FAULT DETECTION, ACCOMODATION AND ISOLATION
FOR PAPER MILL INDUSTRIAL CONTROL
PREM KUMAR A/L APASAMY
UNIVERSITI TEKNOLOGI MALAYSIA
2
DEDICATION
For the loved Father and Mother
3
ACKNOWLEDGEMENT
My special thanks to Dr. Mohd Fauzi Othman for this guide and
supervise along this research study.
Special appreciation to Malaysian Newsprints Industries Sdn .Bhd
(Maintenance Department) stuffs for the cooperation through out the research.
And also not forget those who directly and indirectly help success
this research program.
4
ABSTRAK
Kerosakan instrumen di industri adalah gangguan untuk pengeluaran
industri. In bukan sahaja gangguan kepada pengeluaran malah kepada
pembaikpulihan. Gangguan in menyebabkan kehilangan pengeluaran and masa
pengeluaran. Industri sedang berkembang pesat dan teknik baru diperlukan
untuk mengesan kerosakan instrumen lebih awal sebelum ia rosak and
menganggu
pengeluaran
industri.
Teknik
tradisional
iaitu
‘Predictive
Maintenance’ dan ‘Run to Fail’ tidak mempunyai ruang untuk peningkatan dari
segi pengesanan kerosakan instrumen .Banyak kajian telah dilakukan dalam
topik pengesanan kerosakan instrumen dan teknik baru Kecerdikan Palsuan (
Artificial Intelligent ) digunakan dalam pengesanan kerosakan instrumen.
Dengan mengunakan parameter dan pemalar dari prosess dan instrumen, AI
boleh di bina dan digunakan untuk melihat kebolehpercayaan teknik AI dalam
pengesanan kerosakan instrumen. Kebolehan pengesanan teknik AI digunakan
untuk mengukur kebolehan pengesanan kerosakan instrumen.
5
ABSTRACT
Instrument fault in industries is an interruption to industries production
and maintenance. Its cause loses of production and time .As industries growing
faster and rapidly, there are a need of method to detect the fault of instrument
before its can disturb the production of industries. Conventional technique such
as Predictive Maintenance and Run to Fail not did not give much room in term
of predictability of instrument fault. Many study and research done on the fault
detection technique and new technique Artificial Intelligent technique (AI)
application in fault detection be come one of major studies in research. Using
the parameter and variable from process and instruments, AI can be constructed
and applied to view the reliability of the AI technique in the fault detection
method. The predictability of AI will be measure as performance indicator of
fault detection.
6
CONTENT
CONTENT
CHAPTER
1
PAGE
Declaration
1
Dedication
2
Acknowledgement
3
Abstrak
4
Abstract
5
CONTENT
PAGE
INTRODUCTION
13
1.0
Problem Statement
13
1.1
Objective
15
1.2
Scope Of Study
16
1.3
Thesis Outline
17
1.3.1
Chapter 1 – Introduction
17
1.3.2
Chapter 2 – Fault Detection of Servo
17
Valve in Paper Machine
1.3.3
Chapter 3 – Literature Review
17
1.3.4
Chapter 4 – Methodology
18
1.3.5
Chapter 5 – Result
18
1.3.6
Chapter 6 – Conclusion
18
7
2
FAULT DETECTION OF SERVO
19
VALVE IN PAPER MACHINE
2.0
Introduction To Servo valve and Paper Machine
19
2.1
Overview Paper Making Process
21
2.2
Back ground of Paper Machine and Servo valve
23
2.3
Nip Control
23
2.4
De-watering in the press section
23
2.5
Fault Detection Of Servo Valve
26
In Paper Machine Industries
2.5.1
Problem And Effect Of Servo
26
Valve In Paper Machine
2.5.2
3
Problem Of Servo Valve In Paper Machine
28
LITERATURE REVIEW
30
3.0
Method of solution for fault detection in process control.
30
3.1
Method of solution in similar research
33
RESEARCH METHODOLOGY
36
4.0
Methodology
36
4.1
Data and parameter acquisition from the real
37
4
plant or system
4.2
From Paper Making Manual
39
4.3
Knowledge from Instrument Personal
40
4.4
Introduction to Fuzzy Logic and Neuro fuzzy
40
4.4.0
43
Part In This Research
8
4.4.1
Fuzzy logic model
44
4.4.2
The Modeling Mamdani
44
4.4.3
Neuro Fuzzy Model
45
4.4.4
The Modeling Sugeno
45
4.4.5
The construction of Sugeno model in Mat lab
45
4.4.6
The construction of Mamdani model in Mat lab
61
4.4.7
The construction of Early Prediction model
73
in Mat lab
5
RESULTS
70
5.0
Result of Neuro Fuzzy
75
5.1
Result of Mamdani
83
5.2
Result of Prediction
85
5.3
Analysis
95
5.3.1
Fuzzy Logic
95
5.3.2
Neuro Fuzzy
96
5.3.3
Neuro Network
96
CONCLUSION
97
6
6.0
What have been done during this research
98
6.1
What have been gains from this research
99
6.2
Recommendation
101
6.3
Reference literature
102
9
TABLE LIST
TABLE NUMBER
TITLE
PAGE
Table 2.6
Syndromes of servo valve abnormal in paper making
27
Table 2.7
Common fault of servo valve.
