CN106250631B - It is a kind of based on failure-test correlation matrix method for diagnosing faults - Google Patents
It is a kind of based on failure-test correlation matrix method for diagnosing faults Download PDFInfo
- Publication number
- CN106250631B CN106250631B CN201610629637.6A CN201610629637A CN106250631B CN 106250631 B CN106250631 B CN 106250631B CN 201610629637 A CN201610629637 A CN 201610629637A CN 106250631 B CN106250631 B CN 106250631B
- Authority
- CN
- China
- Prior art keywords
- node
- failure
- test
- matrix
- bayesian network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Analysis (AREA)
- Probability & Statistics with Applications (AREA)
- Algebra (AREA)
- Computer Hardware Design (AREA)
- Computational Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Geometry (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The present invention is a kind of based on failure-test correlation matrix method for diagnosing faults, comprising the following steps: 1) determines the system failure-test correlation matrix D by the dependency graph representation model of system;2) Bayesian network node is established;3) according to each element value Connection Step 2 in matrix D) in two layers of Bayesian network node;4) the Bayesian network root node setting conditional probability table to indicate failure;5) the Bayesian network leaf node setting conditional probability table to indicate test item;6) evidence variable is arranged according to the test result of each test item;7) diagnostic result is obtained;It is proposed by the present invention based on failure-test correlation matrix method for diagnosing faults, the logical relation between failure and test can be described with succinct intuitive figure, the probability that each failure of effective integration system occurs, it overcomes existing direct application correlation matrix and carries out defect existing for method for diagnosing faults, keep diagnostic analysis result more quantitative than correlation matrix.
Description
Technical field
The invention belongs to the Analysis on Fault Diagnosis fields of device integration system, and in particular to one kind is based on failure-test phase
Close the method for diagnosing faults of matrix.
Background technique
With extensive use of the new and high technology in various kinds of equipment, synthesization, informationization, the systematization of device integration system
Degree is higher and higher.While improving systems technology performance, since complexity and unit volume include number of devices
Increase, the factor for influencing system normal operation also increases therewith, causes a possibility that generating trouble or failure increasing.Cause
This, a kind of method of fast and effeciently removal system failure is increasingly important.
Currently, there are a large amount of system fault diagnosis method methods, including it is correlation model, fuzzy theory, expert system, artificial
Neural network, Bayesian network etc..Wherein, the system diagnosis method based on correlation model is widely applied in engineering,
Especially in aerospace field.Correlation models are come across earliest in the 1950s, by DSI company, U.S. founder De
This theory is applied to system fault diagnosis first since the 1960s by Paul.Correlation models are made with Correlation Reasoning
Based on, application failure-test matrix (abbreviation matrix D) carries out fault detection and diagnosis to system or equipment.Therefore based on correlation
Property model method for diagnosing faults, core be the analysis to D matrix handle, to obtain diagnosis.However, D matrix
The middle each element qualitative expression incidence relation of failure and test does not consider the probability that each failure occurs.When multiple events
When barrier can carry out state-detection by same test item, each fault rate is not considered, and only will be unable to judgement by D matrix most has
The failure that may occur, diagnostic method failure.For example, train is made of element A and B, two elements can all be carried out by test T
Detection, i.e.,When working as T=1, i.e. test item T test result exception, it can determine whether that A and B are likely to occur by matrix D
Failure.But the probability of malfunction information of A and B is not merged in view of matrix D, therefore can not specifically judge which element of A and B more has can
It can break down.
It is combined closely based on the method for diagnosing faults of Bayesian network with probability theory, it being capable of effective integration constituent system components
Probability of malfunction information, therefore can effectively make up the deficiency that direct application matrix D carries out fault diagnosis.Currently used for fault diagnosis
Bayesian network is made of two-layer node.Bayesian network for fault diagnosis is also known as failure Bayesian network, and first
Layer is failure cause layer, which expresses the failure of constituent system components, and node state is that failure occurs or do not occur two
Kind;The second layer is failure symptom layer, and each node expresses the phenomenon of the failure as caused by failure cause, and phenomenon of the failure generally passes through survey
Examination item measures, which is that phenomenon of the failure occurs or do not occur two kinds.Each node is with one in Bayesian network
It is associated to open conditional probability table.Wherein, with the first-level nodes associated conditional probability tables, quantitative expression constituent system components
Probability of malfunction;With the associated conditional probability table of the second node layer, incidence relation of the quantitative expression with first layer connected node.So
And the incidence relation (i.e. failure and test incidence relation) between two-layer node is difficult to directly give, which limits base
In extensive use of the method for diagnosing faults in engineering of Bayesian network.
To sum up, traditional method for diagnosing faults based on correlation matrix D, which exists, is unable to the event of effective integration system component units
There is the shortcomings that being difficult to directly give failure and test incidence relation in the shortcomings that hindering probability, traditional failure Bayesian network.With
Upper two disadvantages limit the extensive use of each self-diagnosing method in systems in practice, therefore seek one kind and can effectively overcome
The new method for diagnosing faults of existing two methods disadvantage is of great significance for system fault diagnosis.
Summary of the invention
Make diagnostic result not accurate enough to solve the problems, such as that existing failure-test matrix is difficult to merge probability of malfunction, together
When in order to overcome malfunctioning node and test node incidence relation in existing failure Bayesian network to be difficult to the defect obtained, the present invention
It proposes a kind of based on failure-test correlation matrix method for diagnosing faults.
The technical solution of the invention is as follows: the present invention includes three cores: A is by failure-test correlation matrix structure
The method for building two layers of Bayesian Network Topology Structures;The setting method of each node condition probability tables of two layers of Bayesian network of B;C sheet
Invention propose based on failure-test correlation matrix diagnostic method process.
