fair agreement over the range of pressure levels, as the mean deviation is roughly 4% for FRV1 - VF1, and 1% for FRV2 - VF2. Table (2) presents the respective Re-number for the FRV-model. Table (2). Variation of flow rate and Re-number for a selected set of simulations. VVX (air) 1. 2. 3. 4. 5. ∆p [Pa] 35 40 45 50 55 ̇ [m3/h] 220.0 238.1 255.2 271.4 286.9 Re [-] 329 352 373 394 414 The volume force model representing the paper sheet is exposed to higher vacuum pressure levels in which the estimation of the Forchheimer coefficients are calculated for a broader range of Re-numbers. Therefore, the relation of the mass flux to the pressure levels exhibits a nonlinear relationship compared to Fig. (27). 76 Fig. (28). Comparison of mass flux and pressure drop relation for a structureless model with volume forces to its respective FRV-model. The mean deviation for this case is approximately 2.9%. The volume force model presents a fair agreement over the range of pressure levels as the mean deviation is roughly 2.9%. The Re-numbers for the spectrum of vacuum pressure levels are presented in Table (3). Table (3). Variation of water mass flux of and Re-number for a selected set of simulations. Paper (water) 1. 2. 3. 4. 5. ∆p [kPa] 1 20 40 60 80 ̇ [kg/m2s] 93 687 1005 1249 1457 Re [-] 2.64 19.6 28.7 35.6 41.5 The deviation between the FRV and volume force model is more pronounced at higher vacuum levels. This deviation could indicate that a second order fitting procure of the coefficients is not suitable at the Re-levels presented in Table (2). Overall, sink terms in the volume force models provides a good prediction. 77 In the case of implementing volume forces in a two-phase flow model, which in this case is a vacuum dewatering process, the influence of capillary forces are excluded, as presented in Fig. (22). The vacuum dewatering process was modelled with the LS-method based on a 2-dimensional pore-scale simulation with volume forces representing flow resistance in the forming wire, see Fig. (29a). The other approach is a 2-dimensional volume force model, see Fig. (29bc) with sink terms representing flow resistance in a 3-dimensional fibrous space representing the paper sheet, see Fig. (24b). Moreover, the concentration of fibers packed at the top section of the forming wire is described by implementing volume forces based on the same Forchheimer coefficients in the lower section of the paper sheet, see Fig. (29c). These sink terms are, however, regulated with a time constant which states that there are no volume forces present at the start of the process, and they gradually increase to a maximum value which represents a final packing state. A simplified representation of the forming wire is presented in Fig. (29c) in order to analyse the influence of physically reducing the open area. These models simulate dewatering of a sheet with a basis weight of 50 g/m2 and all models has a porosity of approximately 0.72 and a basis weight at 50 g/m2. 2 Fig. (29). Representation of a paper sheet with a basis weight at 50 g/m . Three models are presented in which (a) a pore-scale model with volume forces is representing flow 78 resistance in the wire. Model (b) presents flow resistance in the paper sheet and the wire with volume forces. Model (c) presents flow resistance in the paper sheet with volume forces and flow resistance representing fiber packing at the top layer of a simplified wire structure. Fig. (30). Dry content and dwell time relations for three simulation models (Fig. (29)) of a 2 paper sheet with basis weight 50 g/m . Based on the cases presented in Fig. (30), it seems that implementing volume forces at the top region of the simplified wire representation (red dashed line) presents the highest flow resistance as the dry content is not increasing at the same rate. Some interesting remarks on this distinction are that volume forces in model (a) and (b) in Fig. (29) seems to affect the dewatering process slightly. In model (a), volume forces are determined based on a trivial representation of the forming wire in 3-dimensions. Numerical results from models in Fig. (29ab) indicate that flow resistance due to the wire is neglectable indicating that averaging over a coarse solid arrangement is pointless in case of denser regions in the system controlling the permeability. In model (b), neither volume forces representing the paper sheet and the wire seems to influence the dewatering. In model (c) however, the system reacts on reducing the open area which forces the fluid to spatially accelerate which increases the volume forces according to Eqs. (53) and (54). The influence of excluding capillary forces is visibly illustrated by analysing the dewatering rate in relation to the dwell time, see Fig. (31). The models in comparison are from Fig. (29ac). The first distinction made between the models 79 is that the pore-scale model exhibits various peaks of the dewatering rate whereas the volume force model presents a single peak at the start of the process. The first peak in the pore-scale model represents the time for air penetration as the dewatering radically decreases afterwards. The other following peaks present infrequent water release from dense regions which capillary forces dictate. The peak in the volume force model represents the effect of fibers packing at the top layer of the forming wire. The volume force model presents a smooth dewatering rate. Fig. (31). Comparison of the dewatering rate between the pore-scale and volume force 2 model. These models presents dewatering in a paper sheet with a basis weight at 50 g/m . The computational time of the volume force models is significantly shortened as the geometrical complexity of the simulation domain is reduced, along with the amount of DOF. Table (4) presents a list in which the computational time of simulation models based on volume forces as sink terms are compared to pore-scale (FRV) models of the same system. The comparison is made for both 80 single and two-phase flow processes, steady state and transient systems, as well as 2 