29
Table 4.5
Output Assigned Table
46
Table 4.6
Data to Mat lab
46
Table 4.22
List of fault from Database report
62
Table 4.23
List of fault from Analysis
60
Table 4.24
Constructed rules base on knowledge of fault analysis
63
10
FIGURE LIST
FIGURE NUMBER
TITLE
PAGE
Figure 2.1
Paper Machine
21
Figure 2.2
Press Section
22
Figure 2.3
Nip between roller with piston
24
Figure 2.4
Hydraulically pressed shoe
25
Figure 2.5
Servo Valve
25
Figure 4.0
Fault detection research methodology
37
Figure 4.1
Record database
38
Figure 4.2
Darwin system
39
Figure 4.3
Fuzzy logic as in Mamdani concept
44
Figure 4.4
Block diagram of TS Fuzzy Modeling
45
Figure 4.7
Loaded data
47
Figure 4.8
Selection of Membership function
48
Figure 4.9
Trained data
49
Figure 4.10
The generated Sugeno model
50
Figure 4.11
The input membership of input 1
51
Figure 4.12
The input membership of input 2
52
Figure 4.13
The input membership of input 3
53
Figure 4.14
The output membership model generated
54
Figure 4.15
The output model rules generated
55
Figure 4.16
The output model rules generated in Rules Editor
56
Figure 4.17
The generated Sugeno Model Structure
57
11
Figure 4.18
The generated Sugeno Model from Surface viewer
58
Figure 4.19
The command to test the output of generated
59
Sugeno model
Figure 4.20
The tested output generated in Mat Lab
60
Command Window
Figure 4.21
Database report which will be report by
61
personal involve in troubleshoot
Figure 4.25
Construction of the Mamdani model
64
Figure 4.26
Membership function for set point deviate from
65
Measure value in the Mamdani model
Figure 4.27
Membership function for Measure value offset
66
set point from Mamdani model
Figure 4.28
Membership function for output for electronic
67
card fault from Mamdani model
Figure 4.29
Membership function for output for filter blockage
68
in Mamdani model
Figure 4.30
Membership function for output for o ring breakage in
69
Mamdani model
Figure 4.31
Apply the constructed Rules from Figure 4.24
70
into the Rules Editor in Mamdani model
Figure 4.32
Ruler Editor generated by Mamdani model
71
Figure 4.33
Ruler Surface generated by Mamdani model
72
Figure 5.1
The output membership of Neuro Fuzzy
75
model generated
Figure 5.2
The output Neuro Fuzzy model rules generated
76
Figure 5.3
The output Neuro Fuzzy model rules generated
77
in Rules Editor
Figure 5.4
The generated Sugeno Model Structure
78
Figure 5.5
The generated Sugeno Model from Surface viewer
79
Figure 5.6
The command to test the output of generated
80
Sugeno model
12
Figure 5.7
Output generated by Sugeno model
81
Figure 5.8
To check with the output generated by Sugeno model
82
Figure 5.9
Output of Sugeno model rules
83
Figure 5.10
Output of Sugeno model rules
84
Figure 5.11
Comparison predicted data and actual data
85
Figure 5.12
Prediction Error of the neuro network
86
Figure 5.13
Comparison predicted data and actual data
87
Figure 5.14
Prediction Error of the neuro Network
88
Figure 5.15
Comparison predicted data and actual data
89
Figure 5.16
Prediction Error of the Neuro Network
90
Figure 5.17
Comparison predicted data and actual data
91
Figure 5.18
Prediction Error of the neuro Network
92
Figure 5.19
Comparison predicted data and actual data
93
Figure 5.20
Prediction Error of the neuro Network
94
13
CHAPTER 1
INTRODUCTION
1.0
Problem Statement
Recent development shows advancement in maintenance tool for
industries growth very fast. Many tool developed to help maintenance to carry
out corrective maintenance and the preventive maintenance. This concept of
corrective maintenance and preventive maintenance is highly used to make sure
the instruments used in production industry working well during production.
The next level need to be seen in this field to enable the instrument reliability
during production is high. One of current research that used is the use of AI
technique to identify the failing instrument before its disturbed the production.
This research regarding the application of Neuro fuzzy in fault
detection of fault in a paper mill process control. This research is to study and
develop an early fault detection system of instrument. The instrument selected
for this research is servo valve due to its critically use in paper mill.
Conventional method is the preventive maintenance where checklist
daily is used to see the functionality of the instrument in field. But the limitation
14
to this method is that analysis method is not good enough to see the future of the
instrument s condition.
The Neuro Fuzzy is an advance technique to analysis the instrument
condition and capable to view the future of this instrument. Data available from
servo valve itself output voltage, current & pressure, flow rate .Its comparable
with indirect pressure measure system (Darwin system). All captured data will
be applied in Neuro fuzzy for fault detection. Other parameter such as machine
speed, product grade will be taken into account. Using this data the fault
detection using Neuro Fuzzy can be develop and help the industrial to make
production more reliable.
1.1
Objective
15
The objective of this research is to study servo valve fault occur in
application in paper machine. Using the information from plant, observe the
pattern and symptoms of failure in servo valve system in act. And study the
method to apply the fault detection method in actual plant, By understand the
changes of parameter in process and its effect in servo valve system in real
application and the change that made by servo valve fault, servo valve fault
detection using the Artificial Intellegent can be constructed .The constructed
fault detection system will be analysis the performances of it and if given
opportunity will be tested in lab or simulation.
1.2
Scope Of Study
16
The scope of research covers wide area of mechanical, electricelectronic, and instrument. Using proper guide, this can be narrow down to
specific range of study. Study the servo valve system in paper mill application
in term of mechanically ,electrically and process control .The literature studied
is to understand the concept of the fault detection in process control and studied
the proposed method for the fault detection . All the data regarding the servo
valve relevant to study collected and analyzed. Analysis the related data and
check the suitable Neuro Fuzzy method to be use in research. Develop the AI
algorithm for the Neuro Fuzzy method and tested by simulation .
1.3
Thesis Outline
17
1.3.1
Chapter 1 – Introduction
In this chapter the introduction to on the fault detection concept in
industries. Also elaborated the problem that exist in current industries
instrument maintenance area. The objective of the research and the scope of
study involve in this research.
1.3.2
Chapter 2 – Fault Detection Of Servo Valve In Paper Machine
Here the overview of process area with instrument involve in this
research will be introduced. The problem – effect involve process and
instrument will be explain in detail.
1.3.3
Chapter 3 – Literature Review
Here the literature, journal and manual that were studied to guide this
research done. There’re some literature that involve indirectly with this
research, which give useful information for research.
1.3.4
Chapter 4 – Methodology
18
In this chapter, the step that have take to complete this research is
elaborated. The introduction to the Artificial Intellegent technique that going to
be constructed will be explained. And how this Artificial Intellegent technique
is implement in Mat Lab will be explained. The construction of the Artificial
Intellegent technique will be shown step by step.
1.3.5
Chapter 5 – Result
In this chapter the output of the constructed Fuzzy Logic, Neuro –
Fuzzy and Neuro Network will be shown and explained .The design will be
simulated in Mat Lab software and the output of simulation will be shown here.
1.3.6
Chapter 6 – Conclusion
Here the overall explanations of the research given. What have been
done throughout the research and what has been gain by this research also
elaborated in this part.
19
CHAPTER 2
FAULT DETECTION OF SERVO VALVE IN PAPER MACHINE
2.0
Introduction To Servo valve and Paper Machine
The rapid growing industrial edge just have came by in our world .So
much of advance in engineering and technology have help boost up the
production speed and capacity of industry .In such world, the need to new
invention and innovation is much need .The industrial have become bigger and
faster. To aid the industry, a lot of research have going on .On of the most rapid
growing area in industry, the defect of instrument and equipment can be find to
interrupt the production of industry. Even in the most advance equipment,
failure of instrument and equipment experienced, and cause a lot of unnecessary
interrupt for production business .The common word used to describe the lost of
production is known as the Down Time. The industry will always try to
minimize this Down Time.
20
Traditionally, there is little method, which used to prevent or minimize
the effect of instrument fault to the process industry. The most common as
Predictive Maintenance or in short for is known as the PVM and also Run To
Fail. The technical personal will go to and check on regular basis while taking
some data regarding the equipment condition, such as Gauge reading, current
reading and other. But often the technique is does not give a very good outcome
because in use human where each personal might have different judgments on
the condition of equipment while inspection carry out and human have too
many variable which effecting their judgments. On other hand technique will
help the technical personal to detect the symptoms of equipment failure by
visual inspection and measurement off line when any abnormal finding is found.