Realization process of the invention the following steps are included: a kind of based on failure-test correlation matrix method for diagnosing faults,
It is characterized by: the described method comprises the following steps:
1) system failure-test correlation matrix D is determined by the dependency graph representation model of system;Wherein, in matrix D, often
The failure of only element in a line correspondence system, unique test item in each column correspondence system;That is, for being formed by m
Unit, the system of n test item, correlation matrix D have m row n column data;
2) Bayesian network node is established;The failure for being followed successively by the corresponding system component units of each row in correlation matrix D is built
Unique corresponding node is found, the root node as Bayesian network;It is followed successively by correlation matrix D and respectively arranges corresponding test item foundation only
One corresponding node, the leaf node as Bayesian network;
3) according to each element value Connection Step 2 in matrix D) in two layers of Bayesian network node;The operation of line walks
It is rapid as follows:
For the i-th row jth column element d of matrix Dij(1≤i≤m, 1≤j≤n), if dij=1, then by failure FiIn step
2) corresponding Bayesian network root node line is directed toward test item T injThe corresponding Bayesian network leaf node in step 2);If
dij=0, it is operated without any line;
I and j is traversed, the line operation is successively carried out to each element in matrix D;
4) the Bayesian network root node setting conditional probability table to indicate failure;
5) the Bayesian network leaf node setting conditional probability table to indicate test item;If leaf node Tj(1≤j≤n) has k
A father node, then father node combinations of states has 2kKind situation;And when having any node state in k father node is 1, leaf node
TjThe probability that state is 1 is set as 1.0;When k father node state is 0, leaf node TjThe probability that state is 1 is set
It is set to 0.0;
6) evidence variable is arranged according to the test result of each test item: the leaf node T passed through for testjSetting state Tj
=0, leaf node T unsanctioned for test resultjSetting state Tj=1,1≤j≤n;
7) diagnostic result is obtained;The method for obtaining diagnostic result is as follows:
Calculate Bayesian network posterior probability Pr (Fi=1 | T=t) (1≤i≤m, Fi∈ F), knot is calculated according to posterior probability
Fruit is from big to small to failure FiIt is ranked up, according to FiRanking results are successively carried out troubleshooting to system.
Above-mentioned steps 1) the specific step of the system failure-test correlation matrix D determined by the dependency graph representation model of system
Suddenly it is:
1.1] for m unit, the system of n test item establishes two-dimensional correlation matrix D, and matrix D size is m
Row n is arranged, the failure of unique component units in each row correspondence system dependency graph representation model, each column correspondence system correlation models
In unique test item;Initializing each element in D is 0;
1.2] i=1 is enabled, wherein i representing matrix D the i-th row data, FiThe corresponding system group of representing matrix D the i-th row data
At the failure of unit;
1.3] in dependency graph representation model, all boxes and circle are regarded as node, all node states, which are arranged, is
0, that is, set Fi=0 and Tj=0;
It 1.4] is F in number in dependency graph representation modeliNode in place a token, setting number is FiNode
State is 1, i.e. Fi=1, and whole nodes that the token can reach are searched for by searching algorithm along directed line;
1.5] in dependency graph representation model, if node FiThe node that can be reached is the circle for indicating test item, that
It determines respective column j of the test item in matrix D represented by the circle, and sets d in matrix Dij=1;
1.6] 1:i=i+1 is added certainly to i;If i < m+1 then goes to step 1.3];Otherwise, each element has determined in matrix D
Finish, algorithm terminates.
Step 1.4] searching algorithm specifically includes the following steps:
1.4.1 an empty linear list L] is established;
1.4.2 it] using node where token as starting point, is directed toward along directed line segment and determines whether the adjacent node of starting point meets two
A condition: the adjacent node state of starting point is that the adjacent node of 0, starting point is not present in linear list L;For meeting two conditions
The adjacent section of starting point is stored in linear list L;
1.4.3] if linear list L is sky, the token in dependency graph representation model is rejected, terminates searching algorithm;Otherwise, exist
In dependency graph representation model, token is moved into the node that linear list L end data indicates, linear list L end data is indicated
The state of node be changed to 1 from 0, delete linear list L end data, go to step 1.4.2];
Above-mentioned steps 4) if failure FiProbability of happening be Pi, then it represents that failure FiNode condition probability tables parameter setting
It is as follows:
Root node Fi Pr(Fi)
0 1.0-Pi
1 Pi。
The invention has the benefit that
It is proposed by the present invention based on failure-test correlation matrix method for diagnosing faults, pass through and constructed according to correlation matrix
Bayesian network, not only succinct intuitive figure can be used to describe the logical relation between failure and test, it is often more important that effectively
The probability that each failure of system occurs has been merged, existing direct application correlation matrix has been overcome and carries out existing for method for diagnosing faults
Defect, so that diagnostic analysis result is more quantitative than correlation matrix;
It is proposed by the present invention based on failure-test correlation matrix method for diagnosing faults simultaneously, it overcomes and is currently used for event
Incidence relation (i.e. failure and test incidence relation) between two layers of Bayesian network node of barrier diagnosis is difficult to directly give tired
Difficulty can promote extensive use of the method for diagnosing faults based on Bayesian network in engineering.
Finally, proposed by the present invention based on failure-test correlation matrix method for diagnosing faults effective integration Correlation Moment
Battle array and the respective advantage of Bayesian network can quantitatively calculate the probability that each failure occurs in known test result, and give
System maximum possible is explained to support fault diagnosis out, and the research for carrying out the strong failure diagnosis tool of maneuverability, which has, uses for reference
Meaning can assist the intelligence for realizing fault diagnosis, improve diagnosis efficiency.
