PROCESS FAULT ANALYSIS USING DIGRAPH METHOD :

PROCESS  FAULT  ANALYSIS  USING  DIGRAPH METHOD :
PROCESS
FAULT
ANALYSIS
USING
DIGRAPH
METHOD
A
THESIS
SUBMITTED
IN
PARTIAL
FULFILLMENT
REQUIREMENTS FOR THE DEGREE OF:
Bachelor of Technology
In
Chemical Engineering
By
KREETI DAS
Department of Chemical Engineering
National Institute of Technology, Rourkela
2012
OF
THE
PROCESS
FAULT
ANALYSIS
USING
DIGRAPH
METHOD
A
THESIS
SUBMITTED
IN
PARTIAL
FULFILLMENT
REQUIREMENTS FOR THE DEGREE OF:
Bachelor of Technology
In
Chemical Engineering
By
KREETI DAS
Under the Guidance of:
Dr. Madhusree Kundu
Department of Chemical Engineering
National Institute of Technology, Rourkela
2012
OF
THE
National Institute of Technology
Rourkela
CERTIFICATE
This is to certify that the thesis entitled, “Process Fault Analysis using Digraph Method”
submitted by Miss Kreeti Das in partial fulfilment of the requirements for the award of
Bachelor of Technology in Chemical Engineering at the National Institute of Technology,
Rourkela is an authentic work carried out by her under my supervision and guidance.
To the best of my knowledge, the matter embodied in this thesis has not been submitted to
any other University / Institute for the award of any degree or diploma.
Date:
Dr. Madhusree Kundu
Department of Chemical Engineering
National Institute of Technology, Rourkela- 769008
ACKNOWLEDGEMENT
I wish to convey my deepest gratitude to my guide, Dr. MadhusreeKundu for her timely
guidance and utmost patience during my project work. I must also thank the HOD of
Chemical Department, Dr. R. K. Singh for providing me with the opportunity of carrying out
this research work.
And I would like to convey my appreciation to all those who directly or indirectly supported
me in carrying out this thesis work successfully.
Kreeti Das
1|Page
ABSTRACT
Fault detection in chemical processes and their isolation is becoming a pressing demand due
to rising concerns regarding safety. An undetected fault can often get out of control and cause
massive losses to an industry in terms of infrastructure and personnel. . When a fault is
detected the next steps comprise of identifying the root of the fault, determining the extent to
which the system functioning can be maintained despite the fault and to find a repair or
solution to the fault. Traditional methods of fault diagnosis include real time analysis of faults
or/and sequential testing procedures but none guarantee an early detection; thus confusion is
inevitable in case of multiple faults. Due to these limitations, the digraph method of fault
detection has been an area of interest lately. The foundations of this method are developed
based on human reasoning and analytical skills and later the prepared model is validated with
the help of simulation. The advantages of this method being the relative simplicity of the
model to be referred in the case of a fault and the fast diagnosis of root of fault; this process
has been applied to three processes in this paper and its effectiveness observed.
The three processes that have been studied are: Jacketed Continuous Stirred Tank Reactor,
Binary Distillation Column and Drum Boiler. Mechanism of each process was studied in
detail; their mathematical modelling was done for the purpose of simulation using MATLAB.
Using the working principle of each process and theoretical knowledge, digraph was
developed for each process. The variables were denoted by nodes and the relationship among
them by continuous and dashed lines. Once the digraph was developed, simulation of each
process was carried out and the results compared to the digraph in order to validate it.
2|Page
The disadvantage of the digraph method is that fault detection is only qualitative which
reduces the resolution of fault detection. Also a variable value may fluctuate due to the action
of control loops giving compensatory variables (CV) or inverse variables (IV) and the fault
diagnostic system should be efficient enough to monitor or predict such changes as well.
Keywords: Fault detection, Digraph method, Process Variables, Compensatory Variables,
Inverse Variables
3|Page
Table of Contents
1) ACKNOWLEDGEMENT ...................................................................................................................... 1
2) ABSTRACT ......................................................................................................................................... 2
3) INTRODUCTION ................................................................................................................................ 6
1.1.
INTRODUCTION ....................................................................................................................... 7
4) LITERATURE REVIEW ...................................................................................................................... 12
2.1. THE DIGRAPH METHOD ............................................................................................................. 13
2.2.1. DETAILS OF THE SYSTEM ..................................................................................................... 15
2.2.2. OPERATING MODES ............................................................................................................ 16
2.2.3. STEPS FOR DEVELOPING DIGRAPH...................................................................................... 16
2.2.4. WATER TANK DIGRAPH ....................................................................................................... 17
2.2.5. DIGRAPH DIAGNOSTIC METHODS....................................................................................... 20
5) JACKETED NON-ISOTHERMAL CONTINUOUS STIRRED TANK REACTOR ......................................... 24
3.1 INTRODUCTION ........................................................................................................................... 25
3.2 FAULT ANALYSIS .......................................................................................................................... 29
3.3 COMPENSATORY VARIABLE AND INVERSE VARIABLE................................................................. 30
6) BINARY DISTILLATION COLUMN ..................................................................................................... 31
4.1 INTRODUCTION ........................................................................................................................... 32
4.2. DIGRAPH ANALYSIS .................................................................................................................... 37
7) DRUM BOILER ................................................................................................................................. 39
5.1. INTRODUCTION .......................................................................................................................... 40
5.2. DIGRAPH ANALYSIS .................................................................................................................... 43
8) CONCLUSION AND RECOMMENDATION ........................................................................................ 48
6.1 CONCLUSION AND RECOMMENDATION..................................................................................... 49
9) REFERENCES ................................................................................................................................... 51
10) NOMENCLATURE ............................................................................................................................ 52
4|Page
LIST OF FIGURES
Figure 1.1: Jump fault behaviour ............................................................................................................ 9