This technique does help to detect failure of equipment which cause sometime
by external factors such temperature and vibration. But this technique does not
help for online monitoring and inspection because human will get very fast tired
if they were as to continuously monitoring and become bored very fast. But
using human intelligent where any kind of change in the signal, we as human
find it’s as easy to tell something is fault from our experience. But machine is
unable to learn as human and don’t know what is wrong and normal. But using
both of human ability to learn and machine, which have longer working hours,
we can construct a system, which capable learn and react as human. This
technique is known as the Artificial Intelligent. There are few Artificial
Intelligent technique can be used to make a fault detection system. In this
research the Fuzzy Logic, Neuro –Fuzzy and Neuro Network is introduced for
the fault detection research.
The use of AI technique is not something new in current technology.
In the paper machine the use of servo valve was discussed as in the early
chapter .The servo valve use in paper machine is to control the gap between two
rollers with the actuation power of piston.
21
2.1
Overview Paper Making Process
Figure 2.1 Paper Machine
[12] Processes, which involve the making of dry paper in sheet, form
using wood fiber or recycle paper. Supplying the stock will make paper or raw
material and paper formation will take place in Paper Machine. The Paper
Machine is a very large piece of machinery. A typical machine is about the
length of two football pitches and around 4 meters wide. It can run up to speeds
of 2000 m per minute - or 60 miles per hour! The machine itself consists of 7
distinct sections. The head box, wire, press section (As in Figure 2.2), dryer
section, calendar and reel section. The first section of the machine is called the
'Wet End'. This is where the diluted stock first comes into contact with the paper
machine. It is poured onto the machine by the head box, which is a collecting
box for the dilute paper stock. A narrow aperture running across the width of the
head box allows the stock to flow onto the wire with the fibers distributed
evenly over the whole width of the paper machine. The machine is operated by
computer control such as DCS (Distributed Control system). The computer will
22
monitor the paper for moisture content, weight etc and computer screens will
show pictures of the process and should any adjustments need to be made, an
alarm will sound.
Figure 2.2 Press Section
23
2.2
Back ground of Paper Machine and Servo valve
Servo valve use in paper making machine to control the gap (nip)
between 2 rollers which dewatering paper run. Refer Figure 2.3 .The example
paper machine in Malaysia is in MALAYSIAN NEWS
PRINTS
INDUSTRIES SDN.BHD (MNI) located in Temerloh, Pahang. The type Paper
Machine use is the VOITH SULZER from Germany. The servo valve is used in
wet end area to control the formation of paper between 2 roll. This Servo valve
is a critical instrument where the failure of it can cause down time to the
machine. Machine down mean lost of production time and cost. On this
research, the fault detection applied using neuro - fuzzy techniques on to this
servo valve system in order to find the fault earlier. This fault detection method
not only can be use to reduce down time but also give maintenance people
information which can be analysis to make this system more reliable and better.
This project also allows more room for the research in fault detection in process
control.
2.3
Nip Control
Servo valve use in paper making machine to control piston, which set
the gap (nip) between 2 rollers, which run dewatering in papermaking. 3 servo
valve used to control the bottom roller nip pressure line in paper making
machine.
24
The picture shows the overview of servo valve application in wet end in
papermaking. This is the control loop to control the nip (gap between 2 roller).
Paper with
from
sectio
Top
Wet
Paper Dewatered
Piston
Bottom
Figure 2.3 Nip between roller with piston
2.4
De-watering in the press section
[11] After formation of the sheet, a process which determines the most
important sheet properties, the paper sheet has to be further drained and
compressed. In this next phase, mechanical pressure exerted vertically to the
sheet surface is used to further increase the proportion of dry content. In the
press section, the web runs between a series of rolls, which exert specifically set
amounts of pressure. The water pressed out of the paper is absorbed by felts and
transported off. Refer Figure 2.4
In recent years, shoe presses have been developed to increase the
efficiency of the traditional roll presses. In these press units, one of the rolls is
replaced by a hydraulically pressed shoe. This creates a bigger press nip, which
makes the process more effective.
25
Figure 2.4: Hydraulically pressed shoe
Figure 2.5: Servo Valve
26
2.5
Fault Detection Of Servo Valve In Paper Machine Industries
2.5.1
Problem And Effect Of Servo Valve In Paper Machine
In paper making ,when the raw material known as stock which contain
fiber and water have to pass thought nip between 2 roller , top and bottom in
wet end press section. This process to run the dewatering process in paper. This
is a part of paper formation. Dewatering is a process where water will be
removed from formed paper stage by stage in press section. Paper will pass
through 1st press, 2nd press, 3rd press and so on to 4th press to run through the de
watering process each time pass the presses nip. Later paper will lost more
water and moisture drastically when run thought the dryer section .Here the fine
tuning of the moisture level in paper by control loop system .Because the speed
of 2 roll very high and the gap nip size is very small, a very précised of control
gap distance required.
And due to the size and weight of rollers which is very large compared
to the gap nip ,a high actuation power required to actuate the moving part of the
piston which control the rollers distance .Due to a very high vibration and noise
environment ,the best control for this nip control is the hydraulic pressure which
move the piston. This piston will be move as precise as possible because the use
of hydraulics or oil .Air or pneumatic can not give good control due air can be
compress and unable to control for very small displacement and its depend on
the temperature as well.
The servo valve will control the hydraulic pressure supply to the
piston and control this piston movement. This gives smooth movement and,
27
controllable displacement ,will the properties of oil is independent of external
factor .
Why use servo valve, its because of its robustness in industry
application special for in heavy industries. The reliability and high durability is
come with the price of servo valve. It s consume very little power and able to
control in very fine measure of hydraulic pressure. Precision and accuracy of
control is one of important reason, why its have been choose in paper mill
control system. Its requires very little maintenance and its universal control
where its can be control using 4 - 20 mA either from DCS or PLC or any other
common control system and tools.
The failure of servo valve function will cause upset in paper formation.
Possibility of paper break happen in presses section is high. The other effect of
servo valve fault to process paper making as below :
Table 2.6 : Syndrome of servo valve abnormal in paper making:
No Syndrome of servo valve abnormal in paper making:
1
Number of paper break high at wet end press section
2
Paper wrinkle at in some part on paper
3
Profile nip balance
4
Dewatering unbalance across roll
5
Unbalance moisture profile
6
Felt condition abnormal
7
Felt life span reduce
8
Paper formation unstable
9
Moisture control unstable
10
Steam control on dryer will change
28
2.5.2
Problem Of Servo Valve In Paper Machine
The servo valve is an very robust instrument, but its still will fault due
its have an electronic card which to receives signal from DCS or PLC .Its have
oil filter internally where filter the oil that passing inside and be control to the
piston. The oil circulation from oil tank to pump to servo valve and to piston
again back to tank. This oil circulation will carry not only carry oil but also
sometime particles such as rust, mud and other unwanted element in oil
pipeline. This sometime will cause blockage to the oil line. And cause the
internal filter in servo valve coated. But however if the hydraulic line is
monitored frequently, the possibility of oil contamination can be avoided.
The critical issue or problem of servo valve fault is when its internal
card fault. Because it’s happen without any early sign that can be recognize by
normal technical personal. Visual inspect is impossible. The failure due to
electronic card is sometime very difficult to diagnosis compared to other type of
fault. There in no indication in DCS or in servo valve itself. Its worse case if it’s
an intermitted problem where sometime its give correct and incorrect reading
.In some paper mill to detect the faulty of servo valve require and high sampling
time .This is due to the DCS and PLC sampling time very slow and does not
shows actual reading to operation or maintenance .
