Advances in Engineering Research, volume 170

7th International Conference on Energy and Environmental Protection (ICEEP 2018)

Condition Analysis of Steam Turbine DEH Control System based on

Data Fusion
Zhenhe Wang
College of Sciences, Hebei University of Science and Technology, Shijiazhuang, Hebei
050018, China
e-mail:[email protected]

Keywords-DEH control system; condition analysis; data fusion

Abstract-Aiming at the problem that the system of digital electro-hydraulic(DEH) regulation
system is complex and difficult to analyze its condition, a method based on characteristic
extraction and information fusion is proposed. By making RCM analysis, the indexes for
quantitative risk evaluation of fault modes are determined, and the fuzzy rule base for risk
evaluation is built. Using fuzzy inference, the system fault modes are ranked according to their
risk level. Then, the condition characteristic parameters are extracted according to the pivot fault
modes. Using the extracted characteristic parameters, an information fusion method based on
evidence theory is put forward to evaluate the system’s condition. It is shown by the instance that
this method is feasible and effective and the condition analysis results can be used as a support
for next maintenance decisions.

INTRODUCTION
With the rapid developing of electric power industry in our country, the maintenance mode
is changing from planned maintenance to condition-based maintenance. The pivotal task of
condition-based maintenance is to evaluate the equipment condition synthetically and arrange
maintenance times and items reasonably. DEH regulation system is complex and includes many
components. Meanwhile, the relations between subsystems have strong relationships. In
condition analysis, reasonable condition characteristic extraction method and accurate condition
evaluation model are two pivotal aspects. But, in most methods[1], condition characteristics are
extracted just using qualitative analyzing and expert opinion, and the condition evaluation model
is just linear integration. The accuracy of condition evaluation results is weaken greatly. So based
on RCM analysis, fuzzy theory and evidence theory, a method of condition analysis is put
forward for DEH system.
In the paper, take DEH regulating system as an example, all subsystems are analyzed in
RCM method based on system division and the fault modes and their risk evaluation indexes are
determined. Then based on fuzzy theory, the risk levels of all fault modes are evaluated
quantificationally using fuzzy inference method and the characteristics are extracted by
analyzing the fault modes with higher risk levels. Finally, the system condition is analyzed by
information fusion arithmetic based on D-S theory.

RCM ANALYSIS ON POWER STATION EQUIPMENT

A. System division of DEH regulating system

According to the principle of function correlation, the system division is shown in Fig.1

Copyright © 2018, the Authors. Published by Atlantis Press.

This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). 321
 Advances in Engineering Research, volume 170

DEH regulating system

Digital system
Electronic sets
Analog system
Elec-hydro converter
Actuators Oil motor
Regulating valves
Oil pump
EH oil sys. Oil box
Pipes
OPC
Protection sys. AST
Manual snap
Fig.1 The division of DEH regulating system

B. FMEA on DEH regulating system

In the process of fault mode and effective analysis, the possible fault modes and their effects
of subsystem or components are listed in the form of table[2,3]. In the same time, the risk
evaluation indexes of fault modes are also determined. Based on the actual operation condition
of DEH regulating system, the components prone to fail are selected to be analyzed.The possible
fault modes of the components are determined. In Tab.1, the fault modes and their risk evaluation
factor values of oil motor are listed.
Tab.1 Oil motor’ failure modes and linguistic descriptions
FM no. Failure modes Severity Occ Freq. Dete.
F1 Inlet jammed Mod Vlow Mod
F2 Outlet jammed Mod Low Low
F3 Spring rapture Mod Low Mod
F4 Unfit preload Mod Low Low
F6 Piston jammed Mod High Low
F7 Piston wear Mod High Low
F8 Sealing ring bad Mod Mod Mod
F9 LVDT break Mod Low High

C. Quantitative indexes of failure mode risk analysis

During the risk evaluating of failure mode, the 3 indexes of failure severity, probability of
occurrence and detectability should be considered. The evaluation criterions of the there indexes
are shown in table 2, table 3 and table 4. In addition, the risk evaluation index of failure mode is
shown in table 5.
Tab.2 The scale for severity

Severity Rating
Very low 1
Low 2
3
Moderate 4
5
6
High 7
8
Very high 9
10

322
 Advances in Engineering Research, volume 170

Tab.3 The scale for probability of occurrence

Probability of occurrence Failure rate(day) Rating
Very low <1:20000 1
Low 1:20000 2
1:10000 3
Moderate 1:2000 4
1:1000 5
1:200 6
High 1:100 7
1:20 8
Very high 1:10 9
1:2 10