Figure 1.2: Intermittent fault behaviour……………………………………………………………………………….8
Figure 1.3: Drift Fault Behaviour…………………………………………………………………………………........8
Figure 2.1: A simple digraph representation……………………………………………………………………………………13
Figure 2.2: Water tank system……………………………………………………………………………………………12
Figure 2.3: SDG of water tank system…………………………………………………………………………………14
Figure 2.4: Valve 2 (V2) unit digraph…………………………………………………………………………………..16
Figure 2.5: Valve 3 (V3) and control loop (C2) unit digraph………………………………………………..16
Figure 2.6: SDG of the whole system…………………………………………………………………………………..17
Figure 2.7: SDG of a PVC plant (partially)……………………………………………………………………………………19
Figure 3.1: Jacketed non-isothermal CSTR model along with level controller……………………23
Figure 3.2: SDG Model of JCSTR………………………………………………………………………………………..25
Figure 4.1: Schematic diagram of a distillation column ........................................................................ 32
Figure 4.2: SDG of binary distillation column………………………………………………………………………33
Figure 5.1: Schematic diagram of a drum boiler ................................................................................... 40
Figure 5.2: SDG of drum boiler……………………………………………………………………………………………38
Figure 5.3:Open-loop Response for step increase in Mf…………………………………………………….44
Figure 5.4: Open-loop Response for step increase in Ms……………………………………………………45
Figure 5.5: Open-loop respose for step increase in Q……………………………………........…………...46
Figure 5.6: Closed-loop response to a step increase in Q……………………………………………………47
LIST OF TABLES
Table 2.1: Loop status of nodes in PVC plant (partial)…………………………………………………………………….20
5|Page
Chapter1
INTRODUCTION
Background and Objective
6|Page
1.1.INTRODUCTION
The two primary concerns of any industry are to ensure production of economically efficient
and quality goods and to take safety measures for the workers. In chemical industries highly
toxic substances are dealt with and an accident can cause huge scale damage such as that in
Bhopal Gas Tragedy. It is of utmost importance that faults are detected early and proper steps
be taken to repair those faults so that no harm is done either to the workers or to the
machinery of the industry. Several sciences like statistics, system science, signal processing,
fuzzy logic and computer science have contributed to the development of fault detection and
diagnosis (FDD) techniques. FDD generally includes the following steps [1]:
(i)
Fault Detection: Identification of abnormal system behavior and deviation of
process values from their expected ones.
(ii)
Fault Isolation: Determining the location and accurate cause of the fault among
many other possible causes.
(iii)
Fault Identification: Determination of the magnitude of fault or the degree by
which the observed values differ from the expected ones.
Faults can be categorized into three categories: Sensor Fault, Actuator Fault and Process
Fault. Sensor function is to measure process variable. If the measured value is different from
actual variable, it is known as sensor fault. Actuator is the one actually carrying out the
operation and provides output. If there is discrepancy between command given to the
actuator and the output, it is known as actuator fault.. Process faults include the other faults
of the system which may be additive or multiplicative. Additive faults are unknown inputs
like a leak whereas multiplicative faults are gradual changes like fouling of heat exchanger
surfaces or rusting of iron parts of a machine.
7|Page
For FDD, two types of methods can be used, model-based method and model-free method.
Model-based methods use a mathematically developed model of the process to estimate the
correct values of the process variables whereas model-free methods do not use a
mathematical model and find faults using defined laws and theories. The performance of
FDD is characterized by the following:
(i)
Sensitivity: It is the ability to detect or diagnose a fault of a specific size. The size
of a fault is defined by the range of its effect.
(ii)
Discrimination Power (Isolation Power): It is the ability to detect correct fault
when several faults happen simultaneously mask each other.
(iii)
Robustness: It is the ability to detect a fault among noise, disturbance and
modeling errors.
(iv)
Missed Fault Detection and False Alarm: It indicates the number of faults that
went undetected and the number of times alarm was issued when there were no
faults.
(v)
Detection and Diagnosis Speed: Indicates time taken by the system to detect and
diagnose faults after their occurrence.
Redundancy is an important criterion in FDD. If duplicate sensors are used to detect a fault, it
is called physical redundancy. If a process model is used to estimate process variables and
the difference between measured and estimated values forms the basis of diagnosis, then it is
called analytical redundancy. Physical redundancy is expensive as more hardware is required
and concentrates on one variable.
Faults can be further classified into abrupt (sudden faults) and incipient (slowly developing
faults). Abrupt faults are dangerous and need to be detected immediately. Incipient faults
8|Page
develop over a period of time, e.g., deposits on heat transfer surface in an exchanger. They
are difficult to detect and model-based techniques are more useful in such cases. Time
behavior of faults can be classified into:
(i)
Jump: Jump in sensor reading is generally caused by bias change or breakdown.
Figure 1.1: Jump fault behaviour
(ii)
Intermittent: Loose wiring or erroneous data recording results in intermittent
fault-time behavior.
Figure 1.2: Intermittent fault behaviour
9|Page
(iii)
Drift: If the sensor is warming up or actuator is wearing out, it results in drift
fault time behavior.
Figure 1.3: Drift Fault Behaviour
In model-based fault detection, a mathematical model of the fault is used and estimation of
variables is done based on the model. Then the measured values are compared to estimated
values. The difference between these two values is called residual and residuals indicate the
presence of faults.
The digraph method is the mode of fault detection that has been studied in this paper.
Digraph method, unlike other mathematical models, does not give a quantitative idea of the
process under study. Rather mathematical modelling of the process is carried out beforehand
and the obtained residues indicate fault, after which the role of digraph comes in.The greatest
advantage of the digraph method is the utter simplicity of the figure obtained; even a layman
can figure out the relationship between the various process variables and how they affect each
other. Digraphs are prone to errors; to validate a given digraph, simulation results of the
10 | P a g e
particular process come to use. Once the tedious task of validating the digraph is over, fault
detection by back-tracing through the digraph is child’s play.
11 | P a g e
cHAPTER 2
LITERATURE REVIEW
The Digraph Method Fundamentals
12 | P a g e
2.1. THE DIGRAPH METHOD
Digraph, also known as directed graph illustrates the fault propagation through a system. It
comprises of a set of nodes (V) representing the system process variables and edges (E)
representing the relation between the nodes. The digraph model is represented by:
DG= (V,E)
Examples of nodes/process variables: Temperature, pressure, mass flow rate, signals from
sensors. Digraphs, being qualitative in nature, deviations in variables in digraphs may be
represented by five discrete values: +10 (very high), +1 (moderate high), 0 (normal), -1
(moderate low) and -10 (high low).