In my research the servo valve data available in other resources ,data
were collected from Darwin system, a system which used to monitor the
performance servo valve with high sampling time. This give more accurate data
regarding the symptoms that happen to the servo valve. Common fault of servo
valve shown in table 2.7.
29
Table 2.7: Common fault of servo valve.
Internal cause servo valve fault External cause servo valve fault
Electronic card
Hydraulic oil properties
Cable connector
Hydraulic oil temperatures properties
o - ring
Surrounding vibration
Filter
DCS system
PLC system
Hydraulic pump
Pipeline condition
There is some external factor that cause the servo valve control
abnormal such as the oil pump which give the inlet pressure to servo valve and
the oil properties. These symptoms also will be show in DCS or other system,
which look as the servo valve fault. Fault of servo valve can be detect by the
machine when upset happen to paper or abnormal changes happen to the paper.
30
CHAPTER 3
LITERATURE REVIEW
There are some research that have been done as in related to this
research thesis Fault detection and Isolation Paper Machine Instrument. There
are suggested method which can be see from the literature that been studied for
this research on fault detection and isolation in paper machine instrument. Its
have guide the research process directly or in directly.
3.0
Method of solution for fault detection in process control.
Here s the general literature study which related to the research.
[6] Leon Reznik and Vladik Kreinovichis, studying a feasibility of fuzzy models
application in measurement procedures .The problem measurement data fusion
,when the sources can provides prediction or combination sources .The data
applied to fuzzy prediction models and been observed the outcome.
31
[7] Rajiv Sreedhar and Benito Fernandez. In his literature have presented the
preliminary results regarding the design of neural network based adaptive fault
detection method. The error between the actual system and model system will
be applied to the observer to obtain accurate estimates of states or output.
[8] Tracy Dalton in his literature, mentioning the type of fault, and method of
diagnosis that can have been done . Its give more detail regarding fault detection
method in general.
[9] Reza Asgary and Karim Mohammadi proposed the use a probabilistic
neural network for fault detection in MEMS .A fuzzy system is implement to
improve performance of the network.
[12]Erik Dahlquist ,Tomas Lindberg from Malardalen University in Integrated
Process control, fault diagnostics ,optimization and production planning –
Industrial , literatures the different types of diagnostics with respect to control
sensors ,control loop and processes .
[13]A.M.Bhagwat ,R.Srinivasan and P.R.Krishnaswamy
from National
University of Singapore ,fault detection approaches during transition based on a
nonlinear process model ,and operating procedures .
[14]Bruno HEIM, Sylvie CAUVIN, Sylviane GENTIL presents the diagnostics
system with output on real data scenario and its tested on line on a FCC (Fluid
Catalytic Cracking) pilot plant.
32
[15]Pascal Dufour ,Sharad Bhartiya ,Thomas J.English ,Edward P.Gatzke
,Prasad S .Dhurjati and Francis J.Doyle 3 , shows the diagnostics for fault
detection and isolation in a continuous pulp digester with comparing for
accuracy and real time data .
[16]Dauren F .Akhmetov ,Tomofumi Miura ,Masataka Itou , Manabu Tamamori
,Hiroto Murabayashi and Yuichi Hamade attempts the intelligent system for
unified point for learning process optimization .Also practical implementation
of faults detection is demonstrated .
[17] B.Jiang ,M.Staroswiecki and V.Cocquempot in paper discuss the fault
diagnosis problem for nonlinear system with unknown parameters .Discussion
on Strictly Positive Realness requirement and extension to fault cases .The DC
motor will be experimented here.
[18] Sheng Yung Chang, Cheng Ren Lin and Chuei Tin Chang in paper mention
the Signed Directed Graphs (SDG) and Fuzzy Interference techniques for fault
diagnosis. The fault derived from SDG model and encoded ,interfaced with
fuzzy rules .
[19] Piotr Kulczycki , used statistical interference system for fault detection and
diagnosis also prediction . Using kernel estimator techniques, an output wide
range of application will be produced.
[20] Bentiz – Perez H and Rendon – Torres P approach Fuzzy ART2 network
to study the on - line for the pattern recognition rather than faults detection.
33
[21] In Soo Lee ,Doug Hyung Lee ,Euy Soo Lee ,Sang Jin Park a predictive
method of identifying the feed and product using neural network based software
is proposed .Data preprocessing and post processing methods are utilized to
enhance the predictability of the software sensor model.
Although many research have been studied, few paper was view as
to related to this research study. Its give some guide for this research directly
and indirectly .
3.1
Method of solution in similar research
Here some of the principles from literature is observed for this
research.
[1] Rolf Isermann and Mihaela Ulteru, proposed the unified frame for
integration of different fault detection and decision techniques via approximate
reasoning on analytical and heuristic fuzzified symptoms.
[2] Rolf Isermann have elaborate regard the fault detection method ,fault
diagnosis as using knowledge based fault detection which divided to analytical
symptom generation and heuristic symptom generation, and knowledge based
fault diagnosis . Model based fault detection and its method has been drop
written here. Rolf Isermann mention fault diagnosis method, which used to
34
identify the fault, was based on the observed analytical and heuristic symptoms
and heuristic knowledge on process.
[3] Dr.Rahmat Shoureshi and team member ,s have proposed on line equipment
diagnosis using artificial intelligent and non linear observer to prevent
catastrophic failures I substation equipment ,and aid the preventive maintenance
.
[4] L.B.Palma ,F.J.Coito ,R.N.Silva ,shows the online model based procedure
for fault detection and diagnosis of sensor and process faults in plant. The
method was discuss here is based on parity equations and no an adaptive
observer based approach. The researcher propose use of dynamic features of
black box (Auto Regressive with eXogenous input - ARX)
[5] Alessandro De Carli ,Rccardo Cirillo applied the latent faults prevision
applied to a component of an automatic plant glass industry. Here a clear picture
of Fuzzy logic applied using the Mamdani Method. Thought the construction of
linguistic rules ,fuzzy logic gives mathematical instrument to transpose the
typical nuances of person ,s reasoning and experience in a precise and strict
treatment ,fuzzy logic is also particularly useful in research .
[10] R.A Reese, In “Press Operations “ literature have describe the nip function
and the important of nip control in paper making process.
35
[11] Venkat Venkatasubramaniam ,In Process fault detection and diagnosis
:Past ,Present and Future proposed the various approaches to fault diagnosis and
the challenges in this field .