and 3-dimensional models. Table (4). List of computational time for pore-scale and volume force models FRV-model CPU time DOF Heat exchanger 58.82 min 2.8*106 Paper single-phase 13.38 min 9.3*105 Paper two-phase 846.0 min 7.4*104 Volume force model CPU time DOF 0.25 min 4.3*104 0.53 min 2.5*104 23.0 min 7.3*103 6.2.1 Concluding remarks In single-phase simulations, volume forces are successfully implemented based on achieving acceptable flow rate and pressure drop relations in comparison to their respective FRV-model. In two-phase flow simulations, capillary forces are excluded along with the removal of the internal structure which affects the dynamics of the dewatering process. The dry content and dwell time relations could still be valid in terms of predicating the bulk properties of the system. However, the details of the dynamics are eliminated in the process. The computational time for the volume force based models were reduced significantly, as presented in Table (4). This provides time efficient simulations during pressure drop and volume flow rate predictions in the case of the heat exchanger and prediction of dry content and dwell time relation based on system parameters in a vacuum dewatering process. Certainly, it should be possible to predict such relations in up-scaled systems with the same porous configurations. Another concluding remark on time efficient simulation is the implementation of the volume force model in larger computational domains as a result of reducing DOF of the entire system. For instance, the internal duct system presented in Fig. (3). As it was stated before, the details of the flow process is eliminated which should be considered when implementing a volume force model in a lager system. The reason is that subsystems following a volume force model could be miscalculated due to the homogenous flow patterns entering the system. For instance, dealing with a system with high Re-numbers, turbulence is an essential parameter. As a result of smoothing out heterogeneities, the build-up of 81 turbulence is mistreated which affects key inlet parameters such as turbulence intensity and temperature distribution. 6.3 Sensitivity study of pore-scale simulations: Paper II, III, IV, V Several aspects of conducted sensitivity studies should be considered in a modelling process heavily based on numerical analysis of constitutional laws of physics. Besides experimental validation, which is a requirement in numerical analysis, sensitivity studies of simulations models and processes should be analysed further. The sensitivity of the mesh grid is presented as the estimated Forchheimer coefficients in the flow directions are presented for cases in Table (1). The variation of the mass flux is presented in Fig. (32). The mass flux is defined based on the volume-averaged flow velocity. Fig. (32). Mass flux and vacuum level relation for various mesh grid configurations. The mass flux based on the volume-averaged z-component velocity increases as the grid is more refined. At the highest vacuum levels, the mass flux increases with approximately 19%, as the grid is refined to the settings labelled as “coarser”. The next refining steps yield 8% and 5%, respectively. The resistance coefficients acting in the flow direction are chosen as parameters for validation; see the coefficients with a double index of z-components in Eq. 82 (50). At low sample points, the coefficients fluctuate considerably and unpredictably, indicating that the set of data points is sparse. As the amount of sample points are increased, the value of the resistance terms seems to stretch towards a certain value with no fluctuations along the increasing sample points, see Fig. (33). Fig. (33). Variation of estimated viscous and inertial resistance coefficients based on the amount of numerical sample points. The maximum Re-number increases with the amount of sample points. The friction factor is based on the Forchheimer coefficients, which are averaged over a set of data points. The modified friction factor and Re-number relation are examined for a range of sample points which is presented in Fig. (34). It is obvious that the friction factor is depending on the range and frequency of Re-numbers indicating that it is sensitive to the amount of sample points. The factor is converging towards a fixed value as the deviation of the curves in Fig. (34) in decreases along with the higher sample points. The sensitivity is visible at low Re-numbers as the friction factor for all sampling cases seems to increase towards the same value. 83 Fig. (34). Modified friction factor and Re-number relation based on the amount of sample points. The sample points presented in Figs. (33) and (34) are equivalent to Renumbers ranging from 0.002 to 1000. 6.3.1 Concluding remarks A few concluding remarks are that the sensitivity analysis of the box size and mesh grid should be conducted early in order to assess the robustness of the representative structure as well as the resolution of the flow field. Increasing the box size and grid resolution would yield a more robust FRV-model, as well as increase computation time. Therefore, refinements should be made wherein both these requirements are met. It is obvious that the friction factor is sensitive to the amount of numerical sample points used in the estimation of the Forchheimer coefficients. As sample points are increasing, the Forchheimer coefficients stabilises towards a constant value. The variation is still hard to predict as analysis has indicated that the Forchheimer coefficients are dependent on the porosity and level of orientation in the fibrous structure. 