Figure 2.1: A simple digraph representation [2]
In the above figure, M1 and M2 are representing mass flow rates at location 1 and 2. M1 is
independent and M2 is dependent. This figure indicates that under normal conditions, the +1
edge represents the relationship between the two nodes. The two other edges are applicable
only if the mentioned conditions are satisfied. The gain of such a loop is given by:
Gain= Δ (M2)/ Δ (M1)
Control loops are present in the process and they consist of sensor, controller and control
device. Two basic control loops that can be represented by digraphs:
(i)
Negative Feedback Loops: Corrects moderate deviations in variables. Path on a
digraph start and end on the same node and product of all normal gains around
13 | P a g e
this loop is negative. It measures the output, finds the difference between set-point
and output and thus controls the output.
(ii)
Negative Feedforward Loops: In theory, any disturbance can be cancelled in this
loop but practically that is not possible. In digraph, represented as two or more
paths from one node to another node and sign of product of all normal gains is
different on each of the paths. Feedforward controllers measure load directly and
accordingly control output.
Negative loops are very important in any process as they play the role of stabilizers. If we
have a positive loop and the gain is positive, it means disturbance in a particular direction in
a variable will lead to similar disturbances in other related variables and the fault will keep
on multiplying until it gets out of control. On the other hand, if there is a negative gain
between two variables, then an increase in one will lead to a decrease in the other and this
counter effect stabilizes the system. By drawing a digraph of a system, we can accurately
keep track of the number of negative and positive loops and try to keep the number of
negative loops to maximum.
Following is the example of a water tank system, courtesy literature review, in which the use
of digraph method has been studied to determine the causes of possible faults [2].
14 | P a g e
2.2. THE WATER TANK SYSTEM
It is the simplest system to explain the digraph method. The schematic diagram is given below:
Figure 2.2: Water tank system [2]
The aim of this arrangement is to maintain the level of water between pre-determined levels.
The water level is monitored by sensor1 (S1). Under normal conditions, water flows in
through the valve1 (V1) and flows out through valve2 (V2). Valve V3 is given as a safety
valve. The tray is provided to collect water from any leakage or spill due to overflow.
2.2.1. DETAILS OF THE SYSTEM
The system consists of three valves V1-3, two level sensors S1-2, two controllers C1-2 and a
spill tray. S1 senses the level of water and sends signal to controller C1 which in turn
controls the valve V1. If the level raises more than the desired level, the controller sends
command to V1 to shut down so that supply from mains is cut off and water is drained out.
When the water level drops V1 is opened by C1 allowing water in. V2 is a manual valve and
15 | P a g e
thus no controller is associated with it. V3 is a safety valve and is dormant under normal
conditions. In case if C1 fails and water level becomes very high, the sensor S2 detects the
level and sends signal to C2 which in turn opens V3 to let the excess water out. Even after
that if overflow occurs and water is spilled then it is stored in the tray. Control loop 2 is thus
redundant until control loop 1 is working properly. The flow sensors VF1-3 measure the
flow-rates through the valves V1-3. A sensor SP1 is located in the tank to inform whether
there is a leak or not.
2.2.2. OPERATING MODES
There are two operating modes: Active and Dormant. In active state, the valves V1 and V2
are open and V3 is closed. The system is dormant when all the valves are closed.
2.2.3. STEPS FOR DEVELOPING DIGRAPH
There is a step by step process of developing digraph of a system.
(i)
The process to be analyzed is defined in details.
(ii)
All the possible component failures are considered.
(iii)
The system is classified into sub-systems and components.
(iv)
The control loops present in the system are identified.
(v)
The digraphs of sub-systems are developed first by considering all process
variable deviations and the effect they have on the variable present in the model.
Also the magnitude of the effect is to be represented by using the values of 10, 1,
0, -1 and -10.
(vi)
The models of sub-systems are joined to form the digraph of the whole system.
(vii)
Causes of faults are detected by back-tracing from the node showing fault.
16 | P a g e
2.2.4. WATER TANK DIGRAPH
The following assumptions are made while making the digraph:
(i)
If there is a pipe rupture, it is not detected by flow sensors.
(ii)
A rupture in the tank causes more volume loss than a tank leakage.
(iii)
The system is at steady state initially.
Digraph for each sub-unit is developed as shown in the figures below. The digraphs are
constructed based on the proper knowledge of the process and the components, basic laws of
science and human reasoning. The variables are represented by:
M= mass flow rate; L= level; P= pressure
Control loops within the water tank system arerepresented as negative feedback loops
(NFBL), where the product of the normal gains around the loop isnegative. NFBL’s are used
since they introduce theability to correct any moderate disturbances which maybe present in
one of the process variables.
Figure 2.3: SDG of water tank system [2]
17 | P a g e
The air to open characteristics of V1 is represented by the relationship between M1, M2 and
P3. The control action of C1 is represented by the negative feedback loop of M2-L4-P5-P3; it
is a negative feedback loop as the product of normal gains is negative. Now the positive
―signed‖ edges represent a directly proportional relation between the two nodes where as the
negative ―signed‖ edges show an inversely proportional relation. The magnitude 1, 10 or 0
just represent the degree of effect either positive or negative. Now let us take a look at the
various possible faults and their causes. M1 entering V1 can be affected by three kinds of
component failures P1B, P1R or NMWS, all of which will decrease the mass flow. Checking
the node M2, we see that M2 can be affected by four types of component failure: P2B, P2R
and V1FC decrease M2 whereas V1FO increases the mass flow M2. M2 is also related to M1
via two ways. In the normal condition, an increase in M1 increases M2 proportionately. But
in case V1 does not work (given by condition V1C), there is no effect of M1 on M2. A
disturbance in the control loop M2-L4-P5-P3 can also disturb M2.