Detailed description of the invention
Fig. 1 is that the present invention is based on failure-test correlation matrix method for diagnosing faults flow charts;
Fig. 2 is the dependency graph representation model of system in the present invention-example;
Fig. 3 is the Bayesian network of equal value of dependency graph representation model in the present invention-example.
Specific embodiment
The present invention is a kind of based on failure-test correlation matrix method for diagnosing faults.It is right with reference to the accompanying drawings of the specification
Technical solution of the present invention is described in further detail.The specific implementation process of the embodiment of the present invention is as follows:
Step 1, the system failure-test correlation matrix D is obtained by the dependency graph representation model of system.
The failure of only element in the every a line correspondence system of matrix D, unique test in each column correspondence system
?;That is, for having m component units, the system of n test item, correlation matrix D has m row n column data.The each group of the system
It only include normal and failure two states at unit, the event of the corresponding system component units of the i-th row (1≤i≤m) in matrix D
Barrier uses FiIt indicates, jth arranges (1≤j≤n) corresponding test item TjIt indicates, matrix D representation method is as follows:
I-th row jth column data d in matrix DijShow failure FiWith test item TjCorrelation.Work as TjFailure can be measured
FiFault message when, then test item TjWith failure FiCorrelation, dij=1;Work as TjFailure F cannot be measurediFault message when, then
Test item TjWith failure FiIt is unrelated, dij=0.
System refer to by interact and complementary several units be combined into realize the organic of specific function
It is whole.Wherein, unit can be element, equipment, subsystem and the module with specific function.The correlation models are expression
The model of cell failure and test correlation logic relationship in system, including dependency graph representation model and correlation mathematical model two
Kind form.The dependency graph representation model can intuitively show the relationship between unit and test item;And correlation mathematical modulo
Type describes the correlation between unit and test with failure-test matrix D, and matrix D is also known as correlation matrix.
Specifically, the representation method of the dependency graph representation model is respectively to form Elementary Function and structure conjunction in systems
After reason divides, on the basis of functional block diagram, functional information stream direction and each component units interconnected relationship are clearly indicated,
And position and the number for understanding test item are marked, to show the correlative relationship of each component units Yu each test item.Dependency graph
Representation model includes three classes graphic element: box, circle, directed line segment.Wherein, box illustrates each component units in system,
Circle illustrates that the test item for each unit in test macro, the functional information that directed line segment shows each unit in system pass
Pass direction.In addition, each box is number, each circle with the failure of its system component units indicated in dependency graph representation model
Circle is number with the test item that it is indicated.
Dependency graph representation model is the basis for establishing matrix D.It is described to obtain correlation matrix D's by dependency graph representation model
Algorithm is as follows:
1. the system of n test item establishes two-dimensional correlation matrix D, and matrix D size is m row n for m unit
It arranges, the failure of unique component units in each row correspondence system dependency graph representation model, in each column correspondence system correlation models
Unique test item.Initializing each element in D is 0.
2. enabling i=1.Wherein, the i-th row of i representing matrix D data, FiThe corresponding system composition of representing matrix D the i-th row data
The failure of unit.
3. all boxes and circle are regarded as node in dependency graph representation model, it is 0 that all node states, which are arranged,
Set Fi=0 and Tj=0.
4. being F in number in dependency graph representation modeliNode in place a token, setting number is FiNode shape
State is 1, i.e. Fi=1, and all nodes that the token can reach are searched for along directed line segment.The searching algorithm is in correlation
Property diagram model in include 3 steps, it is specific as follows:
I) an empty linear list L is established;
Ii it) using node where token as starting point, is directed toward along directed line segment and determines whether the adjacent node of starting point meets two
Condition: the adjacent node state of starting point is that the adjacent node of 0, starting point is not present in linear list L.For meeting rising for two conditions
The adjacent node of point is stored in linear list L.
The adjacent node refers between two nodes, is directed toward the oriented of another node if there is by a node
Line segment, then the latter node is the adjacent node of previous node;
Iii) if linear list L is sky, the token in dependency graph representation model is rejected, end step 4. calculate by the search
Method;Otherwise, in dependency graph representation model, token is moved into the node that linear list L end data indicates, by the end linear list L
The state for the node that end data indicates is changed to 1 from 0, deletes linear list L end data, goes to step ii).
Token refers to that one of dependency graph representation model marks, and the node that searching algorithm is being searched for is marked, and leads to
Often indicated with stain.
Complete step i)~iii) after, the node that state is 1 is node FiAccessibility all nodes.
5. in dependency graph representation model, if node FiThe node that can be reached is the circle for indicating test item, then
It determines respective column j of the test item represented by the circle in matrix D, and sets d in matrix Dij=1.
6. adding 1:i=i+1 certainly to i.If i < m+1, then step is gone to 3.;Otherwise, each element determination finishes in matrix D, calculates
Method terminates.
Step 3.~5. constitute one by variable i control circulation: as i=1, step 3.~5. by the 1st time execute, can
Complete the determination of the 1st row data of matrix D;As i=2, step 3.~5. by the 2nd time execute, achievable the 2nd row data of matrix D
Determination;And so on, step 3.~5. by the 3rd, 4 ... m times execution, be sequentially completed matrix D the 3rd, 4 ... m row data really
It is fixed.Wherein, step be 6. by variable i rate-determining steps 3.~5. carry out circulate operation sentence: as i < m+1,3. to step
~5. operated, that is, enter and recycles next time;Otherwise, step 3.~be 5. no longer performed, i.e., circulation terminate.