Tab. 4 The scale for detectability

Detectability Probability of detection(%) Rating

Very high 85 -100 1
High 75-85 2
65-75 3
Moderate 55-65 4
45-55 5
35-45 6
Low 25-35 7
15-25 8
Very low 5-15 9
0-5 10

Tab.5 The scale for risk evaluation

Risk None Elow Vlow Rlow Low
Rating 1 2 3 4 5
Risk Mod High Rhigh Vhigh Ehigh
Rating 6 7 8 9 10

RISK ANALYSIS ON FAILURE MODES BASED ON FUZZY INFERENCE

A. Fuzzy membership function[4,5]

Depending on the table 2~5 and the experiences of Depending on the table 2~5 and the
experiences of domain experts, the fuzzy membership functions for the severity, the probability
of occurrence, the detectability and risk of failure mode is set up, and shown in figure2~5.
In the figures, the scope of abscissa is between 1 and 10. Using figures of fuzzy membership
function, the corresponding fuzzy membership value can be gained when input the evaluation
value of a certain index. The value of a certain index can be decided by the there evaluation
criterion tables.
Very low Low Moderate High Very high
Fuzzy membership

1 .0
0 .8
function

0 .6
0 .4
0 .2
0 .0
0 2 4 6 8 1 0

Fig. 2 The severity

Very low Low Moderate High Very high
Fuzzy membership

1 .0
function

0 .8
0 .6
0 .4
0 .2
0 .0
0 2 4 6 8 1 0

Fig. 3 The probability of occurrence

323
 Advances in Engineering Research, volume 170

Very high High Moderate Low Very low

Fuzzy membership
1 .0
0 .8
function

0 .6
0 .4
0 .2
0 .0
0 2 4 6 8 1 0

Fig. 4 The detectivity

None Elow Vlow Rlow Low Mod High Rhigh Vhigh Ehigh
Fuzzy membership

1 .0
0 .8
function

0 .6
0 .4
0 .2
0 .0
0 2 4 6 8 1 0

Fig. 5 The risk level

B. Fuzzy rule base development

Fuzzy inference systems are knowledge-based or rule-based systems constructed from
human knowledge in the form of fuzzy IF-THEN rules. In fuzzy inference systems, the rule base
describes the synthetic risk caused by input variables with different combination. The rules
usually are described by natural language, and in the form of IF-THEN. The IF-THEN rule
consists two parts, one is precondition which can compares with input, the other is the result
under the precondition. In this paper, the input of rule base is severity, probability of occurrence
and detectability, the output is a linguistic variable, that is, the criterion of fault mode risk
evaluation.

C. Principle of fuzzy inference

The min-max fuzzy inference form is used in the paper[6,7 ],there are 3 important formulus
as follows.
According to the fact“x0 and y0”and the inference of rule“ A i andBi ⇒ Ci ”（i=1，2，…，n）,
the fuzzy membership grade of C 'i can be got using formula (1)
µ C ' ( z i ) = µ Ai ( x 0 ) ∧ µ Bi ( y 0 ) (1)
i

Suppose a certain fact x1 , y1 , from m rules, a same inference result Ci' (i = 1,2,L, m) is
obtained. The m rules correspond to m fuzzy membership µ Ci'1 , µ Ci' 2 , L , µ Cim' . The fuzzy
membership grade of result Ci' can be calculated using formula (2)
µ C ' ( zi ) = µ C' ( zi ) ∨ µ C' ( zi ) ∨ L ∨ µ C ' ( zi ) (2)
i i1 i2 im

According to the fact x 0 , y 0 , the fuzzy membership grade of C1'，C 2' is µ C1' ( z1 ), µ C2' ( z 2 ) .Then,
a method of weighted averaging can be used to get the numeral value of the fuzzy result
C ' (C1' , C 2' ) , and the representative value z 0 is obtained.
2 2
z0 = ∑ zi µ C' (zi ) / ∑ µ C' (zi ) zi ∈ Z (3)
i i
i =1 i =1

Thus, using formula (1)~(3), the fuzzy inference from fact “x0 and y0” to the result
representative point z0 can be realized.