Concentrating on the control loop, let us assume that M2 increases. As positive relation exists
between M2 and L4, the level of water in tank L4 also increases which is detected by a
sensor and it sends a high signal, in this case P5 (higher pressure) to the controller. There is
an inverse relation between P5 and P3 which means that the controller on receiving high
signal from P5 gives a low signal to P3 (lower pressure). As V1 is an air to open valve, it
closes due to low pressure P3 and M2 decreases. This results in the reduction of water level.
If M2 decreases for some reason then just the reverse action takes place and liquid level
becomes higher. TR (tank rupture) and TL (tank leak) as obvious will have a negative effect
on liquid water level and positive effect on water collected in the tray. Similarly based on
logical reasoning other nodes can be studied and interlinked to each other. The digraph for
the manual valve V2 is given in figure 2.4.
18 | P a g e
Figure 2.4: Valve 2 (V2) unit digraph [2]
The different component failures affecting M6 and M7 are shown. An increase in L4,
increases M6 which in turn increases M7. The relation between M6 and M7 is nullified if V2
is closed.
Figure 2.5: Valve 3 (V3) and control loop (C2) unit digraph [2]
Figure 2.5 illustrates the air to close valve V3 through the relationship between M8, M9 and
P10. The loop M8-L4-P11-P10 depicts the negative control loops; here also the product of
19 | P a g e
normal gains is negative. If the level becomes very high sensor S2 sends a high pressure
signal (P11) to the controller. Negative relation exists between P11 and P10, hence the
controller sends a low signal (P10) to the valve. As V3 is air to close valve, low pressure
opens the valve and flow increases giving high M8 and M9.
The complete digraph system shown in figure 2.6 is obtained by combining figures 2.3, 2.4
and 2.5.
Figure 2.6: SDG of the whole system [2]
2.2.5. DIGRAPH DIAGNOSTIC METHODS
For fault diagnosis, the system sensor readings are compared with the expected values while
the system is in operating mode. If a node registers a deviation, diagnosis involves backtracing from the node through which it is possible to determine the failure nodes. Backtracing is done until all the nodes possibly causing the fault are marked. It is done in two
ways:
20 | P a g e
(i)
Method One: In this approach back-tracing is done from the faulty node until the
point where no further back-tracing is possible. The disadvantage is that many
fault options are generated for a single fault and some of them are contradictory.
So this creates an ambiguous situation.
(ii)
Method Two: From sensor readings it is observed that which particular areas is
showing deviation and that particular area is flagged off for back-tracing, leaving
the non-deviating nodes intact. Back-tracing from a node ceases as soon as the
boundary of flagged section is reached. This method is particularly useful in case
multiple sensor faults.
Let us consider an example of Method Two. A deviation from the normal active mode is
taken in which VF1 and VF2 are registering no flow. As there is no problem with VF3 or
SP1, the part of digraph containing these nodes is flagged off. Now no flow in V1 will cause
M2 to decrease which can be caused by P2B, P2R or V1FC. It could also have been caused
by decrease in M1 which in turn could have been caused by P1R, P1B or NMWS. Going by
the control loop, the decrease in M2 can also be caused due to high liquid level L4. Similar
reasoning can be applied to the no flow condition registered by VF2.
Another example taken from literature review, illustrating the use of digraph method, is the
PVC plant [3].
2.3. PVC PLANT
In a PVC plant, hydrogen and chlorine are obtained by electrolysis and coupling reaction of
hydrogen and chlorine is done to produce hydrochloric gas. Part of the process is represented
by the digraph given below [4]:
21 | P a g e
Figure 2.7: SDG of a PVC plant (partially) [4]
This figure can be used to determine the types and number of control loops present in the
system. The order of a loop is nothing but the number of nodes present in the loop. As
already mentioned above, the loop the product of normal gains of whose edges is negative is
a negative loop otherwise it is positive. A positive loop is dangerous from safety point of
view as the effect of deviation whether increasing or decreasing, advances or goes on
multiplying making the system unstable. To make a system stable more and more negative
loops are used so that no one effect dominates and contradictory effects exist to balance.
In the above figure we see there are two 3-order positive loops: Cy1-Y1-Z1 and Cy2-Y2-Z2.
There are four 3-order negative loops: Cz1-Z1-Zp; Ck2-K1-Zp; Cz2-Z2-Zp and Cx2-X2-Xp.
Similarly there are positive and negative 4-order loops. The complete list of number and
types of loops related to a node is known as its loop status. The loop status of nodes in the
above digraph is given in the following table.
22 | P a g e
Table 2.1: Loop status of nodes in PVC plant (partial)[4]
Following chapters are based on the three processes that have been studied as a part of this
project and their digraphs have been developed.
23 | P a g e
CHAPTER 3
JACKETED NON-ISOTHERMAL CONTINUOUS STIRRED TANK
REACTOR
Mathematical modeling
Development of digraph
24 | P a g e
3.1 INTRODUCTION
A JCSTR was used for analyzing the effect of deviation in one variable on all the other
variables. Liqiang Wang [5] has shown that in this case, fault in one node propagates through
various pathways and the most dominant of these pathways is determined. The concept of
compensatory response (CR) and inverse response (IR) has also been explained by Wang. To
study the concept of CR and IR, three qualitative states are identified: 0 (normal), +1 (high)
and -1 (low). It is assumed that initially all the nodes are at steady state and hence are
assigned 0 before fault is introduced. When a fault occurs, deviation of the root node causes
all variables accessible from the root node to change sign. The first sign change (0 to +1 or 0
to -1) is considered as the initial response of the system. Due to the presence of numerous
feed forward and negative feedback loops and control action, the signs of variables may
change during propagation of the fault. The final state or ultimate response of a node can be
to retain the initial state; return to the steady state or to have the inverse state. If the variable
returns to the steady state value of 0, then it is known as compensatory variable (CV) and if
the variable takes the opposite value of that in initial response, then it is known as inverse
variable (IV). The whole rule base for fault detection has three parts: the normal state before
fault introduction, the initial and ultimate response after fault introduction.