When after circulation terminates, step 1.~6. terminate, so far complete failure-test correlation matrix D determination.
In the present embodiment, as shown in dependency graph representation model in Fig. 2, system includes 5 component units, 2 test items.
Each component units include two states: normal and failure.The failure of 5 component units uses F respectively1、F2、F3、F4And F5It indicates,
And F1~F5Probability of malfunction be followed successively by 0.001,0.002,0.003,0.004 and 0.005, furthermore test item uses variable respectively
T1And T2It indicates.T1It is directly used in test failure F1And F5, work as F1And F5Middle Arbitrary Fault occurs, test item T1Test result not
Pass through;T2It is directly used in test failure F3, work as F3Failure occurs, test item T2Test result do not pass through.
In the present embodiment, determine that the implementation process of correlation matrix D is as follows by dependency graph representation model shown in Fig. 2:
1. there are 5 component units and 2 test items for the dependency graph representation model of system shown in Figure 2, then m=5 and n=
2 set up.The correlation matrix D with 5 row, 2 column data, the i-th row of matrix D (1≤i≤m, m=5) are established for dependency graph representation model
With the failure F of component unitsiIt is corresponding, the jth column (1≤j≤n, n=2) and test item T of matrix DjIt is corresponding, the matrix D of foundation
Expression-form is as follows:
In matrix D, dijValue is 0 or 1.Initialize each element d in DijFor 0 (1≤i≤m, 1≤j≤n):
2. enabling i=1.Wherein, the i-th row of i representing matrix D data, FiThe corresponding system composition of representing matrix D the i-th row data
The failure of unit.
3. in dependency graph representation model shown in Fig. 2, F is arranged1=0, F2=0, F3=0, F4=0 and F5=0, setting
T1=0 and T2=0;
4. being F in number in dependency graph representation model shown in Fig. 2iNode in place a token, token is with black
Point indicates, sets Fi=1, i=1.F is searched for along directed line segmentiWhole nodes that middle token can reach.Steps are as follows:
I) an empty linear list L is established;
Ii it) using node where token as starting point, is directed toward along directed line segment and determines whether the adjacent node of starting point meets two
Condition: the adjacent node state of starting point is that the adjacent node of 0, starting point is not present in linear list L.For meeting rising for two conditions
The adjacent node of point is stored in linear list L.
In Fig. 2, node FiIt is placed with token, then with FiFor starting point, it is directed toward along directed line segment and determines FiAdjacent node be
Two conditions of no satisfaction: FiAdjacent node state be 0, FiAdjacent node be not present in linear list L.In Fig. 2, meet
The node of two conditions is F2And T2, two nodes are stored in linear list L, then L={ F2, T2}。
Iii) if linear list L is sky, the token in dependency graph representation model is rejected, terminates searching algorithm;Otherwise, in phase
In closing property diagram model, token is moved into the node that linear list L end data indicates, linear list L end data is indicated
The state of node is changed to 1 from 0, deletes linear list L end data, goes to step ii).
In the present embodiment, linear list L={ F2, T2, end T2, then in Fig. 2, by token from node F1It moves into
Node T2, by T2=0 is changed to T2=1, delete linear list L end data T2, L={ F at this time2, go to step ii);
It is transferred to step ii) after, to be placed with the node T of token2For starting point, there is no meet step ii) described two
The node of part, thus be not required to execute step ii) in storing step, be directly entered step iii);
Enter step iii) after, linear list L={ F at this time2, end F2, then in Fig. 2, by token from node T2It moves
Ingress F2, by F2=0 is changed to F2=1, delete linear list L end data F2, chained list is sky at this time, goes to step ii);
It is transferred to step ii) after, to be placed with the node F of token2For starting point, meet step ii) sections of described two conditions
Point is F3, then by F3It is stored in linear list L, then L={ F3};
Enter step iii) after, linear list L={ F at this time3, end F3, then by token from node F2Move into node
F3, by F3=0 is changed to F3=1, delete linear list L end data F3, chained list is sky at this time, goes to step ii).
And so on, repeat step ii) and iii), until linear list L is sky in step iii), searching algorithm terminates.
At the end of searching algorithm, the state of each node in Fig. 2 are as follows: F1=1, F2=1, F3=1, F4=0, F5=0, T1=1, T2=1
4. described, the step i)~iii according to step) after, the node that state is 1 is FiAll sections that can be reached
Point.That is, in the present embodiment, node Fi(i=1) reachable node has F1、F2、F3、F4、T1And T2。
5. in dependency graph representation model shown in Fig. 2, node Fi(i=1) circle for the expression test item that can be reached
For T1And T2, respective column of two test items in matrix D is followed successively by the 1st and the 2nd column, therefore d is set in matrix Di1=1 and di2
=1, i=1.
6. adding 1:i=i+1 certainly to i.If i < m+1, step is gone to 3.;Otherwise, each element determination finishes in matrix D, calculates
Method terminates.
In the present embodiment, i=1 at this time, so far step 3.~be 5. finished by the 1st time, the 1st row data in matrix D
It has determined that and finishes, be followed successively by d11=1 and d12=1;
To i from after adding 1: i=2.I < m+1 (m=5) set up, therefore, step need to be gone to 3., the 2nd execution step 3.~
⑤;When step 3.~be 5. finished by the 2nd time after, it may be determined that the 2nd row data are followed successively by d in matrix D21=1 and d22=0;
To i from after adding 1: i=3.I < m+1 (m=5) set up, therefore, step need to be gone to 3., the 3rd execution step 3.~
⑤;When step 3.~be 5. finished by the 3rd time after, it may be determined that the 3rd row data are followed successively by d in matrix D31=1 and d32=0;
And so on, i is carried out from add operation: when i < m+1 is set up (m=5), going to step 3., i-th executes step
3.~5., complete the determination of the i-th row data in matrix D;Otherwise end step 3.~the 5. circulation.