84 6.4 Experimental validation of simulation models: Paper II, III, IV Volume forces presented as sink terms in the NS equations were implemented in the heat exchanger unit in the absence of the internal pin-fin configuration. A comparison of the magnitude of the volume flow rate and pressure drop is presented in Fig. (35). Fig. (35). A comparison of the pressure drop and volume flow rate relation of the volume force models to experimental data. The volume force model based on FRV1 significantly deviates from the experimental data due to the absence of inlet regions of each heat exchanger, which in turn excludes the high viscous shear stresses as the boundary layers are developed along the fins. The volume force model presented in Fig. (29c) is further evaluated by comparing the dewatering rate with experimental data from Rezk et al. [98] and experimental data from Pujara et al. [44]. 85 Fig. (36). Dewatering rate and dwell time relations at three different vacuum levels for the volume force model to experimental data from Rezk et al. (2013) and experimental data from Pujara et al. (2008b). Comparison is made for paper sheet with a basis weight of 50 2 g/m . The dynamics of the dewatering rate in the experimental setups are still hard to comprehend, due to the sample frequency. However, experimental data indicates that the rate of packing at the contact surface of the wire plays an essential role in the process, as the dewatering rate decreases with the dwell time. The sudden dip of the dewatering rate for the volume force models in Fig. (36) is a result of air penetrating the paper sheet. The models indicate that reducing the vacuum level delays the time for air penetration. There is a slight offset between the first sample points and the peaks of the curves which could be a result of the time delay of sampling. The offset is visible, due to the fact of the short time span of the dewatering process. 6.4.1 Concluding remarks Comparison to experimental data indicates that the proposed volume force models predict the flow process convincingly. (i) The volume force model for the heat exchanger unit based on averaging over the FRV2-model shows a fair agreement, as the flow rate and pressure drop relation differed by approximately 8% from the experimental data. (ii) The simulated values for the dewatering rate as a function of time showed a fair agreement with the experimental data, as the data shows a rapid decreasing dewatering rate. 86 6.5 Assessment of flow resistance coefficients: Paper V The aim with the validation in this chapter is to assess the flexibility of the algorithm, in which volume forces are characterised based on the averaging procedure presented in chapter 5.2. In the algorithm presented in Fig. (23), the simulated flow regimes can be predetermined by configuring the Re-number with parameters such as material properties of the fluid, porosity and pressure loss over the simple domain. Depending on the system at hand, the algorithm could be configured to create numerical data sets at highly turbulent flow regimes, as well as flow regimes with Re-numbers well below creeping flow conditions. In this thesis, the friction factor is investigated for aligned fibrous structure with porosities at 0.7 to 0.8. In Fig. (37), the friction factor is estimated with Re-numbers ranging from approximately 0.0005 to 1000 for an aligned disordered fibrous arrangement with a porosity at 0.8, see Fig. (24a). The figure displays the numerically determined flow resistance coefficients over a wide range of flow regimes in comparison to the formulations of Ergun [80], Montillet et al. [90] and Tamayol et al. [96], which consider both viscous and inertial effects. The expressions from Ergun [80], Montillet et al. [90] are based on the 3-dimensional packing of spheres. Montillet et al. [90] concluded their empirical formulations based on packed beds of spheres also accounting for the bed to particle ratio and level of packing, which is determined by the bed porosity. Tamayol et al. [96] established their formulation based on numerical simulations, see Eq. (64). These simulations are based on a 3-dimensional ordered fibrous structure representing a simple cubic arrangement. Based on Fig. (37), the presented friction factor is in fair agreement with Erguns formulation. 87 Fig. (37). Logarithmic plot of the friction factor and Re-relation for a wide range of Renumbers, ranging from 0.0005 to1000 in an aligned fibrous structure with a porosity at 0.8. A comparison of the friction factor and Re-number relation between the numerical simulations and Blake-Kozeny expression shows the departure from the linear relation between the friction factor and Re-number. In Fig. (37), the numerical friction factor is in good agreement with the BlakeKozeny (BK) expression, which solely considers viscous dissipation and is expressed for packed bed columns as ( ) (65) The expression is valid for Re < 10, which is visible in Fig. (37), as the presented friction factor along with the expression from Ergun [80] are visibly deviating as the Re-number reaches the value 10. Fig. (38) presents the friction factor for an aligned structure at a porosity of 0.7. 88 Fig. (38). Logarithmic plot of the friction factor and Re-relation for a wide range of Re-6 numbers ranging from approximately 10 to 1000 in an aligned fibrous structure with a porosity at 0.7. The figure illustrates a better comparison between the presented friction factor and the expression presented by Tamayol et al. [96]. Fair comparison with Erguns formulation was observed for both porosities. However, the expression presented by Montillet et al. [90] underestimates the friction factor in both cases. By analysing the sensitivity of the sample points presented in chapter 6.3, it is obvious that comparisons to classical empirical correlations is not generalised for a number of cases regarding the porosity. Current numerical data from Yazdchi and Luding [59] was compared to our simulations in Fig. (39). Yazdchi and Luding analysed flow resistance for 2dimensional disordered unidirectional fibers in which their numerical data presents friction factors at porosities of 0.7 and 0.8. 