36
CHAPTER 4
RESEARCH METHODOLOGY
4.0
Methodology
In order to comply this research thesis, a proper step sequence have
been follow to keep the research on track with the objective and progress of this
project as refer in Figure 4.0 Fault detection research methodology
37
Data and parameter acquisition from the actual system
Selection of Neuro Fuzzy method based on input and output
Construct a Neuro Fuzzy algorithm
Simulate and train the Neuro Fuzzy algorithm if required
Test fault detection algorithm with actual future data
Figure 4.0: Fault detection research methodology
4.1
Data and parameter acquisition from the real plant or system
Here the data collected from actual plant. The data from DCS and other system
such Darwin was selected to be use for research. The target data for this
research was:
Servo valve parameter from DCS :
1. Calculated Pressure from PLC system
2. Outlet Pressure from Servo valve in DCS
3. 4-20mA output current from servo valve to DCS
4. Flow rate from internal flow meter in side servo valve itself
5. Out let pressure direct measure after the servo valve connected Darwin
system
But the data of servo valve fault case available in Darwin only. This is
due to DCS did not keep or record any data more that 2 days. Data from the
Darwin system available since year 2000. Date and time of previous failure
38
cases been trace and recorded in database as in Figure 4.1 Record database. The
previous cases date and time of have been recorded. Using Darwin system, data
been trace back from the similar date and time. Here only data from Darwin
which is the outlet pressure from servo valve. Other data that have collected and
studied .For the normal run or good condition running data available easily from
plant by daily observation or PVM (Predictive Maintenance) check list which
done by technical personal in regular basis. The Darwin system measure using
pressure transmitter at the out let of the servo valve as in Figure 4.2.
Figure 4.1: Record database
39
DCS
Darwin
Monitoring
System
Top
Roller
Darwin
Pressure
Hydraulic pump
Servo valve
Pisto
n
Bottom
Roller
Hydraulic tank
Figure 4.2: Darwin system
4.2
From Paper Making Manual
We can understand on nip control process from point of view control
process .The journal by R.A Reese, Senior Process Consultant CRS Sirrine
Engineers, Inc Greenville, SC .XI PRESSING OPERATIONS explains the
fundamental of nip control in papermaking, limitations to pressing ,press from
aspect mechanical ,type of presses ,sheet transfer from wire to press section
,type of press arrangements ,press configuration consideration ,press clothing
and press section variables .
Manual, MOOG, D656 PQ Proportional Control Valve by MOOG
GmbH, explain the construction of servo valve, and also the theoretical
calculation of the function of servo valve and the electrical signal control of
servo valve.
40
Transfer Function for MOOG Servo valves
by W.J. Thayer
,December 1958 Rev .January 1965 from Technical Bulletin 103 ,MOOG INC
,Control Division ,East Aurora ,NY .In this literature ,the fundamental
calculation of the control of the servo valve was reviewed.
4.3
Knowledge from Instrument Personal
The maintenance personal have kept record of symptom, which can
cause servo valve failure, and also the effect of servo valve to process
papermaking. This data recorded by personal involve directly in troubleshooting
the servo valve failure involved.
4.4
Introduction to Fuzzy Logic and Neuro fuzzy
Fuzzy logic had a rapid growth in the number and variety of
applications since few years .The applications range from consumer products
such as cameras, camcorders, washing machines, and microwave ovens to
industrial process control, medical instrumentation, decision-support systems,
and portfolio selection. The meaning of fuzzy logic is needed to be understand
for reasoning of fuzzy logic.
Fuzzy logic has two different ways to be interpreted. In straight words,
fuzzy logic is a logical system, which is an extension of multi valued logic. But
in a other word, its can mean in predominant use today, fuzzy logic (FL) is
almost synonymous with the theory of fuzzy sets, a theory which relates to
41
classes of objects with un accurate boundaries in which membership is a matter
of degree. Here its view as, fuzzy logic in its specific meaning is a branch of
FL. What is important is to identify it’s even in its specific meaning; the agenda
of fuzzy logic is very different both in spirit and substance from the agendas of
traditional multi valued logical systems.
In the Fuzzy Logic Toolbox, fuzzy logic should be interpreted as FL,
that is, fuzzy logic in its wide sense. What might be added is that the basic
concept underlying FL is that of a linguistic variable, that is, a variable whose
values are words rather than numbers. In effect, much of FL may be viewed as a
methodology for computing with words rather than numbers. Although words
are inherently less precise than numbers, their use is closer to human intuition.
Furthermore, computing with words exploits the tolerance for imprecision and
thereby lowers the cost of solution.
Another basic concept in FL, which plays a central role in most of its
applications, is that of a fuzzy if-then rule or, simply, fuzzy rule. Although rulebased systems have a long history of use in AI, what is missing in such systems
is a machinery for dealing with fuzzy consequents and/or fuzzy antecedents. In
fuzzy logic, this machinery is provided by what is called the calculus of fuzzy
rules.
A change that is visibility relates to the use of fuzzy logic in
combination with neuro computing and genetic algorithms. More generally,
fuzzy logic, neuro computing, and genetic algorithms may be viewed as the
principal constituents of what might be called soft computing. Unlike the
traditional, hard computing, soft computing is aimed at an accommodation with
the pervasive imprecision of the real world. The guiding principle of soft
42
computing is: Exploit the tolerance for imprecision, uncertainty, and partial
truth to achieve tractability, robustness, and low solution cost. In coming years,
soft computing is likely to play an increasingly important role in the conception
and design of systems who’s MIQ (Machine IQ) is much higher than that of
systems designed by conventional methods.
Among various combinations of methodologies in soft computing, the
one that has highest visibility at this juncture is that of fuzzy logic and neuro
computing, leading to so-called neuro-fuzzy systems. Within fuzzy logic, such
systems play a particularly important role in the induction of rules from
observations. An effective method developed by Dr. Roger Jang for this
purpose is called ANFIS (Adaptive Neuro-Fuzzy Inference System). This
method is an important component of the Fuzzy Logic Toolbox. Fuzzy logic is
all about the relative importance of precision. All books on fuzzy logic begin
with a few good quotes on this very topic, and this is no exception. Here is what
some people have said in the past.
Precision is not truth.
—Henri Matisse
Sometimes the more measurable drives out the most important.
—RenéDubos
Vagueness is no more to be done away with in the world of logic than friction
in mechanics.
—Charles Sanders Peirce
I believe that nothing is unconditionally true, and hence I am opposed to every
statement of positive truth and every man who makes it.
—H. L. Mencken
43
So far as the laws of mathematics refer to reality, they are not certain. And so
far as they are certain, they do not refer to reality.
—Albert Einstein
As complexity rises, precise statements lose meaning and meaningful
statements lose precision.
—LotfiZadeh
4.4.0 Part In This Research
There 3 major parts in this research:
1. Fault Detection - Neuro - Fuzzy –Sugeno Method
2. Fault Isolation – Fuzzy Logic - Mamdani Method
3. Early Prediction – Neuro network
There are 2 model need are to be constructed. There are:
1. Sugeno model - to train data and construct the model
2. Mamdani model – is the decision making model
3. Early Prediction – to detect symptoms of the change in signal
44
4.4.1
Fuzzy logic model
The fault isolation is the part fault detection system will Fuzzy logic
.The fuzzy logic the used to construct the Isolation model. This Isolation Model
is done using Mamdani structure concept using the Mat lab.
4.4.2
The Modeling Mamdani
The information from real plant will be used to construct the Rules.
This rule are base on the analyze data and experience from personal.