The JCSTR model is given in figure 3.1. The objective was to control the volume in the tank
at desired value using the level control (LC) loop. As a non-isothermal reactor was
considered, temperature could vary but extreme values had to be avoided. Hence, cold water
was circulated through the jacket to cool the reaction solution inside the reactor heated up by
an exothermic reaction. The volume was controlled by adjusting the tank outflow and the
temperature was controlled by adjusting cold water inflow valve. Assumptions made were
25 | P a g e
that no phase change occurs either in the tank or the jacket. Also the jacket volume, density
and heat capacity of fluids remain constant. Heat transfer to surroundings was neglected.
Figure 3.1: Jacketed non-isothermal CSTR model along with level controller
Mass balance inside the reactor:
(Assuming constant density)
(3.1)
Component Balance inside the reactor:
(
)
(
)
(3.2)
Energy Balance in the reactor:
(
)
(
)
(
)
(3.3)
Energy Balance inside the jacket:
(
)
(
)
(3.4)
26 | P a g e
These are the four equations of state which tell us about the behavior of the system and these
equations are used for the simulation of the jacketed CSTR. Now, in any system, there are
some variables and some equations. Under ideal conditions, the number of equations is equal
to the number of variables and all the variables have unique values. When the number of
equations is more than the number of variables, there is no feasible solution of the system.
Again, if the number of variables is greater than the number of equations, then there are
infinitely many solutions as the variables can take any arbitrary value. Keeping the above
points in mind, it is very important to check the degree of freedom (Number of variablesNumber of equations) of any system. In case of a positive degree of freedom, control
parameters are introduced to balance the extra variables.
From
the
above
figure,
the
variables
Total we get 11 variables. But out of these 11,
in
the
JCSTR
are:
are constant as the jacket volume
remains constant. Thus the actual number of variables is 9. Hence degree of freedom:
Degree of Freedom= 9-4 =5
Out of these 5 degrees of freedom, Fo, Cao and To can be considered as forcing variables or
disturbances that can be controlled manually. The rest 2 can be balanced by the following
equations:
Level Controller Equation:
(
)
(3.5)
27 | P a g e
Here we can see that F is the manipulated variable and V is the controlled variable
(indirectly, as we are controlling the height). Kv represents the outflow valve attached to the
level controller.
Heat Transfer Equation:
(
)
(3.6)
Where
(3.7)
After the development of mathematical model of the process, its digraph was developed. The
digraph is shown in figure 3.2.
Figure 3.2: SDG Model of JCSTR
28 | P a g e
3.2FAULT ANALYSIS
In the above digraph, the continuous lines between two variables suggest a directly
proportional relationship between the two, whereas the dashed lines represent inversely
proportional relationship. The direction of effect is represented by the direction of the arrow.
This digraph consists of one three-order negative loop and no positive loops and thus is
stable. A probable disturbance was assumed and its propagation through the system was
observed. The propagation of fault through the system, when back-traced from the node
showing fault, gives us the originating node.
A sudden increase in the inflow rate Fo was assumed. The consequent effects on other
variables as traced in the digraph are:
(i)
Fo ---- V ; thus V increases
(ii)
Fo --V --- F ; thus outflow rate F increases to bring down V
(iii)
Fo -- V -- A -- T; hence T decreases. This is because when V increases, the
area of heat transfer increases and more heat is given out to the jacket. Hence,
temperature inside the tank is reduced.
(iv)
Fo -Ca; the concentration of reactant increases with increase in inflow rate
(v)
Fo --Ca --- T; as reactant concentration increases, reaction rate is faster and
more heat is released
(vi)
Fo - T --Tj; Tj decreases. Due to low inside temperature, jacket fluid does
not heat up much
(vii)
Fo - V -- A -- T --Tj; Tj decreases
(viii) Fo ----Ca --- T ---Tj; Tj increases
29 | P a g e
In the last three points it was observed that contradictory results were obtained. In two cases,
Tj decreased with an increase in inflow rate. But in the last case Tj increased with increase in
flow rate. In any process, a particular variable is affected by many other variables and their
effects may be contradictory. According to work carried out by AtallaSayda[6], such
problems can be overcome with the help of simulation. If the Tj versus time graph shows a
positive slope when Fo is changed positively, path (viii) is dominant. On the other hand, if Tj
versus time shows negative slope with an increase in Fo, any of the pathways (vi) or (vii) is
dominant. To decide which among the two is more dominant, the slopes of the corresponding
graphs are compared. The one with a steeper slope will have higher dominance.
3.3 COMPENSATORY VARIABLE AND INVERSE VARIABLE
Now, let us test the theory of CV and IV. For this purpose, pathway (viii) was assumed to be
dominant, i.e., Tj increases when Fo increases. Any other pathway can also be assumed. The
measured out variables of the system are F, T, Ca, V, Tj. These five variables are represented
in a set: S [F, T, Ca, V, Tj]. (+) sign indicates a positive deviation and (-) indicates negative
deviation. In case of Fo(+), the initial response of the system (IR) is as follows: S [+, -, +, +,
+]. But after some time, the controller C1 will come into action whose objective is to
maintain desirable V value. As V goes on increasing, the controller will try to counter the
effect with the help of manipulating variable F. Thus the final steady state condition is that V
is decreased. The final result after controller action is known as the ultimate response (UR).
The controller may bring the value of V to the initial steady state value (represented by 0).
Thus the ultimate response will be S [+, -, +, 0, +]. In his case the variable V is known as
compensatory variable or CV. But if the controller action overshoots the steady value and the
final value of V shows a negative deviation, then the ultimate response is S [+, -, +, -, +]. In
this case the variable V is known as inverse variable or IV.
30 | P a g e
CHAPTER 4
BINARY DISTILLATION COLUMN
Mathematical modelling
Development of digraph
31 | P a g e
4.1 INTRODUCTION
The working of distillation column was studied in detail, its mathematical modeling done and
then its digraph prepared. The digraph was validated after simulating the model in MATLAB.