89 Fig. (39). Logarithmic plot of the modified friction factor and Reynolds number relation in comparison with numerical data from Yazdchi at porosities of 0.7 and 0.8. Comparisons are made for aligned disordered fibers. The factor concerning the porosity at 0.7 is overestimated, compared to the simulations of Yazdchi and Luding. They estimated their friction factor at Renumbers ranging up 30; whereas our friction factors are fitted for a broader range which results in an offset. However, reducing our data set at the high Renumbers resulted in a better comparison, which is explained in Fig. (34). Overall, the successful comparison is satisfactory and the observation strengthens the flexibility of the algorithm presented in Fig. (23), which could, in a time efficient manner, generate extensive numerical data sets and accurately estimate flow resistance coefficients for further usage in time efficient simulations of flow in porous media for single and two-phase flow, as well as stationary and time dependant flow processes. 6.5.1 Concluding remarks A few concluding remarks are: (i) The comparison to the classical empirical correlation by Ergun display a fair agreement to the presented friction factor. However, empirical formulations seem to deviate from the presented friction factor depending on the range of Re-numbers the resistance coefficients are fitted over and the porosity of the fibrous structure. (ii) The friction factor for aligned fibrous structures is in good comparison with the numerical study of 90 Yazdchi and Luding provided that the estimation of the friction factor is determined over the same range of Re-numbers. 6.6 Assessment of numerical based design process: Paper I In commercial CFD codes a single user can perform different tasks, such as designing geometry (CAD) and simulating a physical phenomenon at various degrees of complexity. However, when dealing with the design of industrial components and systems, the geometry becomes complex and in many cases extensive. It is reasonable to believe that working with such a geometry design based on fluid mechanic parameters requires knowledge in different fields. Depending on the material and the magnitude of the project, researchers and engineers of different fields are usually involved. In large-scale companies, it may be more effective to create a virtual experiment system that consists of the integration of various elements of technology, something which is acknowledged by Koikekoi [1]. Small-scale projects may, however, be utilised differently. When working with a heuristically based design process, several geometric modifications are conducted based on guesses. Without experimental validation or indicative parameters, modifications such as these cannot be assessed properly. When working with CFD, it is always beneficial to use an experimental laboratory for validation. Then again, certain components are intricate to measure. The establishment of design criteria is, together with visual analysis, used to assess modifications. It should be noted, however, that the validity of the simulation in terms of absolute values such as pressure drop is questionable, due to certain assumptions on boundary conditions and approximations on numerical instabilities. In a design process, such as the one described in Paper I, the issue is to make improvements based on previous models. A constant process of knowledge sharing is established, as there is a continuous verbal communication about the dependency of geometry structure and flow patterns. A summary of the beneficial aspects of implementing CFD in a design cycle are presented in Fig. (40). 91 Design cycle Heat pump tumble dryer Paper machine Internal duct system Vacuum dewatering process Numerical Experimental Numerical Experimental Spatial resolution of flow field Few measurment points Extensive data set of flow process Few sample points Knowledge sharing Trial and error Thorough analysis of system parameters Trial and error Few prototype manufacturing Multiple prototype manufacturing Time efficient simulations Time and resource demanding exp. Fig. (40). Flow chart describing the impact of implementing CFD analysis in the application fields encountered in this thesis. The significance of the knowledge sharing is noticeable, as small-sized industrial companies generally lack a fluid mechanic knowledgebase. As a design process is completed, the mechanical engineer passes on the knowledge to further design projects. It is hence, a favourable approach to sustainable development. In addition, it is beneficial for a company to work with this design process. It is time consuming to design several prototypes and conduct flow and pressure measurements. In addition, it generally lacks the detailed analysis that is important to create a relation of fluid and geometric parameters. It is therefore, profitable to take a numerical based approach. The benefits of implementing CFD analysis on the vacuum dewatering system are similar to those presented for the heat pump tumble dryer. This includes the generation of data sets presenting a highly detailed resolution of the dynamics in the flow process. Along with establishing time efficient simulation models, a thorough analysis of system parameters is possible, which provides an insight in the process which a classical experimental analysis would fail to do. However, it 92 should be pointed out that experiments are crucial in these design cycles in terms of validating simulation data. The involvement of CFD, however, requires analysis of the resolution and the sensitivity of the mesh grid. In more or less every CFD study, a sensitivity analysis of the mesh grid is required, as it plays an essential role in presenting fluid mechanic parameters. There are no specific guidelines for creating certain mesh grids, as flow processes take various forms depending on the geometry of the domain. Hence, establishing a good feel for meshing requires experience in modelling. Moreover, the demand for computer power is reduced by doing so. 6.6.1 Concluding remarks A few concluding remarks based on the assessment made in previous chapter is that; (i) Implementing CFD-models in a design cycle generally yields extensive data points with high resolution in space and time compared to experiments; (ii) Thorough analysis can be conducted on the system instead of the traditional trial and error approach. Moreover, it is possible to share knowledge based on this approach. Time efficient simulations can reduce manufacturing of prototypes and extensive experimental runs, both of which are time and resource consuming factors. 