Case reported
People experiences
Fuzzy logic
(Fuzzification)
Normal run data
Theoretical/Analyzed
data
Figure 4.3: Fuzzy logic as in Mamdani concept
Output/conclusion/
Action
45
4.4.3
Neuro Fuzzy Model
Using all the data collected as been in the (4.1) data and parameter
acquisition from the real, model of system is to be constructed. The neuro fuzzy
will be used to construct model the system .The type of fuzzy system used in
this research is Takagi –Sugeno (TS) Fuzzy Model. Figure 4.4 shows the
general block diagram of TS Fuzzy Model.
Fuzzification
Layer
Inference Mechanism
De Fuzzification
Rule
Base
Figure 4.4: Block diagram of TS Fuzzy Modeling
4.4.4
The Modeling Sugeno
4.4.5
The construction of Sugeno model in Mat lab
The faulty signal captured during instrument fault will be presented a numerical
manner. The trend that capture by Darwin system will be converted to table
form .The 3 type fault signal, signal during the electronic card fail, filter
experience blockage and the O ring break will complied and the output is
categories for each fault signal.1 is the output assign for the O ring breakage,2 is
46
for the filter blockage ,and 3 is for electronic card fault. Assign output can view
from Table 4.5 as below
Table 4.5: Output Assigned Table
Faulty part in Servo
Symptom in Darwin Trend
Assign output
valve
for Modeling
O ring breakage
Reading fluctuate
1
Filter blockage
Reading Offset
2
Electronic card fault
Reading lost/lower the minimum
3
The example of actual data as below,
The Example of actual data as in table 4.6,
Table 4.6 Data to Mat lab
Fluactutate
16.46
16.21
16.46
15.39
16.23
16.34
Reading
Offset
13.87
13.87
13.87
13.87
13.87
13.87
Lost signal
13.87
13.87
13.87
13.87
13.87
13.87
Fault Type
O
O
O
O
O
O
ring break
ring break
ring break
ring break
ring break
ring break
47
Data will be load to Anfis and be trained in Mat lab.
>> Anfisedit
Figure 4.7: Loaded data
48
Figure 4.8: Selection of Membership function
49
Figure 4.9: Trained data
50
The model will generated by the Mat lab
Figure 4.10: The generated Sugeno model
51
The membership function generated by Mat lab can be seen as below:
Figure 4.11: The input membership of input 1
52
Figure 4.12: The input membership of input 2
53
Figure 4.13: The input membership of input 3
54
The output membership function is as below:
Figure 4.14: The output membership model generated
55
The rules from generated model :
Figure 4.15: The output model rules generated
56
The generated output model rules generated in Rules Editor
Figure 4.16: The output model rules generated in Rules Editor
57
The generated Sugeno Model Structure:
Figure 4.17: The generated Sugeno Model Structure
58
Figure 4.18: The generated Sugeno Model from Surface viewer
The model can be tested by running the command in Editor
a = readfis(‘the model name ’);
Evalfis([ X Y Z ],a)
Where :
X - INPUT1
Y – INPUT2
Z – INPUT3
a – output model
59
Figure 4.19: The command to test the output of generated Sugeno model
60
The tested output generated in Mat Lab Command Window
Figure 4.20: The tested output generated in Mat Lab Command Window
61
4.4.6 The construction of Mamdani model in Mat lab
The reported case in database will be listed and analysis one by one in order to
find the symptoms of fault during operation plant and trouble shooting step
taken by personal. From the gathered data,
Figure 4.21: Database report which will be report by personal involve in
troubleshoot
62
List will be as below,
Table
Date
03-Nov-05
05-Dec-05
27-Dec-05
23-Nov-05
29-Mar-05
16-Feb-05
01-Feb-05
27-Jul-06
13-Jan-06
05-Jan-06
4.22: List of fault from Database report
Possiblility fail
Symptom
Action
electronic card
reading fluctuate 10 bar
change servo
electronic card
reading fluctuate 8-30 bar (sp 17.5)
change servo
electronic card
feedback higher than setpoint
change servo
electronic card
reading output lost
change servo
reading did not change according change
to setpoint
servo
electronic card
fuse blown
reading from sv lost
replace new fuse
electronic card
reading fluctuate
change servo
loose mounting
leakage oil physical
retighten mount
unstable oil system reading fluctuate when ramping monitor system
reading intermitterly reading
high/lowcan not follow set point(unstable)
change servo
Compile and investigate more deep from the gather data, until the final
element of servo valve which fail recognized. The investigation until the
recognized fault can be correlated theoretically with failure part of servo valve .
Recognized fault can investigate more deep by run the Why - Why - Why
analysis in such as cases. The symptoms also can be under stand by understand
the design and structure of servo valve from manual.
One all the compiled the reasoning for servo valve failure can be as below in
Table 4.23: List of fault from Analysis. But for this thesis research will focus on
O ring, Filter and Electronic card problem as was record this type fault happen
twice more of the servo valve operation life in the plant.
63
Table 4.23: List of fault from Analysis
Faulty part in Servo
Symptom in Darwin Trend
valve
O ring breakage
Reading fluctuate from set
point
Filter blockage
Reading Offset from set
point
Electronic card fault
Reading lost/lower the
minimum
Construct the rules from the analysis data and concept of the servo valve
application.
No
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Deviation
Offset
Signal Lost Electronic cardFilter
negative low deviate negative low have signal ok
block
negative low have signal ok
block
normal
positive high deviate negative low have signal ok
block
negative low deviate normal
have signal ok
ok
normal
normal
have signal ok
ok
positive high deviate normal
have signal ok
ok
negative low deviate positive highhave signal ok
leak
positive highhave signal ok
leak
normal
positive high deviate positive highhave signal ok
leak
negative low deviate negative low no signal fail
ok
negative low no signal fail
ok
normal
positive high deviate negative low no signal fail
ok
negative low deviate normal
no signal fail
ok
normal
normal
no signal fail
ok
positive high deviate normal
no signal fail
ok
negative low deviate positive highno signal fail
ok
normal
positive highno signal fail
ok
positive high deviate positive highno signal fail
ok
Table 4.24
O ring
broken
ok
broken
broken
ok
broken
broken
ok
broken
ok
ok
ok
ok
ok
ok
ok
ok
ok
Constructed rules base on knowledge of fault analysis
64
The apply written rules Mamdani as in Figure 4.31
Construction of the Mamdani model, by calling the anfisedit function in Matlab
Figure 4.25: Construction of the Mamdani model
65
Figure 4.26: Membership function for set point deviate from Measure value in
the Mamdani model
66
Figure 4.27: Membership function for Measure value offset set point from
Mamdani model
67
Figure 4.28: Membership function for output for electronic card fault from
Mamdani model
68
Figure 4.29: Membership function for output for filter blockage in Mamdani
model
69
Figure 4.30: Membership function for output for o ring breakage in Mamdani
model
70
Figure 4.31: Apply the constructed Rules from Figure 4.24 into the Rules Editor
in Mamdani model
71
Figure
4.32: Ruler Editor generated by Mamdani model
72
Figure 4.33 : Ruler Surface generated by Mamdani model
73
4.4.7 The construction of Early Prediction model in Mat lab
The prediction principle is from the AI technique neuro network.