The schematic diagram of the distillation column is given below:
Figure 4.1: Schematic diagram of a distillation column
The following assumptions were made while studying the system [7]:
32 | P a g e
(i)
There are N trays in the column and all the trays are 100 % efficient
(ii)
The molal vapor flow-rate remains constant throughout the tower
(iii)
The heat of vaporization of the down-coming fluid is almost equal to the heat of
condensation of the upcoming vapour; sensible heat difference are also same
(iv)
There is no heat loss to the surrounding
(v)
The liquid flow rate from each tray depends on its liquid hold-up
(vi)
Vapour hold-up on each tray is neglected
(vii)
Relative velocity of the two components remains constant throughout the tower and
momentum balance on each tray is neglected
Mass balance in feed plate (i=f):
(
)
(4.1)
Component balance in feed plate (i=f):
(
)
(4.2)
Mass balance for top plate (i=N):
(4.3)
Component balance for top plate (i=N):
(
)
(4.4)
Mass balance for bottom tray:
33 | P a g e
(4.5)
Component balance for bottom tray:
(
)
(4.6)
Mass balance for rest trays:
(4.7)
Component balance for rest trays:
(
)
(4.8)
Mass balance for reflux drum:
(4.9)
Component balance for reflux drum:
(
)
(
)
(4.10)
Mass balance for column base:
(4.11)
Component balance for column base:
(
)
(4.12)
The above were differential equations. The algebraic equations governing the process are:
Equilibrium relation:
34 | P a g e
(
)
(4.13)
Francis Weir relation:
(
)
(4.14)
Total number of variables = 4N+11
Tray liquid and vapour composition (xi and yi) = 2N
Tray liquid hold-ups (Mi) = N
Tray liquid flow-rates (Li) = N
Bottom liquid and vapour composition (xb and yb) = 2
Molal vapour flow-rate throughout the tower = 1
Reflux drum composition (xd) = 1
Reflux flow-rate (Fr) = 1
Distillate flow rate (Fd) = 1
Bottom liquid flow rate (Fb) = 1
Reflux drum and bottom hold-up (Mrd and Mb) = 2
Feed flow-rate and composition (F and xf) = 2
Total number of equations = 4N+5
Tray mass balance = N
Tray component balance = N
Mass balance in reflux drum and bottom = 2
Component balance in reflux drum and bottom = 2
Equilibrium relation on trays = N+1
Francis Weir relation on all trays = N
35 | P a g e
Hence, degrees of freedom = (4N+11) – (4N+5) = 6
Out of these six, xf and F can be considered as the forcing variables or disturbances. The rest
four degrees of freedom are balanced by the use of four controllers whose control objectives
are:
(i)
To maintain desirable reflux drum hold-up
(ii)
To maintain optimum composition in reflux stream
(iii)
To maintain desirable bottom hold-up
(iv)
To maintain optimum composition in bottom stream
These four controllers have been shown in figure 4.1 above.Figure 4.2 shows the binary
distillation column digraph.
Figure 4.2: SDG of binary distillation column
36 | P a g e
4.2. DIGRAPH ANALYSIS
In the above digraph, there are four 3-order negative loops and no positive loops. Hence, the
system is stable. There are two forcing functions: F and xf. A probable disturbance was
assumed in both the variables and the propagation of fault through the system was observed.
If there is a sudden decrease in the feed rate F, the hold up in plates from i=F to i=1 decreases.
As the liquid flow rate from trays is a function of the hold-up, the liquid flow-rates (L) of the
above mentioned trays are also reduced. Due to low liquid flow rate, the bottom liquid level
starts decreasing. The objective of controller C3 is to maintain an optimum level of liquid
level in column bottom. When the liquid level reduces below the optimum level, it sends a
signal to the controller, which in turn reduces the output flow rate (Fb) of bottom liquid thus
stabilizing the liquid level. The propagation of fault induced by disturbances in the other
forcing function xf can be similarly traced out in the digraph.
Now let us test the back-tracing technique by assuming a fault in any random variable. Let it
be Mrd. Back-tracing is always done in the direction opposite to that of the arrows as we are
tracing the cause and not the effect. If the level sensor is showing negative reading (level
lower than desired), then as per the digraph, we can trace the fault through the following
paths:
(iv)
A positive deviation (increase) in Frd. It is in turn caused by a negative deviation in
xd. The negative deviation can be finally traced back to xf through the joining lines
and thus it may be a possible cause of the original deviation.
(v)
Another path is that through V (V decreases). Tracing the lines we see that the
negative deviation is finally traced to xf through xb and x.
37 | P a g e
Although we took two different routes to back-trace, we finally reached the same fault root,
i.e., a negative deviation in xf.
38 | P a g e
CHAPTER 5
DRUM BOILER
Mathematical modelling
Development of digraph
39 | P a g e
5.1. INTRODUCTION
Drum boiler is a device in which water is the input feed and steam is the output product. It
consists of a drum to which water is fed, the water flows down the downcomer connected to
the drum; the riser section of the tube is exposed to radiation and thus the water heats up. The
steam formed travels back to the drum via connecting tubes. When steam enters the drum, the
drum pressure increases. To reduce the pressure inside and bring it to the saturation pressure
of the drum liquid, some of the steam is converted back to water while the other fraction
escapes the drum through outlet. The schematic diagram of a drum boiler is given in figure
5.1.
Figure 5.1: Schematic diagram of a drum boiler
The inputs to the system are heat flow to the riser (Q) and the input feed water rate (Mf) while
the outputs of the system are the drum level (l) and drum pressure (P). The following
assumptions were considered to develop the mathematical model [8]:
(i)
Water is in thermodynamic saturated state
40 | P a g e
(ii)
Instantaneous and uniform thermal equilibrium is maintained between water and
metal everywhere
(iii)
Steam and water readily release and absorb energy with changing pressure
(iv)
At steady state metal temperature is very close to the saturation temperature of
water and thus there is significant sensible heat transfer
Overall mass balance for the drum boiler:
(5.1)
Overall energy balance:
[
]
(5.2)
As the water passes through the riser region, it gets heated up by the heat radiation and a
mixture of steam and water is formed. To composition of the mixture is denoted in terms of
and
which represent the volume fraction of vapour and mass fraction of vapour
respectively.
Mass balance for riser:
(
)
(5.3)
Energy balance for riser:
[
(
)
]
(
)
(5.4)
The main forces acting on the system are:
(i)
Inertial force:
(
)
41 | P a g e
=(
)
(ii)
Driving force due to vapour-liquid density difference: (
(iii)
Friction force :
)
Thus overall momentum balance for drum boiler:
(
)
(
)
(5.5)
Drum level (l) is given by:
(
)
(5.6)
The above six equations determine the behaviour of the system and are called state equations.