93 7 Conclusions The emphasis of the research conducted in this thesis is to present methods for reducing the complexity of geometrical structures. The focus is to establish time efficient simulations in which the concept is to reduce the geometrical complexity of systems by replacing the internal structure with volume forces as sink terms in the momentum equation. The flow rate and pressure drop relations are well predicted in the volume force models to their respective FRV-simulations. The relation for the heat exchanger presented a mean deviation of approximately 4% whereas the relation for the paper sheet presented a mean deviation of approximately 2.9%. The reliability of these comparisons are based on establishing FRV-models with a consistent mesh grid configuration and a sufficient amount of numerical data points for volume force characterisation. The method of establishing volume forces as sink terms are strengthened based on the comparison to experimental data in which the pressure drop and the dewatering rate are fairly predicated in the heat exchanger and paper sheet respectively. The comparisons indicate that the bulk properties of the system are well predicted, however, the implementation of volume forces eliminates the spatial resolution of the detailed system. In two-phase flow cases, capillary forces are excluded along with the internal solid matrix. Moreover, considerations have to be made when implementing a volume force model as a subdomain in a larger control volume. Key inlet parameters such as turbulence could be mistreated and have significant effects on the following domain. The comparison to classical empirical correlations seem to deviate from the presented friction factor depending on the range of Re-numbers the resistance coefficients are fitted over and the porosity of the fibrous structure. In this thesis however, the formulation by Ergun displays a fair agreement to the presented friction factor for an aligned fibrous structure. Moreover, the friction factor for aligned fibrous structures is in good comparison with the numerical study of Yazdchi and Luding provided that the estimation of the friction factor is determined over the same range of Re-numbers. The computational time of the volume force models was significantly shortened as the geometrical complexity of the simulation domain was reduced, thus 94 reducing the amount of DOF. The volume force model of the heat exchanger successfully predicted pressure drop and flow rates at simulation times of 15 seconds, wherein the computational time of the respective FRV-model took approximately 1 h. Given the fact that the FRV-model only considers a small segment of the heat exchanger, resolving the entire solid matrix would yield an unfavourable amount of DOF. The reduction in computational time enabled time efficient simulations when determining the size of heat exchangers for a given system, with a demand on a certain flow rate and pressure drop. Even though there is a wide access of empirical formulation on flow rate and pressure drop relations for various types of heat exchangers, this method should be applicable on new design configurations in order to quantify such relations. The computational time of a two-phase flow model of the paper sheet with a basis weight of 50 g/m2 was approximately 14 h, wherein implementing volume forces in the paper model reduced computational time to roughly 23 min. In turn, key system parameters in a vacuum dewatering process could be analysed in a more time efficient manner. Such parameters include variation of vacuum levels and duration of pulses, different basis weights, and various arrangements of the forming wire. These parameters are analysed experimentally these days in pilot-scale machines which require preparation, extensive measurement equipment and facilities, pulp and water among other resources. Reducing experimental runs based on accurate numerical models would reduce the time and cost requirements during the design and development cycle. Moreover, dimensioning the components more accurately of a full-scale paper machine would reduce the probability of failed production runs which would result in a considerable amount loss of income. 95 8 Future Research There are several interesting issues that have yet to be studied in this thesis. The method of implementing volume forces have solely been focused on in the momentum equation. Naturally, the next step in the analysis of the heat exchanger unit is to investigate the possibility of solving the energy equation with a similar approach. Hence, certain options will be examined in replacing the tube bundles with a heat source and sink. In order to do so, the NS equations are required to couple with the heat transfer equations. The establishment of a volume force model coupled with heat transfer enables the user to analyse the efficiency of the heat exchanger unit using time efficient numerical simulations. In the presented results, the interaction between the paper sheet and forming wire seems to affect the dewatering process the most. Based on this conclusion, a more physically representable configuration of the forming wire should be investigated further. Moreover, capillary forces have yet to be included in the volume force models. Moving forward, these forces should be analysed further in order to improve the prediction of the models. Even though successful comparison of the friction factor has been established to classically empirical relations, the sensitivity of the resistance coefficients in terms of quantifying them in a certain Re-level require more experimental data in order to strengthen their validity. Overall, there is still a lack in the literature on the knowledge of these resistance coefficients at higher Re-numbers; variation of the coefficients based on the amount and frequency of sample points at higher Re-numbers; variation of the coefficients with the arrangement of 3-dimensional disordered fibrous structures. Based on the established numerical algorithm presented in chapter 5.3.1, an extended set of numerical experiments can be conducted with increased sample points in order to gain understanding regarding the issues just stated. 