When after circulation terminates, step 1.~6. terminate, so far complete failure-test correlation matrix D determination, determine
Correlation matrix D is as follows:
Step 2, Bayesian network node is established;It is followed successively by the event of the corresponding system component units of each row in correlation matrix D
Barrier establishes unique corresponding node, the root node as Bayesian network;It is followed successively by correlation matrix D and respectively arranges corresponding test item and build
Unique corresponding node is found, the leaf node as Bayesian network.
The Bayesian network node of the foundation, if the failure of the corresponding system component units of its representing matrix the i-th row of D,
The then Bayesian network root node number F to establishi;If its representing matrix D jth arranges corresponding test item, for the shellfish of foundation
This network leaf node code T of leafj。
The Bayesian network node of foundation only includes 0 and 1 two states, root node FiState is 0 expression failure FiIt does not send out
It is raw, root node FiState is 1 expression failure FiOccur;Leaf node TjState is 0 expression test item TjMeasured result passes through or just
Often, leaf node TjState is 1 expression test item TjMeasured result does not pass through or exception.
In the present embodiment, correlation matrix D has 5 row data, is corresponding in turn to system component units failure F1、F2、F3、F4With
F5, it is F1~F5Corresponding node is established respectively, the root node as Bayesian network;Correlation matrix D has 2 column datas, is corresponding in turn to
Test item T1And T2, it is T1And T2Corresponding node is established respectively, the leaf node as Bayesian network.
As shown in figure 3, the root node established is 5 total, corresponding to the row data in matrix D, the number of root node is successively
For node F1~F5;The leaf node of foundation is 2 total, and corresponding to the column data in matrix D, the number of leaf node is followed successively by node
T1And T2.The root node and leaf node of foundation are two condition: root node Fi(1≤i≤5) state is 0 expression failure FiDo not occur,
Root node FiState is 1 expression failure FiOccur;Leaf node Tj(1≤j≤2) state is 0 expression test item TjMeasured result passes through
Or normal, leaf node TjState is 1 expression test item TjMeasured result does not pass through or exception.
Step 3, according to two layers of Bayesian network node in matrix D in each element value Connection Step 2;The line
Operating procedure is as follows:
For the i-th row jth column element d of matrix Dij(1≤i≤m, 1≤j≤n), if dij=1, then by failure FiIn step
2) corresponding Bayesian network root node line is directed toward test item T injThe corresponding Bayesian network leaf node in step 2);If
dij=0, it is operated without any line;
I and j is traversed, the operation of line described in step 3) is successively carried out to each element in matrix D.
In the present embodiment, according to described in step 3, concrete operations are as follows:
The 1st column element d of the 1st row of matrix D11=1, by node F in Fig. 31Line is directed toward node T1;
The 2nd column element d of the 1st row of matrix D12=1, by node F in Fig. 31Line is directed toward node T1;
The 1st column element d of the 2nd row of matrix D21=1, by node F in Fig. 32Line is directed toward node T1;
The 1st column element d of the 3rd row of matrix D31=1, by node F in Fig. 33Line is directed toward node T1;
The 2nd column element d of the 4th row of matrix D42=1, by node F in Fig. 34Line is directed toward node T2;
The 2nd column element d of the 5th row of matrix D52=1, by node F in Fig. 35Line is directed toward node T2;
Remaining element of matrix D is 0, is operated without any line.
Step 4, conditional probability table is set for the Bayesian network root node in step 2.The conditional probability table of each root node
In include 2 conditional probability values, each parameter is shown in Table 1 in conditional probability table.In table 1, as root node FiWhen being 0, conditional probability value
Pr(Fi)=1.0-Pi;As root node FiWhen being 1, conditional probability value Pr (Fi)=Pi.Wherein PiFor the corresponding system of the i-th row of matrix D
The probability of malfunction of system component units.In conditional probability table, FiValue 0 and 1 respectively indicates failure FiDo not occur and occur two kinds of shapes
State.
1 root node conditional probability table of table
Root node Fi | Pr(Fi) |
0 | 1.0-Pi |
1 | Pi |
Conditional probability table refers to the corresponding conditional probability table of each node in Bayesian network.In conditional probability table
Each row of data include node conditional probability value corresponding with the combinations of states of father node, combinations of states two.Conditional probability table
The meaning of middle each row of data expression are as follows: when father node state is listed state in table, node is that listed shape probability of state is equal to table
In listed conditional probability value.
In the present embodiment, the probability of malfunction of the corresponding system component units of the 1st~5 row of matrix D be followed successively by 0.001,
0.002,0.003,0.004 and 0.005.The root node conditional probability setting method according to step 4, Bayesian network root
Data in the conditional probability table of node F1 are as follows: Pr (F1=0)=0.999, Pr (F1=1)=0.001;Root node F2Condition it is general
Data in rate table are as follows: Pr (F2=0)=0.998, Pr (F2=1)=0.002;Root node F3Conditional probability table in data are as follows: Pr
(F3=0)=0.997, Pr (F3=1)=0.003;Root node F4Conditional probability table in data are as follows: Pr (F4=0)=0.996,
Pr(F4=1)=0.004;Root node F5Conditional probability table in data are as follows: Pr (F5=0)=0.995, Pr (F5=1)=
0.005.Root node F1、F2、F3、F4And F5Conditional probability table it is as shown in table 3.
