Você está na página 1de 5

Study of Fault Diagnosis Approach Based on Rules

of Deep Knowledge Representation of Signed


Directed Graph
Wen-liang Cao1, Bing-shu Wang1, Liang-yu Ma1, Qin Yan1, Wei Hao1, Yufeng Xin2
1. School of Control Science & Engineering, North China Electric Power University, Baoding, Hebei 071003, China
2. Yangcheng International Power Generating Company, Yangcheng, Shanxi 048102, China

Abstract-The fault diagnosis method using a Signed Directed


Graph (SDG) based on qualitative model as a model of the system A O sign(A-B) OB
is useful to real-time diagnosis of failures that occur in process.
First, it establishes the SDG of the systems and components,
simplifies these SDG corresponding to the fault modes needing to Fig. 1. Simple Structure of SDG
be diagnosed, at the same time SDG are described the many rules
forms for shortening the calculating time of making use of SDG, adjusting thresholds is beneficial to improve forecast and
then expands the diagnosing rule with expert knowledge to resolution of diagnosis.
construct the diagnosing rule bank of the system; Second, the Directed arcs between nodes in SDG represent the
fault modes can be primary diagnosed by using the constructed qualitative influence of a variable (initial node A ) on another
rules; And then the modes that can not be distinguished are variable (final node B), when the arc sign "sign(A -B) "is
diagnosed by adding adequate quantitative information. The case
studies show that the problem of misoperation autodiagnosis increasing influence "±", the change tendency between branch
during computer simulation training can be solved effectively, and nodes are the same, when the " sign(A - B) "is decreasing
the SDG diagnosis method has good completeness, fine resolution influence "-", the change tendency between branch nodes are
and detailed explanation in actual industrial process. the opposite.
If the 4D(A) or 4D(B) are not "0", this node can be regarded as
I. INTRODUCTION valid, which shows the abnormal state of its according variable,
that is to say, the system has fault. An arc is said to be
Signed Directed Graph (SDG) is a very convenient tool to consistent if 4D(A) * sign(A - B) * 4D(B) =+. It can be regarded
provide a visual description on the industrial system structure that consistent arc is the only path that can propagate the fault
by using graph to show the cause and effect relation between information. The relation between nodes is called Cause-Effect
system variables. Applying the description system SDG-based Graph (CEG), which is composed by all valid nodes and
and causal analysis, it can identify the root cause by using the consistent arcs in SDG, so the problem of fault diagnosis
information saved on SDG to search for the possible fault SDG-based can be described as the search of most connectivity
source of disturbance. Its advantage lies in the less information branches in CEG [1].
in SDG construction and application in diagnosis. First, it
establishes the SDG of the systems and components, simplifies III. SDG MODELING
these SDG corresponding to the fault modes needing to be
diagnosed, at the same time SDG are described the many rules SDG modeling of high quality that is the base that fault can
forms for shortening the calculating time of making use of SDG, be diagnosed successfully in system. Although the form of
then expands the diagnosing rule to construct the diagnosing SDG is only made up of the nodes and signed arcs, each of
rule bank of the system. Second, transforming the quantitative these must be considered carefully in order that SDG model
values of the system's variables into qualitative values, and accords with objective law, so it is essential to realize the
then the fault modes can be diagnosed by using the constructed process deeply and master practical experience. There are three
rules. methods about modeling SDG are introduced in the following.
II. DEFINITIONS AND EXPLANATION OF SDG A. The Experience Method
With the development of network technology especially the
Research on fault diagnosis approach SDG-based, an application of plants' Intranet system, nowadays most real-time
example of three-pattern SDG is shown in Fig. 1. The value of supervisory system already adopted Web distribution. Web
each node represents the qualitative state of a given process distribution works according to database server, so the
variable, it includes three signs "+", "-"or "0", if the node " A " construction of real-time database has great effect on the
or " B " is upper than the threshold, then the node capability of information distribution [2].
signs D(A) or D(B) are "+", if the node variables is lower than SDG modeling has good direct property and the
the threshold, then D(A) or D(B) are "-", if they are at normal non-measurable nodes contained in the SDG is few, but it is

0-7803-9484-4/05/$20.00 §2005 IEEE 778


possible that some faults in system which may not be predicted C. Mathematics Model Method
by human experience would be excluded from system in this The mathematics equation that describes system can reflect
model. The steps are the following: system action more comprehensive, however, it is more
a. To take the critical variables corresponding to the faults of complicated and has more non-measurable nodes that need to
system as nodes. be simplified. Considering the process mathematical model,
b. To Find out the combination of nodes and arcs resulting in generally rewritten in the following form:
the fault as far as possible. At the same time, it will be analyzed
that they are positive or negative influences in principle, then dxt =fi (X1, X2 ..., Xn) (1)
the relations between nodes are connected by "+" or "-".
c. To use experiences, field information and dynamic and xi (to) = xi (i = 1,2 , n) (2)
qualitative simulation model of system to test, modify and The expressions for discretization of eq.(1) and eq.(2) is:
simplify SDG model until the model satisfies requirements.
During the course of modeling SDG, the principle of nodes and i i + 2 n (3)
arcs determined is that it should be favorable to reveal the
cause and result of faults in system on the basis of according Xi (tn ) x i (i 1,2, n; m 0,1,2, ... M)
=
'
= ...