96 9 References 1. Koikekoi, H. (1999). Evolution of CFD software from academic code to practical engineering software. Journal of Wind Engineering and Industrial Aerodynamics, 81 (1-3), 41-55. 2. Stenzel, I. & Pourroy, F. (2008). Integration of experimental and computational analysis in the product development and proposals for the sharing of technical knowledge. International Journal on Interactive Design and Manufacturing (IJIDeM), 2 (1), 1-8. 3. Petridis, M. & Knight, B. (1996). The integration of an intelligent knowledgebased system into engineering software using the blackboard structure. Advances in Engineering Software, 25 (2-3), 141-147. 4. El-Sayed, M., Sun, T. & Berry, J. (2005). Shape optimization with computational fluid dynamics. Advances in Engineering Software, 36 (9), 607613. 5. Yin Bo, Xu Dian, An Yi-ran & Chen Yao-song (2008). Aerodynamic optimization of 3D wing based on iSIGHT. Applied Mathematics and Mechanics-English Edition, 29 (5), 603-610. 6. Yakinthos, K., Missirlis, D., Palikaras, A., Storm, P., Simon, B. & Goulas, A. (2007). Optimization of the design of recuperative heat exchangers in the exhaust nozzle of an aero engine. Applied Mathematical Modelling, 31 (11), 2524-2541. 7. Lakshmiraju, M. & Cui, J. (2007). Numerical investigation of pressure loss reduction in a power plant stack. Applied Mathematical Modelling, 31 (9), 1915-1933. 8. Anderson, T.B. & Jackson, R. (1967). Fluid mechanical description of fluidized beds. Equations of motion. Industrial & Engineering Chemistry Fundamentals, 6 (4), 527-539. 9. Whitaker, S. (1967). Diffusion and dispersion in porous media. AIChE Journal, 13 (3), 420-427. 10. Slattery, J.C. (1967). Flow of viscoelastic fluids through porous media. AIChE Journal, 13 (6), 1066-1071. 11. Whitaker, S. (1999). The method of volume averaging. Springer. 12. Travkin, V. & Catton, I. (1992). Models of turbulent thermal diffusivity and transfer coefficients for a regular packed bed of spheres. 97 13. Travkin, V. & Catton, I. (1995). A two-temperature model for turbulent flow and heat transfer in a porous layer. TRANSACTIONS-AMERICAN SOCIETY OF MECHANICAL ENGINEERS JOURNAL OF FLUIDS ENGINEERING, 117 , 181-188. 14. Travkin, V. & Catton, I. (2001). Transport phenomena in heterogeneous media based on volume averaging theory. Advances in heat transfer, 34 , 1-144. 15. Gray, W.G. & Lee, P.C.Y. (1977). On the theorems for local volume averaging of multiphase systems. International Journal of Multiphase Flow, 3 (4), 333340. 16. Gray, W.G. & Hassanizadeh, S.M. (1989). Averaging theorems and averaged equations for transport of interface properties in multiphase systems. International Journal of Multiphase Flow, 15 (1), 81-95. 17. Whitaker, S. (1973). The transport equations for multi-phase systems. Chemical Engineering Science, 28 (1), 139-147. 18. Whitaker, S. (1986). Flow in porous media II: The governing equations for immiscible, two-phase flow. Transport in Porous Media, 1 (2), 105-125. 19. Horvat, A. & Catton, I. (2001). Development of an integral computer code for simulation of heat exchangers. In Proc. 8th Regional Meeting, Nuclear Energy in Central Europe. 20. Horvat, A., Rizzi, M. & Catton, I. (2002). Numerical Investigation of Chip Cooling Using Volume Averaging Technique (VAT). Advanced Computational Methods in Heat Transfer, 7 , 373-382. 21. Horvat, A. & Catton, I. (2002). Modeling of conjugate heat transfer using Galerkin approach. URL: www.2.ijs.si/~ ahorvat/publications/mchtugm.pdf, . 22. Horvat, A. & Catton, I. (2003). Numerical technique for modeling conjugate heat transfer in an electronic device heat sink. International Journal of Heat and Mass Transfer, 46 (12), 2155-2168. 23. Horvat, A. & Mavko, B. (2005). Hierarchic modeling of heat transfer processes in heat exchangers. International Journal of Heat and Mass Transfer, 48 (2), 361-371. 24. Zhou, F., Hansen, N.E., Geb, D.J. & Catton, I. (2011). Obtaining Closure for Fin-and-Tube Heat Exchanger Modeling Based on Volume Averaging Theory (VAT). Journal of Heat Transfer, 133 (11), 111802-111802. 25. Feng Zhou, Vasquez, D.A., DeMoulin, G.W., Geb, D.J. & Catton, I. (2012). Volume Averaging Theory (VAT) based modeling and closure evaluation of 98 scale-roughened plane fin heat sink. In Semiconductor Thermal Measurement and Management Symposium (SEMI-THERM), 2012 28th Annual IEEE. 260. 26. Zhou, F. & Catton, I. (2012). Volume Averaging Theory (VAT) based modeling and closure evaluation for fin-and-tube heat exchangers. Heat and Mass Transfer, 48 (10), 1813-1823. 27. International Energy Agency, Cool Appliances – Policy Strategies for Energy Efficient Homes, IEA Publications. (2003). 28. Energy Technology Systems Analysis Program (ETSAP). (2012). Dryers. [Online] Available from: http://ieaetsap.org/web/HIGHLIGHTS%20PDF/R09_HL_Dryers_FINAL_GSOK.pdf [2014-04-16]. 29. Stawreberg, L. (2011). Energy Efficiency Improvements of Tumble Dryers:Technical Development, Laundry Habits and Energy Labelling. 30. Werle, R., Bush, E., Josephy, B., Nipkow, J. & Granda, C. (2011). Energy efficient heat pump driers–European experience and efforts in the USA and Canada. EEDAL, Berlin, . 31. Burns, R.I. (1996). Paper Comes to the West: 800-1400. Gebr. Mann Verlag. 32. Ek, M. (2007). Ljungberg textbook. Pulp and paper chemistry and technology. Book 2, Pulping chemistry and technology. Stockholm: Fiber and Polymer Technology, KTH. 33. Holmberg, K., Siilasto, R., Laitinen, T., Andersson, P. & Jäsberg, A. (2013). Global energy consumption due to friction in paper machines. Tribology International, 62 (0), 58-77. 