Each root node conditional probability table in 3 Bayesian network of table
Step 5, the Bayesian network leaf node setting conditional probability table to indicate test item;If leaf node Tj(1≤j≤
N) there is k father node, then father node combinations of states has 2kKind situation;And when having any node state in k father node is 1,
Leaf node TjThe probability that state is 1 is set as 1.0;When k father node state is 0, leaf node TjState is 1
Probability is set as 0.0.
For the present embodiment, in Fig. 3, leaf node T1There are 3 father node { F1,F2,F3, when any father node state is 1
When, the probability that leaf segment dotted state is 1 is 1.0;When 3 father node states are 0, the probability that leaf segment dotted state is 1 is 0.0.
In table 4, as j=1, table 4 is leaf node T1Conditional probability table;
In Fig. 3, leaf node T2There are 3 father node { F1,F4,F5, when any father node state is 1, leaf segment dotted state
Probability for 1 is 1.0;When 3 father node states are 0, the probability that leaf segment dotted state is 1 is 0.0.In table 4, work as j=2
When, table 4 is leaf node T2Conditional probability table.So far, dependency graph representation model equivalence Bayesian network shown in Fig. 2 is completed
The building of network.
4 leaf node conditional probability table of table
Step 6, evidence variable is arranged according to the test result of each test item: the leaf node T passed through for testjShape is set
State Tj=0, leaf node T unsanctioned for test resultjSetting state Tj=1,1≤j≤n.
The evidence variable refers in Bayesian network, variable known to state.According to the test knot of each test item
Fruit, the state of each test item are known --- state is to pass through or do not pass through, therefore in the present invention, evidence variable includes shellfish
All leaf nodes of this network of leaf, i.e. Bayesian network leaf node T1、T2…Tn.Leaf node T1、T2…TnConstitute evidence variables set
T is closed, each variable-value is denoted as t in set T.
It is known in the present embodiment to test T1Do not pass through, tests T2Pass through, then evidence variables collection T={ T1, T2, set T
In each variable-value t={ T1=1, T2=0 }.
Step 7, diagnostic result is obtained;The method for obtaining diagnostic result is as follows:
Calculate Bayesian network posterior probability Pr (Fi=1 | T=t) (1≤i≤m, Fi∈ F), knot is calculated according to posterior probability
Fruit is from big to small to failure FiIt is ranked up, according to FiRanking results are successively carried out troubleshooting to system.
Bayesian network posterior probability problem can be solved using existing software or algorithm routine.In the present embodiment,
Posterior probability Pr (F is successively solved using Bayesian network analysis software SamIami=1 | T1=1, T2=0) (1≤i≤m, m=
5), and according to posterior probability result from big to small to a faulty FiSequence.As a result as follows:
①Pr(F3=1 | T1=1, T2=0)=0.6072874
②Pr(F2=1 | T1=1, T2=0)=0.5951417
③Pr(F1=1 | T1=1, T2=0)=0.0
Pr(F4=1 | T1=1, T2=0)=0.0
Pr(F5=1 | T1=1, T2=0)=0.0
According to posterior probability ranking results F3It is most likely to occur failure, followed by F2, remaining failure F1、F4And F5?
It can not occur.So in known test result T1=1, T2When=0, the sequence of system troubleshooting is F3、F2。
So far, the fault diagnosis to Fig. 2 system is completed by above 7 steps.Next in conjunction with the embodiments, from following
3 aspects illustrate beneficial effect caused by the present invention.
Firstly, the present embodiment diagnostic result has reasonability.In the case where testing completely reliable situation, T2Test passes through, root
It can determine whether according to directions of information flow shown in directed line segment in Fig. 3, F1、F4、F5It will not necessarily break down, the posteriority in embodiment
Probability problem solving result is consistent with the fact;T1Test does not pass through, and can be sentenced according to the directions of information flow of directed line segment in Fig. 3
It is disconnected, F1、F2、F3It is likely to occur failure, and by T2Test is by having excluded F1A possibility that failure, therefore in evidence variable
{T1=1, T2=0 } under support, only F2And F3May break down, the posterior probability problem solving in embodiment the result shows that
The probability that the two respectively breaks down is greater than 0.5, is true to life.
Secondly, the probability that each failure of method effective integration proposed by the present invention occurs, diagnostic result is more quantitative, relative to
The method for being directly based upon correlation matrix D has more superiority at quantitative aspect.Event is carried out to system using correlation matrix D directly below
Barrier analysis.Evidence variable { T is known1=1, T2=0 }, match with the 2nd row and the 3rd row data in matrix D, this shows at this
Under evidence variable is supported, the 2nd row of matrix D and the corresponding failure F of the 3rd row2And F3It is likely to occur, but can not further deduce
Who is more likely to break down the two, and a possibility that accidentally tearing open when to increasing troubleshooting, delay system restores the time worked normally.
And diagnostic method proposed by the present invention, in known evidence variable { T1=1, T2It=0 }, being capable of each failure of effective integration in the case where
The probability of generation obtains F by the solution of posterior probability problem3Compare F2Be more likely to occur failure this as a result, in turn can be to
Troubleshooting sequence F out3And F2, a possibility that accidentally tearing open is reduced, system is improved and restores the time worked normally.
Finally, proposed by the present invention based on failure-test correlation matrix method for diagnosing faults, in step 3 directly according to
The root node and leaf node that two layers of Bayesian network is connected according to each element value in correlation matrix D, then utilize in step 4
Conditional probability table has carried out quantitative expression to the bonding strength of root node in step 3 and leaf node, overcomes existing for event
Incidence relation is difficult to the difficulty directly given between failure classes node and test class node in the Bayesian network of barrier diagnosis, can promote
Extensive use based on the method for diagnosing faults of Bayesian network in engineering can assist the intelligence for realizing fault diagnosis, mention
High diagnosis efficiency has reference to exploitation system fault diagnosis tool.