There the 6 stage early prediction concept introduced. The data will be loaded in
to the Trn_ 4 program ,the MF is initialize with the epochs .The data will be
trained in algorithm and the output will recognize to understand the error occur .
Data captured during fault happen , will load in Mat-lab Workspace
.This data will be used to training. Number of times training need depending
epoch number that been training ,And the generated MF will be tested
.Comparison between actual and generated MF output will be analyze .Different
epoch will be tried to get the best comparison .
74
CHAPTER 5
RESULTS
75
5.0 Result of Neuro Fuzzy
Figure 5.1: The output membership of Neuro Fuzzy model generated
76
The rules from generated Neuro Fuzzy model
Figure 5.2: The output Neuro Fuzzy model rules generated
The generated output model rules generated in Rules Editor
77
Figure 5.3: The output Neuro Fuzzy model rules generated in Rules Editor
78
The generated Neuro Fuzzy Model Structure
Figure 5.4: The generated Sugeno Model Structure
79
Figure 5.5: The generated Sugeno Model from Surface viewer
80
Testing input
Figure 5.6: The command to test the output of generated Sugeno model
81
Enter value for input1 = –1.46 , input 2 = 0.3 , input 3 = 0.025
Figure 5.7: Output generated by Sugeno model
82
Generated output at Mat lab [0.7231 0.5 0.1732], which is same as generated by
Anfis rules output.
Compared the rules developed output with tested data
Figure 5.8: To check with the output generated by Sugeno model
83
5.1 Result of Mamdani
Figure 5.9: Output of Sugeno model rules
84
Tested Output set point deviate =-01.46,offset_mv=0.3,Signal_lost=0.025
Electronic card = 0.723 ,filter = 0.5 , O ring = 0.17
Figure 5.10 : Output of Sugeno model rules
85
5.2 Result of Prediction
Output For epoch, n = 500 , MF = 2
Figure
5.11: Comparison predicted data and actual data
Blue color trend is the trend of actual data
Green color trend is the predicted trend output by neuro network
86
Figure
5.12: Prediction Error of the neuro network
87
Output For epoch, n = 1000 ,MF = 2
Figure
5.13: Comparison predicted data and actual data
Blue color trend is the trend of actual data
Green color trend is the predicted trend output by neuro network
88
Figure
5.14 : Prediction Error of the neuro Network
89
Output For epoch, n = 1500 ,MF = 2
Figure
5.15: Comparison predicted data and actual data
Blue color trend is the trend of actual data
Green color trend is the predicted trend output by neuro Network
90
Figure
5.16: Prediction Error of the Neuro Network
91
Output For epoch, n = 2000 ,MF = 2
Figure
5.17: Comparison predicted data and actual data
Blue color trend is the trend of actual data .
Green color trend is the predicted trend output by neuro Network.
92
Figure
5.18: Prediction Error of the neuro Network
93
Output For epoch, n = 3000 ,MF = 2
Figure
5.19:Comparison predicted data and actual data
Blue color trend is the trend of actual data
Green color trend is the predicted trend output by neuro Network
94
Figure
5.20: Prediction Error of the neuro Network
95
5.3 Analysis
5.3.1 Fuzzy Logic
The knowledge that been understand from the fault sign have been use
to construct a fuzzy logic with concept of Mamdani. The decision making of
this Mamdani model very much relies on the knowledge of human. The
collected data from Data base system fine input for the fuzzy logic modeling.
The data shows a clear picture of the fault occur. From the data, study and
analysis need to be done in detail in order to get the fine level of information
which need to be used for modeling the fuzzy logic. Once the construction
model have been made, the next level is to verify the parameter setting in fuzzy
logic. This can be only done with the knowledge of the people who know the
behavior of the system will. The simulation need to be done to verify the
outcome of constructed model. As can be seen , during simulation input set on
the Ruler Editor and also in the in command program where the function
readfis( ) and evalfis () use to check the output of model constructed. The
example as in the Result of Mamdani page:
The
entered
input
:Output
set
point
deviate
=-
01.46,offset_mv=0.3,Signal_lost=0.025.And output produced is : Electronic
card = 0.723 ,filter = 0.5 , O ring = 0.17
96
5.3.2 Neuro Fuzzy
The Neuro Fuzzy ,is constructed by using the captured numerical data
during fault occur in the system. The Neuro Fuzzy is based on Sugeno method
.The Neuro Fuzzy is used to construct model that can recognized the fault occur
from signal input. In order to do that, the data need to be train on Sugeno model.
The step as in the previous chapter in the designing the Sugeno Model. The
parameter to be set during modeling the Neuro Fuzzy such as Number Of
Member function is depend the type of the system to be construct. The Neuro
Fuzzy trained and the structure will be tested as in the Testing of Sugeno model.
The simulation is to confirm the outcome of the system. There are single which
was define by range as recognized output as 1 –o ring broken ,2 –filter damage
and 3 –electronic card fail. This output will define the type of failure from the
data from plant.
5.3.3 Neuro Network
The prediction system is using the concept of neuro network. The data
need to be feed in to train the neuro network system. This system is used to
detect an early sign of error occur from the input data. The system is capable of
predicting the future data that will be produced. From the output simulation, the
Prediction Error , its start to show the error is getting higher before the actual
large variation happen in the shorter time later. This technique help to detect
before the instrument totally fail from its function.
97
CHAPTER 6
CONCLUSION
98
6.0 What have been done during this research
During this research, there are many thing have been carry out. The
first step to do is studies on the behavior of instruments in plant and the type of
fault occur in instrument. The instruments need to know by its structure and
application of it in area of studied. Understand the process and how instrument
affecting the process. The available data of fault is very important in order to
select proper tool for the case studies .The selection of type of data will also
lead the analyze type to be carry out. From the available type of data, its will be
gather and analyze .The analyzing process require more time due some of the
information will be mixed by fact and some information will be need to
interpreted back with some assumption. The construction of AI technique for
this research channel by the literature study .
The have been a lot of literature study which have been carry out in
order to get the correct path of the AI technique guide line .The selected AI
technique in this research will be Neuro Fuzzy, Fuzzy logic and Neuro Network
according to the application of each of this technique. The selected AI technique
will be constructed in Mat – Lab software. During simulation some adjust may
need to be done and some debugging might be required .The adjust will be done
on some of the parameter until the require output is obtain . Once the fine tuning
done, the model that have been constructed will be tested again using the data
from real plant again to verify the out come.
99
6.1 What have been gains from this research
There are a lot of thing gain and learn during this case studies
conducted .The basic of what have been done in the case study as how raw data
from real plant need to be extracted and how to convert it into useful
information in order to run the AI technique .Here the application of Artificial
Intelligent such as the Neuro Fuzzy ,Fuzzy Logic and Neuro Network have
been experience using the actual data from the real plant. The under stand of
technique of Artificial Intelligent in term of conceptual ,application and
practical application.
But there are some knowledge learn indirectly from this case study .
The methodology from other field which can proposed in another field. Where
AI is the field of Computer Science study technique applied in Instrumentation
and Engineering to run the research on fault detection.