34. Austin, P.C., Mack, J., McEwan, M., Afshar, P., Brown, M. & Maciejowski, J. (2011). Improved energy efficiency in paper making through reducing dryer steam consumption using advanced process control. In PaperCon. 1122-1132. 35. Austin, P. (2010). Reducing energy consumption in paper making using advanced process control and optimisation. In ELCF Seminar. 36. Austin, P., Mack, J., Lovett, D., Wright, M. & Terry, M. (2002). Improved wet end stability of a paper machine using model predictive control. Control Systems, , 80-84. 37. Bhutani, N., Lindberg, C.F., Starr, K. & Horton, R. (2012). Energy assessment of Paper Machines. Energy Procedia, 14 (0), 955-963. 38. Britt, K.W. & Unbehend, J.E. (1985). Water removal during paper formation. Tappi Journal, 68 (4), 104-107. 99 39. Neun, J.A. (1994). Performance of high vacuum dewatering elements in the forming section. Tappi Journal, 77 (9), 133-138. 40. Räisänen, K.O., Paulapuro, H. & Karrila, S.J. (1995). The effects of retention aids, drainage conditions, and pretreatment of slurry on high vacuum dewatering: a laboratory study. Tappi Journal, 78 (4), 140-147. 41. Granevald, R. (2005). Vacuum Dewatering of Low-grammage Paper Webs and Fabrics. Department of Chemical Engineering, University of Karlstad. 42. Åslund, P., Vomhoff, H. & Waljanson, A. (2008). The deformation of chemical and mechanical pulp webs during suction box dewatering. Nordic pulp & paper research journal, 23 (4), 403-408. 43. Pujara, J., Siddiqui, M.A., Liu, Z., Bjegovic, P., Takagaki, S.S., Li, P.Y. & Ramaswamy, S. (2008). Method to Characterize the Air Flow and Water Removal Characteristics during Vacuum Dewatering. Part I—Experimental Method. Drying Technology, 26 (3), 334-340. 44. Pujara, J., Siddiqui, M.A., Liu, Z., Bjegovic, P., Takagaki, S.S., Li, P.Y. & Ramaswamy, S. (2008). Method to Characterize the Air Flow and Water Removal Characteristics During Vacuum Dewatering. Part II—Analysis and Characterization. Drying Technology, 26 (3), 341-348. 45. Ramaswamy, S. (2003). Vacuum Dewatering During Paper Manufacturing. Drying Technology, 21 (4), 685-717. 46. Åslund, P. & Vomhoff, H. (2008). Method for studying the deformation of a fibre web during a suction pulse. Nordic Pulp and Paper Research Journal, 23 (4), 398-402. 47. Granevald, R., Nilsson, L.S. & Stenström, S. (2004). Impact of different forming fabric parameters on sheet solids content during vacuum dewatering. Nordic Pulp and Paper Research Journal, 19 (4), 428-433. 48. Hubbe, M.A. & Heitmann, J.A. (2007). REVIEW OF FACTORS AFFECTING THE RELEASE OF WATER FROM CELLULOSIC FIBERS DURING PAPER MANUFACTURE. BioResources, 2 (3), 500-533. 49. Danielsson, M., Martinsson, L. & McVey, D. (2011). Enhanced Capabilities in wet-end Paper Machine Clothing. In PaperCon. 2102. 50. Panton, R.L. (2013). Incompressible flow. John Wiley & Sons. 51. Lautrup, B. (2011). Physics of continuous matter: Exotic and everyday phenomena in the macroscopic world. CRC Press. 100 52. Saylor, J. & Bounds, G.D. (2012). Experimental study of the role of the Weber and capillary numbers on Mesler entrainment. AIChE Journal, 58 (12), 38413851. 53. Hager, W.H. (2012). Wilfrid Noel Bond and the Bond number. Journal of Hydraulic Research, 50 (1), 3-9. 54. Hellström, J. & Lundström, T. (2006). Flow through porous media at moderate Reynolds number. In International Scientific Colloquium Modeling for Material Processing, Riga. 55. Jensen, B., Jacobsen, N.G. & Christensen, E.D. (2014). Investigations on the porous media equations and resistance coefficients for coastal structures. Coastal Engineering, 84 (0), 56-72. 56. Burcharth, H. & Andersen, O. (1995). On the one-dimensional steady and unsteady porous flow equations. Coastal Engineering, 24 (3), 233-257. 57. Ma, H. & Ruth, D. (1993). The microscopic analysis of high Forchheimer number flow in porous media. Transport in Porous Media, 13 (2), 139-160. 58. Panfilov, M. & Fourar, M. (2006). Physical splitting of nonlinear effects in high-velocity stable flow through porous media. Advances in Water Resources, 29 (1), 30-41. 59. Yazdchi, K. & Luding, S. (2012). Towards unified drag laws for inertial flow through fibrous materials. Chemical Engineering Journal, 207–208 (0), 35-48. 60. Whitaker, S. (1986). Flow in porous media I: A theoretical derivation of Darcy's law. Transport in Porous Media, 1 (1), 3-25. 61. Brinkman, H.C. (1949). A calculation of the viscous force exerted by a flowing fluid on a dense swarm of particles. Applied Scientific Research, 1 (1), 27-34. 62. Whitaker, S. (1996). The Forchheimer equation: a theoretical development. Transport in Porous Media, 25 (1), 27-61. 63. Giorgi, T. (1997). Derivation of the Forchheimer law via matched asymptotic expansions. Transport in Porous Media, 29 (2), 191-206. 64. Chen, Z., Lyons, S.L. & Qin, G. (2001). Derivation of the Forchheimer law via homogenization. Transport in Porous Media, 44 (2), 325-335. 65. Abu-Hijleh, B.A. & Al-Nimr, M.A. (2001). The effect of the local inertial term on the fluid flow in channels partially filled with porous material. International Journal of Heat and Mass Transfer, 44 (8), 1565-1572. 101 66. Andrade Jr, J., Costa, U., Almeida, M., Makse, H. & Stanley, H. (1999). Inertial effects on fluid flow through disordered porous media. Physical Review Letters, 82 (26), 5249. 67. Costa, U.M.S., Jr., J.S.A., Makse, H.A. & Stanley, H.E. (1999). The role of inertia on fluid flow through disordered porous media. Physica A: Statistical Mechanics and its Applications, 266 (1-4), 420-424. 68. Rong, X., He, G. & Qi, D. (2007). FLOWS WITH INERTIA IN A THREEDIMENSIONAL RANDOM FIBER NETWORK. Chemical Engineering Communications, 194 (3), 398-406. 69. Yazdchi, K., Srivastava, S. & Luding, S. (2010). On the transition from creeping to inertial flow in arrays of cylinders. 70. Yazdchi, K. & Luding, S. (2012). Towards unified drag laws for inertial flow through fibrous materials. Chemical Engineering Journal, 207–208 (0), 35-48. 71. Wang, X., Thauvin, F. & Mohanty, K.K. (1999). Non-Darcy flow through anisotropic porous media. Chemical Engineering Science, 54 (12), 1859-1869. 72. Hassanizadeh, S.M. & Gray, W.G. (1987). High velocity flow in porous media. Transport in Porous Media, 2 (6), 521-531. 