324
 Advances in Engineering Research, volume 170

CONDITION ANALYSIS BASED ON INFORMATION FUSION

A. Decision tree information fusion model

Typical decision tree information fusion model is as follow[8].
yk (α )
p ( yk (α )) —Condition level

H1 …
Hn …HN —Evaluation level

e1k (α ) … j
e (α )
k … ekLk (α ) —Basic factor level

Fig.6 A representative decision tree information

fusion model

In evaluation level, H n is called an evaluation grade ( n = 1, L , N ). A set of evaluation

grades for condition attribute y k is denoted by
H = {H 1 , L , H n , L , H N } (4)
where N is the number of evaluation grades. H 1 and H N are set to be the worst and the best
grades, respectively, and H n+1 is preferred to H n .
The qualitative evaluation is difficult to give, as it is subjective and sometimes incomplete.
In order to quantify these evaluation grades and eventually to quantify subjective judgments with
uncertainty, the concept of preference degree was introduced. A preference degree takes values
from the close interval [-1,1], the set of evaluation grades may thus be quantified by
p{H } = [ p ( H 1 ),L, p ( H n ),L, p( H N )]
T
(5)
where p( H n) is the scale of H n and satisfies the following basic conditions:
p ( H 1 ) = −1 ， p( H N ) = 1
p( H n + 1 ) > p ( H n ) ， n = 1, L , N − 1 (6)
In basic factor level, the factor set connecting with condition attribute evaluation is as follow:
{
E k = e1k , ek2 ,L , ekLk
(7) }
where e (i = 1,L, Lk ) denotes the factors affecting the evaluation on y k , the evaluation of e can
i
k
i
k

be determined by connected condition characteristic parameters, eki = eki (α ) , A larger

preference degree value is interpreted as a higher evaluation grade. So the preference degree for
the state of an condition attribute y k through the direct evaluations of the relevant factors eki
can then be generated and integrated by using the D-S evidence theory presented follow.

B. Condition analysis arithmetic based on information fusion

D-S evidence theory is a kind of information fusion classification arithmetic based on
statistic, the Dempster rule combining multiple information is as follow:
In the decision tree information fusion model, evaluation grade H n can be regarded as a
basic hypothesis of D-S evidence theory, factor eki as a piece of evidence, and a basic
probability assignment may be obtained from a confidence degree. All of the evaluation grades
in H are defined as distinct grades. With this in mind, the frame of discernment may be defined
by
Θ = H = {H 1 , L , H n , L, H N } (8)
Let M ( H n ek (α )) express a basic probability assignment to which ek supports a
i i

hypothesis that the state of y k is conform to H n , also let β (e ki (α )) be a confidence degree to

which the decision maker considers that the state of eki is confirmed.

325
 Advances in Engineering Research, volume 170

If there is only one factor eki in E k , m( H n eki (α )) should be equal to β (e ki (α )) ; if there are
multiple factors in E k , then:
M ( H n eki (α )) = λik β (eki (α )) (9)
in which , λ are the weights of all factors e in E k .
i
k
i
k

After gained the basic probability assignments, the overall probability assignment can be
calculated by using the following evidence reasoning arithmetic.
Define a factor subset e Lk (i ) (α ) and a combined probability assignment MM riC (α ) as
follow:
{
e Lk (i ) (α ) = e1k (α ),L, eki (α ) }， 1 ≤ i ≤ L k (10)
C
MM r,Ci (α ) = M ( ) = M riC (α ) (11)
e Lk (i ) (α )
Then, initial condition:
MM rn,1 = M rn,1 ； MM rH,1 = M rH,1 (12)
iterative formula:
MM rn,i +1 = K r ,i +1 ( MM rn,i M rn,i +1 + MM rn,i M rH,i +1 + MM rH,i M rn,i +1 )

n = 1,L , N
MM H
r ,i +1 = K r ,i +1 MM M
H
r ,i
H
r ,i +1 (13)
−1
 N N 
in which, K r ,i +1 = 1 − ∑ ∑ MM rt ,i M rs,i +1  ,
 t =1 s =1, s ≠ t 
i = 1,L , Lk − 1 ， r = 1, L , R
Then the overall preference degree can be calculated as follow:
N
p rk = p ( y k (α r )) = ∑ MM rn, Lk p ( H n ) + MM rH, Lk p ( H ) (14)
n =1
N p(H n )
where, p( H ) = ∑ .
n =1 N