The process variables are: Mf, Vwd, Vsd, Vwt, Vst, qdc (= qr), l, Ms, P, Q, αr and αv
Thus total number of variables= 12
Total equations=6
Degrees of freedom= 12-6 = 6
Out of the 12 variables, Mf and Q are forced variables; so we have 4 degrees of freedom to
balance. One equation exists between αv and αr:
(
)
(5.7)
The rest three degrees of freedom are balanced by the use of three controllers as is shown in
the digraph in figure 5.2.
42 | P a g e
αv
αr
Figure 5.2: SDG of drum boiler
5.2. DIGRAPH ANALYSIS
In the above digraph we can see that there are three 3-order negative loops and no positive
loops; hence the system is stable. The objective of the first controller (C1) is to maintain a
desired drum level. If the drum level increases, the controller in turn increases flow-rate
through the downcomer inevitably reducing the drum level. The objective of the other two
controllers C2 and C3 is to maintain safe pressure inside the drum. The manipulated variable
in case of C2 and C3 are steam flow rate (Ms) and heat radiation (Q) respectively. The total
volume of the system can be calculated by adding total volume of steam (Vst) and total
volume of water (Vwt). As the total volume of system remains constant, Vst and Vwt have
inverse relationship among them, i.e., if one increases then the other decreases.
43 | P a g e
A sudden positive deviation in feed water flow-rate (Mf) was assumed to test the propagation
of fault through the system. The digraph showed an increase in drum level (l) in the above
case. This increase is stabilized by controller C1. Increase in Mf also leads to increase in Vwt
which in turn causes a decrease in Vst. Now, if we consider a situation in which the output
steam flow rate Ms shows a sudden increase, then following back-tracing technique, we can
trace it to an increase in heat radiation (Q) through Vsd, αr and αv. Thus it is obvious how easy
determining the root cause of a fault becomes with the usage of digraph method.
The proposed effects of changes in Mf, Ms and Q on drum level (l) and drum pressure (P)
were verified by simulation. The graphs are provided below:
Step Response
Drum level (l)
0.04
0.02
Amplitude
0
Drum presure (p)
0
-0.01
-0.02
-0.03
-0.04
0
100
200
300
400
500
600
Time (sec)
Figure 5.3: Open-loop Response for step increase in Mf
In the digraph, there are three pathways from Mf to l:
(i)
Mf-- Vwd--l
(ii)
Mf-- Vwd--qdc--l
44 | P a g e
(iii) Mf-- Vwd--qdc--qr---l
Steps (i) and (iii) show an increase in l but (ii) shows a decrease. Thus we again encountered
contradictory pathways. From the graph though we can see that path (i) and (iii) dominate as l
increases with increase in Mf. Drum pressure P invariable decreases with an increase in Mf.
Step Response
Amplitude
Drum level (l)
0.02
0
-0.02
Drum pressure (p)
-0.04
0
-0.02
-0.04
-0.06
0
100
200
300
400
500
600
Time (sec)
Figure 5.4: Open-loop Response for step increase in Ms
The behaviour of l and P due to step increase in Ms is shown both by digraph and simulation
graphs and both the results corroborate each other. Both P and l decrease due to an increase in
Ms.
45 | P a g e
-9
Amplitude
Drum level (l)
0
Step Response
x 10
-2
-4
-6
Drum pressure (p)
-8 x 10-8
4
3
2
1
0
0
100
200
300
400
500
600
Time (sec)
Figure 5.5: Open-loop response for step increase in Q
The increase in heat radiation Q brings about an increase in both l and P. This result is shown
in both digraph and figure 5.5. But an extra knowledge that can be derived from the graph is
that upon increasing Q, l first decreases then increases. This may happen because as Q is
increases, Vsd increases. As more steam enters the drum, more is taken out to reduce pressure
and thus initially the drum level drops. This behaviour cannot be shown by the digraph. All
these graphs were plotted for open loop system, i.e., without the use of controllers. Figure
5.6. shows the behaviour of l and P under linear quadratic tracking (LQT) controller action.
46 | P a g e
Step Response
Amplitude
Drum level (l)
1
0.5
Drum pressure (p)
0
0.8
0.6
0.4
0.2
0
0
1
2
3
Time (sec)
4
5
6
7
5
x 10
Figure 5.6: Closed-loop response for a step increase in Q
47 | P a g e
CHAPTER 6
CONCLUSION AND RECOMMENDATION
Possible future work
48 | P a g e
6.1 CONCLUSION AND RECOMMENDATION
Advantage of digraph method:
(i)
The digraph method discusses in detail the various possible faults in a system and
the different paths via which it can be propagated, thus giving an exhaustive
analysis of the system.
(ii)
Simulation results help determine the dominant pathway in case of conflicting
effects of other variables on the monitored variable.
(iii)
Any arbitrary fault can be traced back to its root cause within no matter of time.
(iv)
Digraph, once fully developed and validated, is so easy to study that even an
ordinary worker without any special education or technological degree can carry
out the task of fault detection and analysis.
Disadvantages of digraph method:
(i)
It is a qualitative analysis. The deviations in variables are assigned qualitative
states of high (+) and low (-). The actual quantity of increase and decrease cannot
be determined.
(ii)
The digraph is developed by knowledge of fundamentals and human reasoning; it
is susceptible to errors and need many rounds of checking before actual
implementation.
These are the major two drawbacks due to which digraph method is still in laboratory scale
and has not been practically implemented in any industry.
49 | P a g e
Since 1990s, researchers have been discussing and experimenting the use of fuzzy logic and
fuzzy set theory to improve the diagnostic resolution in SDG-model based approaches. Inclusion
of quantitative information may also help in better understanding of the dynamic working of the
system and false indication of faults may be reduced. The use of fuzzy logic to help address the
problem of setting the threshold (limiting value) of alarm was discussed by Han et al. [9]. In their
approach, after the most probable root causes which are the possible fault origins are located,
fuzzy logic is introduced. Based on their membership degree of the origin nodes, variables are
arranged in an increasing manner and the fault origins having highest memberships are located.