,
=
(4)
,
with the objective law.
Actually, the method based on mathematics model is to
B. Flow Graph Method complete the transformation from quantitative mathematics
It is common knowledge that the fault's propagation is along model to qualitative SDG model, in which the node in SDG is
technological process in system. So on the basis of the above corresponding to the variable in mathematics equation, and the
feature, the SDG model is easily established by the way which state of arc can be acquired from partial derivative of equation.
has direct property and can reflect properly the relations of In other words, in a digraph the gain GiC corresponding to the
fault's propagation in system. However, It is inclined to
construct a partial model or not whole, especially in the part of arc from node xi and node Xi is a qualitative indicator of the
equipment units of system and leave out some important direct effect on Xi due to a perturbation on xi .
variables in whole system. Moreover, the SDG modeling by
this way is more complex and need to simplify by depending
on mathematical model with experience and simulation. The GU = Sign(a xi /axi) (5)
steps are the following: The relationship between node xi and node xj can be
a. To divide technological process into several units of expressed as axja/xi in the mathematic model; accordingly,
equipment; the sign of axj /axi is the state of the arc between xi and xj in
b. To mark a numeral in the joints of equipment units. As an
example of Flow Detector (FT) in Fig.2, the joints of the SDG model.
connecting with other equipment units separately marked In a word, aiming to the different situations, in the process of
number 1, 2 and 3, the following is for number 3 to illustrate. SDG modeling, it should combine the advantages of the above
The number 3 is marked in the joint between FT and PID, the three methods for obtaining the SDG model of high quality.
output signal in FT is showed by F3 and input signal in PID is IV. REASONING RULES OF SDG
also showed by F3. So through this number 3, nodes of two
different equipment units are combined one node during the A. Fault Simulating
course of SDG modeling of larger system, other nodes are Fault diagnosis is in fact the inverse process of fault
similar. simulating; diagnosis is to test the correctness of state
c. To construct SDG model of each equipment unit. The assumption from the observed symptom, while simulating is to
relations between the related variables and events of deduce the prediction from the given state, so the complicated
propagation among equipment units of system will be diagnosis problem can be further analyzed from the angle of
determined by correlation analysis, experience or test of simple simulating.
simulation, that is, it will be determined that nodes and state of According to one given root node (fault source), Fault
arcs. For example, Fig.3 shows SDG model of equipment units simulating can generate a sub SDG, which can be called
of FT. SDG model of system will be constructed by the simulating tree and interpretation; it represents the route of
combination of model of each equipment unit according to fault propagation and generates the information of fault and
technological process and the joints of each equipment unit. deviated tendency of each node. As for one given SDG and
root node, it is possible to have more than one propagating
If FT damages, interpretation, but only one of them can reflect the actual
t3 Relation is 0
process action.
F F 3 Let's take Fig.4 as the example; it is a simplified SDG, in
which if the node "A" occurs positive deviation, the two
interpretations shown in Fig.4 (a) and Fig.4 (b), the
Fig.2. Equipment Unit of FT Fig.3. SDG of FT corresponding fault pattern of each measurable node is shown