C. Condition analysis example of the DEH regulation system

Took a 300MW steam turbine unit DEH regulation system as example. Using the results of
section 3, the condition characteristic parameter values of a certain running time is listed as
tab.9.
In this example, the condition space is defined as follow:
H = {H 1 , H 2 , H 3 , H 4 , H 5 } = {serious , fault，general，good，best}
(15)
Tab.6 Condition characteristic parameter values of DEH regulation system
Threshold ; Actual
Parameter names units
normal values values
Ratio of flutter >0.5; 1 0.84 －
Oil pressure difference
<1; 0.2 0.42 Mpa
between in and out
Actua-t Down oil pressure of oil
<1.5; 1 1.16 －
ors motor
Worn scraps in oil <5; - 1.6 －
Flux warp of valve <0.2; 0 0.042 －
Power attenuation time <1.5; 1 1.16 －
EH oil pump swing <50; - 16 μm
EH oil EH oil pump current <30; 21 25.5 A
system Oil temperature <57; 43 46 ℃
EH oil main pipe pressure 11-16; 14.5 13 Mpa
Protect OPC oil pressure >1; 1.96 1.78 Mpa
ion sys AST oil pressure >1; 1.96 1.86 Mpa

After making unitary treatment and fuzzy transform, the results are listed in tab.10, which

326
 Advances in Engineering Research, volume 170

are used as the input for system condition analysis.

Tab.7 Input data for system condition analysis
Parameter names Weight H1 H2 H3 H4 H5
Ratio of flutter 0.18 (0,0,0.1,0.53,0)
Oil pressure difference
0.21 (0,0,0,0.83,0)
between in and out
Actuat Down oil pressure of oil
0.23 (0,0,0.1,0.53,0)
ors motor
Worn scraps in oil 0.19 (0,0,0.1,0.53,0)
Flux warp of valve 0.11 (0,0,0,0.73,0)
Power attenuation time 0.08 (0,0,0.1,0.53,0)
EH oil pump swing 0.24 (0,0,0.1,0.53,0)
EH oil EH oil pump current 0.27 (0,0,1,0,0)
system Oil temperature 0.23 (0,0,0,0.76,0)
EH oil main pipe pressure 0.26 (0,0,0.65,0,0)
Protect OPC oil pressure 0.5 (0,0,0,0.6,0.1)
ion sys AST oil pressure 0.5 (0,0,0,0,1)

Put above dada into the iterative formula, the condition analysis results of subsystems and
system can be gained as follow:
Tab.8 Condition analysis results
Preference
System Weight H1 H2 H3 H4 H5
degree
Actuators 0.46 (0,0,0,0.29,0.57) 0.686
EH oil system 0.28 (0,0,0.33,0.19,0) 0.076
Protection system 0.26 (0,0,0.032,0.41,0) 0.164
DEH system - (0,0,0.06,0.23,0.22) 0.312

The above data indicate that the three subsystems are in “best”, “general” and “good”
condition respectively, and the whole DEH system is in “good” condition. Analyzing the
preference degrees of the three subsystem, the EH oil system is in relatively poor condition.
Based on this condition analysis, caution should be exercised and measure be taken in
appropriate opportunity.
the condition analysis results can denote the system condition more accurately, and provide
an effective support for next maintenance decision

ACKNOWLEDGEMENTS
The project is supported by Hebei science and technology department (15273727), and the
Foundation of Hebei University of Science and Technology (XI201504)

REFERENCES
[1] Yujiong Gu, Yuliang Dong, Kun Yang, “Synthetic evaluation on conditions of equipment in
power plant based on fuzzy judgment and RCM analysis”, Proceedings of the CSEE, Vol.24,
No.6, pp.189-194, June 2004.
[2] Gilchrist W, “Modelling failure modes and effects analysis”, International Journal of
Quality Reliability management, Vol. 10, No. 5, pp. 16-23, May 1993
[3] Chang C L, Wei C C, Lee Y H, “Failure mode and effects analysis using fuzzy method and
grey method”, Kybernetes, Vol. 28, No. 9, pp. 1072-1080, Sep.1999.
[4] Klir GJ, Yuan B, Fuzzy sets and fuzzy logic: theory and application, New Jersy: Prentice-Hall,
1995
[5] Kuusela H, Spence MT, Kanto AJ, “Expertise effects on pre-choice decision processed and
final outcomes: a protocol analysis”, Eur J Market, Vol.32, No.5/6, pp.559-576, Jane. 1998.
[6] Bowles J B, Pelaez E C, “Fuzzy logic prioritization of failures in a system failure mode,
effects and criticality analysis”, Reliability Engineering and System Safety, Vol. 50, No. 2, pp.
203-213, Oct.1995.
[7] Zeng Huanglin, Intelligent computing, Chong qing university, Chong qing, 2004.

327