Claims (4)
1. a kind of based on failure-test correlation matrix method for diagnosing faults, it is characterised in that: the method includes following steps
It is rapid:
1) system failure-test correlation matrix D is determined by the dependency graph representation model of system;Wherein, in matrix D, every a line
The failure of only element in correspondence system, unique test item in each column correspondence system;That is, for there is m component units,
The system of n test item, correlation matrix D have m row n column data;
2) Bayesian network node is established;The failure for being followed successively by the corresponding system component units of each row in correlation matrix D is established only
One corresponding node, the root node as Bayesian network;Be followed successively by correlation matrix D respectively arrange corresponding test item establish it is unique right
Node is answered, the leaf node as Bayesian network;
3) according to each element value Connection Step 2 in matrix D) in two layers of Bayesian network node;The operating procedure of line is such as
Under:
For the i-th row jth column element d of matrix Dij, 1≤i≤m, 1≤j≤n, if dij=1, then by failure FiIn step 2)
Corresponding Bayesian network root node line is directed toward test item TjThe corresponding Bayesian network leaf node in step 2);If dij=
0, it is operated without any line;
I and j is traversed, the line operation is successively carried out to each element in matrix D;
4) the Bayesian network root node setting conditional probability table to indicate failure;
5) the Bayesian network leaf node setting conditional probability table to indicate test item;If leaf node Tj, 1≤j≤n has k father
Node, then father node combinations of states has 2kKind situation;And when having any node state in k father node is 1, leaf node TjShape
The probability that state is 1 is set as 1.0;When k father node state is 0, leaf node TjThe probability that state is 1 is set as
0.0;
6) evidence variable is arranged according to the test result of each test item: the leaf node T passed through for testjSetting state Tj=0,
Leaf node T unsanctioned for test resultjSetting state Tj=1,1≤j≤n;
7) diagnostic result is obtained;The method for obtaining diagnostic result is as follows:
Calculate Bayesian network posterior probability Pr (Fi=1 | T=t), 1≤i≤m, Fi∈ F, according to posterior probability calculated result from
It arrives greatly small to failure FiIt is ranked up, according to FiRanking results are successively carried out troubleshooting to system, wherein T refers to leaf node T1、
T2…TnThe evidence variables collection of composition, each variable-value is denoted as t in T.
2. according to claim 1 based on failure-test correlation matrix method for diagnosing faults, it is characterised in that: described
Step 1) determines that the system failure-test correlation matrix D is comprised the concrete steps that by the dependency graph representation model of system:
1.1] for m unit, the system of n test item establishes two-dimensional correlation matrix D, and matrix D size is m row n
It arranges, the failure of unique component units in each row correspondence system dependency graph representation model, in each column correspondence system correlation models
Unique test item;Initializing each element in D is 0;
1.2] i=1 is enabled, wherein i representing matrix D the i-th row data, FiThe corresponding system component units of representing matrix D the i-th row data
Failure;
1.3] in dependency graph representation model, all boxes and circle are regarded as node, it is 0 that all node states, which are arranged, i.e.,
Set Fi=0 and Tj=0;
It 1.4] is F in number in dependency graph representation modeliNode in place a token, setting number is FiNode state
It is 1, i.e. Fi=1, and whole nodes that the token can reach are searched for by searching algorithm along directed line;
1.5] in dependency graph representation model, if node FiThe node that can be reached is the circle for indicating test item, then it is determined that
Respective column j of the test item represented by the circle in matrix D, and d is set in matrix Dij=1;
1.6] 1:i=i+1 is added certainly to i;If i < m+1 then goes to step 1.3];Otherwise, each element determination finishes in matrix D, calculates
Method terminates.
3. according to claim 2 based on failure-test correlation matrix method for diagnosing faults, it is characterised in that: step
1.4] searching algorithm specifically includes the following steps:
1.4.1 an empty linear list L] is established;
1.4.2 it] using node where token as starting point, is directed toward along directed line segment and determines whether the adjacent node of starting point meets two items
Part: the adjacent node state of starting point is that the adjacent node of 0, starting point is not present in linear list L;For meeting the starting point of two conditions
Adjacent section be stored in linear list L;
1.4.3] if linear list L is sky, the token in dependency graph representation model is rejected, terminates searching algorithm;Otherwise, in correlation
Property diagram model in, will token move into linear list L end data indicate node in, by linear list L end data indicate section
The state of point is changed to 1 from 0, deletes linear list L end data, goes to step 1.4.2].