779
A A A
Fq TABLE II
K0 -X )B FO B Fc SDG LOGIC RELATION OF SDG
E D c E D C E D C
Types A->B <- (pAB) A->B <- (mAB)

Fig. 4. Fault source A Fig.4. (a) Pattern a IFig.4. (b) Pattern b


B + O- +
O -

+ T F F F F T
TABLE I
FAULT PATTERNS OF CORRESPOND1NG To RoOT NODE O F T F F T F
Patterns (a) Patterns (b) F F T T F F
B CD E F B C D E F
0 0 0 0 0 0 0 0 0 0 B. Simplifies and Reasoning Rules of SDG
What has been discussed above, it can be seen that one fault
0 0 0
0
00 0 source can generate many interpretations, which can be
0O O
- 0 00
0 0 + 0O + 0 O combined and simplified as one rule, However, it is infeasible
O O + - 0 + + O to use this method in actual industrial process, so the following
O O + -- __+ + + procedures need to applied to generate rules from process SDG
+ O O directly [3][4].
-- + O O * To select one root node;
00 + * To delete the arcs that point the root node;
-
0 + -0 0 * To combine the nodes of sensors that have not fault with
its detecting state variables; and delete the node that the
root node cannot reach from SDG.
* As for each measurable node "n1 ", the following rule
in TABLE I. When the actual fault pattern detected online match need to added:
with that in TABLE I, the fault source can be regarded as the ...
and[(*kIni)or(*k2ni) ...or(*kjni)]
positive deviation of "A".
In practice, it is unscientific to apply the diagnosis method * As for positive feedback loop, the following rule need to
above in the large-scale process, so we will introduce a method added:
of transforming the SDG to a series of rules. ... and[(*Bl)or(*B2) ... or(*Bi)]

First, in order to conveniently examine whether one arc is a


consistent arc, we will adopt logic sign to describe, the "P " In which ki is the input node of the ni , Bi is the loop input
and " m " are defined as the logic function form of the arc sign node, * is the logic function form p(+) or m(-) from ki to ni .
of "+" and "-". For example, the logic relation of SDG in Fig. 1
is shown in TABLE II. In addition, the logic function form of v. REASONING MECHANISM AND CASE ANALYSES
node A <O <- A.lt.O; A =O <- A.eq.0; A >O <- A.ht.O.
The reasoning mechanism based on SDG is to acquire the
The interpretations of "A" fault source in Fig.4 (a) can be instant measurable samples from the operating process
described by its logic relationship as: database, and compare with threshold value of the node, once
+
the abnormal deviation of important variables are detected, the
+
A-B *-*(mAB); B-C *-*(pBC); A->D *-*(pAD);
reverse reasoning from this deviation node will be made, then
D-E *-*(mDE); E-F *-*(mEF) the fault source is determined based on the examination of
The interpretations of " A " fault source in Fig.4 (b) can be consistent arc [5][6].
described by its logic relationship as: The Thermal Exchange Equipment (TEE) in industry is
++
A->B<->(mAB); B->C<->(pBC); C-D*-(pCD);
shown in Fig.5, the SDG model of TEE is shown in Fig.6. In
which the irreversible exothermic reaction is going on, the cold
D-E <-*(mDE) ; E-F <->(mEF); fluid temperature T of TEE is adjusted by flow Fo1 of cooling
Thus all the fault patterns of " A" fault source can be described water, the liquid level L of TEE is controlled by outlet flow F2,
in the rule of IF -THEN as following: and the PID controllers are used in control loop. Besides, other
IF[(mAB)and(pBC)and(mAD)and(mDE)and(mEF)] parameters: such as flow F1, inlet valve v1 and temperature T1 of
or[(mAB)and((pBC)and(pCD)and(mDE)and(mEF)] hot fluid; inlet temperature To ,and outlet temperature To1 of
THEN A is the possible fault source
By using distribute disciplinarian to simplify the above rule,
cooling water; outlet valve v and control valve vo of cold fluid,
we can draw out that: level detecting cell LC , set point LS, flow detecting cell FC and
[Rule]: IF(mAB)and(pBC)and(mDE) and(mEF)and[(pAD)or(pCD)] set point Ts, the pipe blockage FoII of cooling water, the pipe
THEN A is the possible fault source blockage F11 of hot fluid and so on.

780
Hot FluidT technique, the problem of on-line misoperation autodiagnosis
during computer simulation training can be solved effectively,
and instructors' burden can be reduced greatly.
FCoToli ____'
B. Application in Actual Industrial Process
In actual application, owing to the cause of technology and
0 0I.
economy, some nodes in Fig.5 could not be measured. Such as
_....__ L v ~~~~~~~~~~Cooling
Water the nodes V1 , Lc and Fc are unmeasurable. According to the
Fig.5. ...Pocs hrto T method mentioned in section IV, for example, the sub SDG1'
Fig5 o P af TTE corresponding to PI is shown in Fig.8, the corresponding
diagnosis rule is [Rule2]. By the same way, we can get the
diagnosis rule bank of the system.
[Rule2]: IF(To.eq.O)and(T1.eq.O)and(F1.It.O) and(mTTc )and(mTcVO)
and[(pF1L)or(mF2L)]and(pL V)and(p VF2)
and[(pLT)or(pTOjT)]and[(m VoToj)or(pTToI)]
The positive feedback loop between T and TO1:
andIF[(pLT)or(m VoTo 1)]
THEN V1 is the possible fault source.