The out of box concept where new type of suggests can be taken in to
consideration.
Analyzing data many not to be as straight and easy. Sometime might
need to look from other point and different kind of interpretation. Any more
data can be extracted from it.
Focusing on the problem occur and find the solution may not only
from the existing data resource but also by understanding the basic concept of
problem and applying a good tool.
101
6.2 Recommendation
1.The other system that can be use in this research is System Identification Tool
box in Mat lab to construct the model of the symptoms. Using the capture fault
signal a fault model can also be constructed .
2.A model of servo valve can be constructed by using the transfer function of
servo valve which available in [22] Transfer Function for MOOG Servo valves
by W.J. Thayer, December 1958 Rev. January 1965. Using the theoretically
calculated system fault model, the output of system can be determined. But this
two comparison might have difference .In actual plan the reading might
oscillate within a tolerance range and the calculated output model might view as
very accurate and constant reading. The acceptable tolerance range for
oscillating reading is depend on the servo valve itself .The other this the tool
that used to measure the error signal might need .
3. The system can be up grade for the universal fault detection system for all
type of instrument. Most instrument use 4 -20 mA .A lot of monitoring device
available in market to capture this signal. Algorithm can be made to use an
general signal 4 – 20 mA to detect and recognized fault occurrences in
instrument. Other that the algorithm ,using other technique AI to detect the early
failure as Neuro Network in servo valve application.
5. More additional features tool and features can be added in the fault system in
order to increase reliability and efficiency of the system develop.
102
6.3 Reference literature
[1]Rolf Isermann and Mihaela Ulteru ,”Integrated Fault Detection and
Diagnosis” Technical University of Darmatdt Institute of Automatic Control
Laboratory for Control Engineering and Process Automation .
[2]Rolf Isermann ,”Model Based Fault Detection and Diagnosis Method”
Technical University of Darmatdt , Institute of Automatic Control Laboratory
for Control Engineering and Process Automation .
[3]Dr.Rahmat Shoureshi ,Tim Norick ,David Linder , John Work ,Paula
Kaptain, ”Sensor Fusion and Complex Data Analysis for Predictive
Maintenance “.Colorado School of Mines,Western Area Power Administartion
[7]Rajiv Sreedhar and Benito Fernandez “A Neural Network Based Adaptive
Fault Detection scheme “ Depart of Mechanical Engineering ,University of
Texas at Austin
[6]Leon Reznik and Vladik Kreinovichis, “Fuzzy Prediction Models in
Measurement” Dept of Computer Science ,Rochester Institute of Technology ,
Dept of Computer Science ,and Pan –American Center for Earth and
Environment Studies University Of Texas
[4] L.B.Palma ,F.J.Coito ,R.N.Silva , “Adaptive Observer Based Fault
Diagnosis Approach Applied To A Thermal Plant,” Proceedings of the 10th
Mediterranean Conference
on Control And Automation MED 2002
,Universidade Nova de Lisboa ,Faculdade de Ciencias e Tecnologia ,Detp .Eng
.Electrotecnica,2002
103
[8] Tracy Dalton ,”Introduction to fault diagnosis ”May 1998 ,University of
Duisburg ,Dept Of Measurement and Control.
[5]Alessandro De Carli ,Rccardo Cirillo “Towards Latent Faults Prevision
”,Cosmo Leccese ,University of Rome “La Sapienza”, Dept of Computer and
systems Science
[9] Reza Asgary and Karim Mohammadi “Using Fuzzy Probabilistic Neural
Network for Network for Fault detection in MEMS ” Iran University of Science
and Technology, Electrical engineering Department. Tehran Iran
[10] R.A Reese, “Press Operations “ Senior Process Consultant CRS Sirrine
Engineers, Inc Greenville ,SC .XI
[11] The Paper Making Process From wood to coated paper, Sappi, The Paper
Making Process, the fifth technical brochure from Sappi Idea Exchange,
www.ideaexchange.sappi.com
[12]http://www.ppic.org.uk/info/process/process.htm
[11] Venkat Venkatasubramaniam , “In Process fault detection and diagnosis
:Past ,Present and Future” ,School of Chemical Engineering ,Purdue University
,West Lafayette ,USA .
[12]Erik Dahlquist ,Tomas Lindberg ,Christer Karslsson ; Galia Weidl ,Carlo
Bigaran ,Austin Davey “ Integrated Process control, fault diagnostics
,optimization and production planning –Industrial” Malardalen University,
Sweden ,ABB ,Vasteras Sweden VISY pulp and Paper ,Tumut , Australia .
104
[13]A.M.Bhagwat ,R.Srinivasan and P.R.Krishnaswamy “Model Based
Detection During Process Transitions ” Dept of Chemical and Environment
Engineering ,National University of Singapore .
[14]Bruno HEIM, Sylvie CAUVIN, Sylviane GENTIL “FCC Diagnosis Based
On a Causal And Heuristics Approach ” Vernaison ,France ,Rueil France ,Saint
Martin d’ Heres Cedex 1 France .
[15]Pascal Dufour ,Sharad Bhartiya ,Thomas J.English ,Edward P.Gatzke
,Prasad S .Dhurjati and Francis J.Doyle 3 , “Fault Detection In A Continuous
Pulp Digester ” Depart of Chemical Engineering ,University of Delaware ,USA.
[16]Dauren F. Akhmetov ,Tomofumi Miura ,Masataka Itou , Manabu Tamamori
,Hiroto Murabayashi and Yuichi Hamade “Aggregative Learning Method And
Its Applications To Fault Detection” NTTDoCoMo Hokkaido ,Inc ,Towa Denki
Seisakushyo Co .LTD .
[17] B.Jiang ,M.Staroswiecki and V.Cocquempot “Fault Diagnosis For A Class
Of Nonlinear Systems With Unknown Parameters ” Universite des Sciences et
Technologies de Lille ,France .
[18] Sheng Yung Chang, Cheng Ren Lin and Chuei Tin Chang ,Shuh Woei Yu
“On – Line Fault Diagnosis Using Dynamics Fault Tree ” Dept of Chemical
Engineering ,National Cheng Kung University ,Tainan ,Taiwan – Center for
Industrial Safety and Health Technology ,Industrial Technology
Research
Institute ,Hsinchu ,Taiwan .
[19] Piotr Kulczycki , “A Statistical Interference System For Fault Detection ”
,Polish Academy of Sciences ,Systems Research Institute , Newelska ,Poland .
105
[20] Bentiz – Perez H and Rendon – Torres P “Fuzzy Classification of Faults in
a nonlinear Actuator ”Dept de Ingenieria de Sistemas Computacionales y
Automatizacion ,Mexico .
[21] In Soo Lee ,Doug Hyung Lee ,Euy Soo Lee ,Sang Jin Park “Neural
Network –Based Software Sensor For Feed Identification In Crude Distillation
Unit , Sangju National University ,Kyungpook ,Korea - Dongguk University
,Seoul ,Korea .
[22] W.J. Thayer “Transfer Function for MOOG Servo valves” ,December 1958
Rev .January 1965 Technical Bulletin 103 ,MOOG INC ,Control Division ,East
Aurora ,NY