This approach was shown to improve the accuracy of diagnosis resolution. Shih and Lee [10, 11]
discussed the removal of bogus solutions using fuzzy logic principles with SDGs, and the Fuzzy
Cause-Effect Digraph was proposed. The faulty interpretation of system due to presence of
compensatory response (CR) and inverse response (IR) from backward loops and forward paths
in the process, have been eliminated. Furthermore, this method also can estimate the state of the
unmeasured variables, to explain fault propagation paths and to ascertain origins.
50 | P a g e
REFERENCES
[1] Batch Fermentation: Modeling, Monitoring and Control by Ali Cinair, Sarish J. Parulekar,
Cenk Undey and Gulnur Birol
[2] E.M Kelly and L.M. Barlett; Application of the Digraph Method in System Fault Diagnostics
[3] WANG Hangzhou et al; A Signed Digraphs Based Method for Detecting Inherently
UnsafeFactors of Chemical Process at Conceptual Design Stage; Chinese Journal of Chemical
Engineering, 16(1) 52—56 (2008)
[4] Shi, Y., Qiu, T., Chen, B.Z., ―Fault analysis using process signed directed graph model‖,
Chem. Ind. Eng. Prog., 25 (12), 1484-1488(2006).
[5]Liqiang Wang; On-line Fault Diagnosis Using Signed Digraphs; Masters’ Thesis, Tiajin
University, China, 1997
[6] AtallaSayda; ―A benchmark model of a jacketed stirred tank heater for fault detection and
isolation‖.Technical report, University of New Brunswick, Fredericton, NB, Canada, May 2004.
[7] Process Modelling, Simulation and Control by William L. Luyben
[8] Damarla, K. Sheshu, Kundu, Madhusree; ―Monitoring of drum-boiler process using statistical
techniques‖, International Journal of Chemical Engineering and Applications, Vol. 2 , No. 3 ,
June 2011
[9] Chung-cheng Han, Ruey-fu Shih, and Liang-sun Lee. Quantifying signed direct graphs with
the fuzzy set for fault diagnosis resolution improvement. Industrial and Engineering Chemistry
Research, 33(8): 1943-1954, August 1994.
[10] Ruey-Fu Shih and Liang-Sun Lee. Use of fuzzy cause – effect digraph for resolution fault
diagnosis for process plants. 1. Fuzzy cause- effect digraph. Industrial and Engineering
Chemistry Research, 34(5): 1688-1702, May 1995.
[11] Ruey-Fu Shih and Liang-Sun Lee. Use of fuzzy cause- effect digraph for resolution fault
diagnosis for process plants. 2. Diagnostic algorithm and applications. Industrial and Engineering
Chemistry Research, 34(5): 1688-1702, May 1995
51 | P a g e
NOMENCLATURE
Chapter 3
Fo = Inlet mass flow rate (l/s)
Cao = Inlet feed concentration ( mol/l)
To = Inlet fluid temperature (°C)
V = Volume of reacting solution inside the tank (l)
T = Reacting solution temperature (°C)
Ca = Reacting solution concentration (mol/l)
F = Output mass flow rate (l/s)
Vj = Volume of the jacket (l)
Fj = mass flow rate of coolant (l/s)
Tjo = Inlet temperature of coolant (°C)
Tj = Outle temperature of coolant (°C)
k = Specific rate constant
n = Order of reaction
U = Overall heat transfer co-efficient (W/
°C)
A = Area of heat transfer between tank and jacket (
)
h = Enthalpy of reacting solution (J/kg)
ho = Enthalpy of inlet feed (J/kg)
hjo = Enthalpy of inlet coolant (J/kg)
hj = Enthalpy of outlet coolant (J/kg)
52 | P a g e
Kv = Controller gain
ρ = Density (kg/
)
= Density of coolant (kg/
)
Chapter 4
N = Number of trays
i = Used to refer a particular number
L = Liquid flow rate from trays (l/s)
Ff = Feed flow rate (l/s)
Fd = Distillate flow rate (l/s)
Frd = Reflux flow rate (l/s)
Fb = Bottom product flow rate (l/s)
V = Molal vapor flow rate (mol/s)
M = Liquid hold-up on trays (l)
Mrd = Reflux drum hold-up (l)
Mb = Column bottom hold-up (l)
x = Mole fraction of more volatile component in liquid phase
y = Mole fraction of more volatile component in vapour phase
xf = Mole fraction of more volatile component in feed
xd = Mole fraction of more volatile component in distillate
xb = Mole fraction of more volatile component in bottom product
yb = Mole fraction of more volatile component in vapor phase at column bottom
α = Relative volatility
53 | P a g e
Chapter 5
Mf = Mass flow rate of water into drum (kg/s)
Ms = Mass flow rate of steam from drum to outside (kg/s)
P = Pressure inside the drum (Pa)
l = Drum level (m)
Vst = Total volume of steam (
)
Vwt = Total volume of water (
)
Vsd = Volume of steam in the drum
Vwd = Volume of water in the drum
Q = Amount of heat irradiated (J)
hf = Enthalpy of feed water (J/kg)
hs = Enthalpy of steam (J/kg)
= Density of steam (kg/
)
= Density of water (kg/
)
Vr = Volume of the riser (
)
αr = Mass fraction of steam
αv = Volume fraction of steam
qdc = Mass flow rate through the downcomer (kg/s)
qr = Mass flow rate through the riser (kg/s)
mt = Mass of metal (kg)
Cp = Specific heat capacity (J/kg°C)
tm = Average change in temperature of metal (°C)
ts = Change in temperature of metal in riser (°C)
54 | P a g e
Lr = Length of riser section (m)
Ldc = Length of downcomer section (m)
g = Acceleration due to gravity (m/ )
K = Coefficient for frictional loss
Adc = Cross-sectional area of downcomer (
Ad = Area of drum (
)
)
55 | P a g e
Was this manual useful for you? yes no
Thank you for your participation!

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

Related manuals

Download PDF

advertisement