4. according to claim 3 based on failure-test correlation matrix method for diagnosing faults, it is characterised in that: described
If step 4) failure FiProbability of happening be Pi, then it represents that failure FiNode condition probability tables parameter setting it is as follows:
Root node Fi Pr(Fi)
0 1.0-Pi
1 Pi。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610629637.6A CN106250631B (en) | 2016-08-03 | 2016-08-03 | It is a kind of based on failure-test correlation matrix method for diagnosing faults |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610629637.6A CN106250631B (en) | 2016-08-03 | 2016-08-03 | It is a kind of based on failure-test correlation matrix method for diagnosing faults |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106250631A CN106250631A (en) | 2016-12-21 |
CN106250631B true CN106250631B (en) | 2019-03-08 |
Family
ID=57606667
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610629637.6A Expired - Fee Related CN106250631B (en) | 2016-08-03 | 2016-08-03 | It is a kind of based on failure-test correlation matrix method for diagnosing faults |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106250631B (en) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108375742A (en) * | 2018-02-05 | 2018-08-07 | 哈尔滨工业大学 | A kind of flyback sourse and test method with testable parameter and fault diagnosis functions |
CN109298706A (en) * | 2018-11-02 | 2019-02-01 | 中国航空工业集团公司西安飞机设计研究所 | A kind of flight control system method for diagnosing faults based on Bayesian network |
CN109872004B (en) * | 2019-03-08 | 2021-06-04 | 北京工商大学 | Equipment fault prediction and health assessment method based on fuzzy Bayesian network |
CN110479636B (en) * | 2019-07-19 | 2021-11-09 | 深圳市微蓝智能科技有限公司 | Method and device for automatically sorting tobacco leaves based on neural network |
CN111061293B (en) * | 2019-11-12 | 2023-11-17 | 中国运载火箭技术研究院 | Multi-parameter coupled aircraft fault positioning method, aircraft and storage medium |
CN111027721B (en) * | 2019-11-22 | 2023-07-18 | 中国铁道科学研究院集团有限公司通信信号研究所 | System fault positioning method |
CN110989558B (en) * | 2019-12-19 | 2021-03-19 | 浙江中控技术股份有限公司 | Fault diagnosis item processing method and system |
CN113962394A (en) * | 2021-09-10 | 2022-01-21 | 重庆邮电大学 | Automatic generation method of correlation model based fault diagnosis inference machine |
CN114978876B (en) * | 2022-04-21 | 2023-04-07 | 北京航空航天大学 | Function test-oriented Bayesian fault diagnosis method for civil aircraft communication system |
CN117061332B (en) * | 2023-10-11 | 2023-12-29 | 中国人民解放军国防科技大学 | Fault diagnosis method and system based on probability directed graph deep learning |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101814114A (en) * | 2010-01-07 | 2010-08-25 | 北京航空航天大学 | Method for performing fault diagnosis by using model conversion |
CN102184291A (en) * | 2011-05-06 | 2011-09-14 | 北京航空航天大学 | System level fault diagnosis method for full-test non-feedback system |
CN106342306B (en) * | 2011-06-24 | 2013-01-16 | 中国人民解放军国防科学技术大学 | Product test index selection method in undetected situation |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7158958B2 (en) * | 2003-12-24 | 2007-01-02 | The Boeing Company | Automatic generation of baysian diagnostics from fault trees |
-
2016
- 2016-08-03 CN CN201610629637.6A patent/CN106250631B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101814114A (en) * | 2010-01-07 | 2010-08-25 | 北京航空航天大学 | Method for performing fault diagnosis by using model conversion |
CN102184291A (en) * | 2011-05-06 | 2011-09-14 | 北京航空航天大学 | System level fault diagnosis method for full-test non-feedback system |
CN106342306B (en) * | 2011-06-24 | 2013-01-16 | 中国人民解放军国防科学技术大学 | Product test index selection method in undetected situation |
Non-Patent Citations (1)
Title |
---|
基于贝叶斯网络的雷达可靠性评估;贾兴亮;《四川兵工学报》;20100930;第31卷(第9期);56-58 * |
Also Published As
Publication number | Publication date |
---|---|
CN106250631A (en) | 2016-12-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106250631B (en) | It is a kind of based on failure-test correlation matrix method for diagnosing faults | |
Cai et al. | Bayesian networks in fault diagnosis | |
Shakeri et al. | Sequential testing algorithms for multiple fault diagnosis | |
CN106200575B (en) | A kind of robustness control method of the automated manufacturing system based on Petri network | |
CN109633372B (en) | Membrane system-based automatic power system fault diagnosis method | |
Darong et al. | Fault diagnosis for the motor drive system of urban transit based on improved Hidden Markov Model | |
Roth et al. | Black-box identification of discrete event systems with optimal partitioning of concurrent subsystems | |
Darwiche et al. | Exploiting system structure in model-based diagnosis of discrete-event systems | |
CN109241583B (en) | Human-computer interaction system reliability solving method based on Markov | |
CN104504248A (en) | Failure diagnosis modeling method based on designing data analysis | |
CN102662831B (en) | Method of diagnosis strategy optimization for fault tolerant system | |
CN102736616B (en) | Dulmage-Mendelsohn (DM)-decomposition-based measuring point optimal configuration method for closed loop system | |
CN110209118A (en) | With the robustness control method based on Petri network for using multiple types multi-quantity resource and flexible path feature | |
Fotuhi-Firuzabad et al. | A novel approach to determine minimal tie-sets of complex network | |
Kurtoglu et al. | First international diagnosis competition-DXC’09 | |
CN104679945B (en) | System comprehensive estimation method based on colored Petri network | |
CN111190759A (en) | Hybrid diagnosis strategy generation method based on function and fault mode | |
Kolcio et al. | Model-based approach to rover health assessment for increased productivity | |
Sanz et al. | Leakage localization in water networks using fuzzy logic | |
Zhong et al. | A novel anomaly detection method for gas turbines using weight agnostic neural network search | |
CN102788955A (en) | Remaining lifetime prediction method of ESN (echo state network) turbine generator classification submodel based on Kalman filtering | |
US7739213B1 (en) | Method for developing complex probabilistic models | |
Li et al. | Predicting remaining useful life of industrial equipment based on multivariable monitoring data analysis | |
Li* | Selecting KBES development techniques for applications in the construction industry | |
CN105224783A (en) | Based on the method for the Train door system hidden trouble location of petri net |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190308 Termination date: 20190803 |