73. Zeng, Z. & Grigg, R. (2006). A criterion for non-Darcy flow in porous media. Transport in Porous Media, 63 (1), 57-69. 74. Sidiropoulou, M.G., Moutsopoulos, K.N. & Tsihrintzis, V.A. (2007). Determination of Forchheimer equation coefficients a and b. Hydrological Processes, 21 (4), 534-554. 75. Geertsma, J. (1974). Estimating the coefficient of inertial resistance in fluid flow through porous media. Old SPE Journal, 14 (5), 445-450. 76. Engelund, F. (1953). On the laminar and turbulent flows of ground water through homogeneous sand. Akademiet for de tekniske videnskaber. 77. Dullien, F.A. (1991). Porous media: fluid transport and pore structure. Access Online via Elsevier. 78. Macdonald, I., El-Sayed, M., Mow, K. & Dullien, F. (1979). Flow through porous media-the Ergun equation revisited. Industrial & Engineering Chemistry Fundamentals, 18 (3), 199-208. 79. Moutsopoulos, K.N., Papaspyros, I.N. & Tsihrintzis, V.A. (2009). Experimental investigation of inertial flow processes in porous media. Journal of Hydrology, 374 (3), 242-254. 102 80. Ergun, S. (1952). Fluid flow through packed colums. Chemical and Engineering progress, 48 , 89-94. 81. Layton, W. (2008). Introduction to the numerical analysis of incompressible viscous flows. Siam. 82. Koch, D.L., Hill, R.J. & Sangani, A.S. (1998). Brinkman screening and the covariance of the fluid velocity in fixed beds. Physics of Fluids (1994present), 10 (12), 3035-3037. 83. Clague, D.S. & Phillips, R.J. (1997). A numerical calculation of the hydraulic permeability of three-dimensional disordered fibrous media. Physics of Fluids, 9 (6), 1562-1572. 84. Fotovati, S., Vahedi Tafreshi, H. & Pourdeyhimi, B. (2010). Influence of fiber orientation distribution on performance of aerosol filtration media. Chemical Engineering Science, 65 (18), 5285-5293. 85. Hosseini, S.A. & Tafreshi, H.V. (2010). Modeling permeability of 3-D nanofiber media in slip flow regime. Chemical Engineering Science, 65 (6), 2249-2254. 86. Jaganathan, S., Vahedi Tafreshi, H. & Pourdeyhimi, B. (2008). A realistic approach for modeling permeability of fibrous media: 3-D imaging coupled with CFD simulation. Chemical Engineering Science, 63 (1), 244-252. 87. Ahmed, N. & Sunada, D.K. (1969). Nonlinear flow in porous media. Journal of the hydraulics division ASCE, 95 , 1847. 88. Tallmadge, J. (1970). Packed bed pressure drop—an extension to higher Reynolds numbers. AIChE Journal, 16 (6), 1092-1093. 89. Kovács, G. (1981). Seepage hydraulics. Elsevier, NY. 90. Montillet, A., Akkari, E. & Comiti, J. (2007). About a correlating equation for predicting pressure drops through packed beds of spheres in a large range of Reynolds numbers. Chemical Engineering and Processing: Process Intensification, 46 (4), 329-333. 91. Ozahi, E., Gundogdu, M.Y. & Carpinlioglu, M.Ö (2008). A modification on Ergun's correlation for use in cylindrical packed beds with non-spherical particles. Advanced Powder Technology, 19 (4), 369-381. 92. Çarpinlioğlu, M.Ö, Özahi, E. & Gündoğdu, M.Y. (2009). Determination of laminar and turbulent flow ranges through vertical packed beds in terms of particle friction factors. Advanced Powder Technology, 20 (6), 515-520. 103 93. Hicks, R. (1970). Pressure drop in packed beds of spheres. Industrial & Engineering Chemistry Fundamentals, 9 (3), 500-502. 94. Foscolo, P., Gibilaro, L. & Waldram, S. (1983). A unified model for particulate expansion of fluidised beds and flow in fixed porous media. Chemical Engineering Science, 38 (8), 1251-1260. 95. Benyahia, S., Syamlal, M. & O'Brien, T.J. (2006). Extension of Hill–Koch– Ladd drag correlation over all ranges of Reynolds number and solids volume fraction. Powder Technology, 162 (2), 166-174. 96. Tamayol, A., Wong, K. & Bahrami, M. (2012). Effects of microstructure on flow properties of fibrous porous media at moderate Reynolds number. Physical Review E, 85 (2), 026318. 97. Papathanasiou, T., Markicevic, B. & Dendy, E. (2001). A computational evaluation of the Ergun and Forchheimer equations for fibrous porous media. Physics of Fluids, 13 , 2795. 98. Rezk, K., Nilsson, L., Forsberg, J. & Berghel, J. (2013). Modelling of water removal during a paper vacuum dewatering process using a Level-Set method. Chemical Engineering Science, 101 (0), 543-553. References used for historical input in thesis Darrigol, O. (2005). Worlds of Flow – A history of hydrodynamics from the Bernoullis to Prandtl, Oxford University Press. 104 Methods for Reducing the Complexity of Geometrical Structures Based on CFD Programming As there is an ongoing expansion of CFD usage in industry, certain issues need to be addressed as they are becoming more frequently encountered. The general demand for the simulation of larger control volumes and more advanced flow processes result in an extensive requirement of computer resources. Moreover, the implementation of commercial CFD codes in small-scaled industrial companies seems to generally be utilised as a black box based on the knowledge of fluid mechanic theory. Increased partnerships between industry and the academic world involving various CFD based design processes generally yield to a verbal communication interface, which is a crucial step in the process given the level of dependency between both sides. Based on these notions, a method for establishing time efficient CFD-models with implementation of volume forces as sink terms in the momentum equation is presented. The internal structure, or parts of the structure, in the simulation domain is removed which reduces the geometrical complexity and along with it, computational demand. These models are the basis of assessing the benefits of utilizing a numerical based design process in industry in which the CFD code is used as a communication tool for knowledge sharing with counterparts in different fields. urn:nbn:se:kau:diva-31983 ISBN 978-91-7063-565-6 ISSN 1403-8099 DISSERTATION | Karlstad University Studies | 2014:32

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

Download PDF

advertisement