Fig.6 SDG of TTE


+T
In the case, seven fault patterns are taken into consideration,
such as the blockage of in v1; the increase in To; the decrease
in T1; the blockage of in v; the jam in FO11; the increase of vo;
Fijg.7. SUb SDG1 Of Pt Fig.8. Sub SDG1' of PI
the jam in F11, they are marked as Pt, P2, P3, P4, P5, P6 and P7
TABLE III
separately. CASES STUDIES
A. Application in Simulation Training Cases 1 2 3 4 5 6 7
In application of simulation training, owing to the all nodes
in Fig.5 are measurable, according to the method mentioned in
section IV, for example, Firstly, SDG should be simplified to F1 0 0 0 0 0
get sub SDG1 corresponding to the PI is shown in Fig.7, F2 0 0 0 0
Secondly, the diagnosis [Rulel] should be established. By the
same way, we can get diagnosis rules of the corresponding rest
fault patterns, and construct the diagnosing rule bank of the L 0 0 ± 0 0
system. LS 0 0 0 0 0 0 0
[Rule ]: IF(To.eq.O)and(T1.eq.O)and(V1j.It.O)and(pV1F1) FC - ° °
and[(pFiL)or(mF2L)]and(mLLc )and(mLcV)and(pVF2) V0 ± ± ± ± ± ±

and[(pLT)or(pTOIT)]and[(mFOITOI)or(pTToI)]
and(mTTC )and(pTCFC) and(mFCoV)and(pVOFO)
Vi 0 0 0 0 0 0
V 0 0 0 0
The positive feedback loop between T and To1:
T ± ± ± ± ±
andlF[(mF1T)or(mFO1T01)]
THEN V1 is the possible fault source.
Using the rules above, the case test is studied on the TS 0 0 0 0 0 0 0
Simulation Training, in TABLE III, there lists SDG instant T, 0 0 0 0 0 0
measurable samples and diagnosis results. This case study is
very successful, not only the diagnosis result is correct but also ±
all the resolutions achieve 100%. That is, by using of SDG TO 0 0 0 0 0 0
Diagnosis PI P2 P3 P4 P5 P6 P7
deep knowledge model and inverse direction inference Results___

781
In like same, by using the rules above, the above fault consumption, while diagnosis rules corresponding to sub SDG
patterns are tested by the instant measurable samples. The case can shorten greatly the time consumption.
studies show the fault pattern P2, P3, P4, P5, P6 is unique (3) High efficiency simulation software platform for process
determined based on the examination of consistent arc. While system with misoperation autodiagnosis can greatly improve
the fault pattern PI and P7 can't be distinguished, this is the training result.
because they are same in qualitative characteristics, that is, (4) The modes that can not be distinguished may be
SDG only applies qualitative knowledge in the process of diagnosis to combine the quantitative knowledge.
establishment without considering other deep quantitative
information. ACKNOWLEDGEMENTS
So we consider to combine the quantitative knowledge to This paper is supported by the Teacher Fund of Doctor's
improve SDG model, For example, by analysing the operating degree of North China Electric Power University (20041209).
data of fault pattern PI and P7, we can distinguish the two fault
patterns by adding the Additional Rules. REFERENCES
[Additional Rule 1]: IF: the changing scope of inlet flow is the [1] Honggang Liu, Jianjun Wu, Xiaoqian Chen. Study on the SDG-Based
most in all relative parameters. Intelligent Fault Diagnosis Method, Systems Engineering and
THEN P7 is the possible fault source. Electronics, 24(1): 103-105, 2002
[2] Chongguang Wu, Tao Xia, Beike Zhang. The Qualitative Simulation
VI. CONCLUSIONS Based on Deep Knowledge Model of Signed Directed Graph, Journal of
System Simulation,15(10): 1351-1355,2003
(1) The diagnosis model based on SDG has high [3] Mianyu Wang. Research for Fault Diagnosis Based on Signed Directed
Graph. M. Sc. Thesis, 49-65,2003, China
completeness, which is suitable to the diagnosis on the [4] Shiozaki,J.et al. An improved algorithm for diagnosis of system failures
problems of root cause and multiple cause-effect relationship; in the chemical process. Computer & Chemical
and the SDG diagnosis method has good completeness, fine Engineering,.9(3):285-293,1985
resolution and detailed explanation. [5] M.A. Kramer and J.B.L. Palowitch. A rule-based approach to fault
(2) As for one given SDG and fault source, because of the diagnosis using the signed directed graph. AIChE. J, 33:1067-1078,1987
[6] H.Vedam and V.Venkatasubramanian. Signed digraph based multiple
uncertainty of SDG in forecasting fault propagation route, there fault diagnosis. Computers & Chemical Engineering, 21:655-660, 1997
are more than one explanation which can lead to the long time

782

Você também pode gostar