CN113887452B - Fault diagnosis optimization method based on correlation matrix - Google Patents

Abstract

The invention discloses a fault diagnosis optimization method based on a correlation matrix, which comprises the following steps: 1) According to a target system to be diagnosed and a test scheme, a multi-signal flow diagram model is established; 2) Generating a correlation matrix of faults and tests according to the multi-signal flow diagram model; 3) Judging whether a redundant test of faults exists or not according to the generated correlation matrix, and rejecting the redundant test; wherein the redundancy test is that there are identical indistinguishable columns in the matrix; 4) And constructing fault diagnosis strategies under different search width and depth combinations by adopting a method based on information gain, comparing average test cost under different search widths and depths, and optimally designing the fault diagnosis strategies. The invention can improve the searching efficiency and precision of the fault diagnosis test sequence.

Description

Fault diagnosis optimization method based on correlation matrix

Technical Field

The invention relates to a fault diagnosis technology, in particular to a fault diagnosis optimization method based on a correlation matrix.

Background

With the development of technology, the performance of a large-scale system is continuously improved, meanwhile, the complexity of the system is increased, the difficulty of fault diagnosis of the system is increased, the diagnosis cost is increased, and therefore research and optimization of a fault diagnosis algorithm of the system are urgent.

The purpose of the integrated diagnostics is to detect and isolate all known or averaged faults in the system, meeting the mission requirements of the system at the lowest cost. The system with good testability can greatly reduce the time required by fault detection and isolation, thereby obviously shortening the maintenance time, and reducing the corresponding skill requirements on maintenance personnel, thereby achieving the aims of improving the reliability of the system and reducing the cost of life cycle. And the testability analysis design is developed in the analysis design stage, so that a large amount of modeling time can be saved, and the system design efficiency is improved. The system which is designed completely and is in running use is upgraded and reformed, and is maintained daily, and the testability analysis is needed to obtain an optimal test scheme for guiding maintenance work, so that the timely, accurate and low-cost overhaul and maintenance work is realized.

At present, algorithms for generating fault diagnosis strategies include a dynamic programming algorithm, a greedy algorithm, an ant colony algorithm, a genetic algorithm and the like. The existing fault strategy algorithm has certain defects and can be roughly expressed as follows:

1) Dynamic programming algorithms are not suitable for complex systems.

2) The greedy algorithm can only find the local optimal solution of the system, and cannot guarantee the global optimal solution.

3) The ant colony algorithm and the genetic algorithm cannot ensure that the globally optimal solution can be obtained, and meanwhile, the convergence rate is slow in the later stage of the algorithm, and the locally optimal solution is easy to fall into.

Disclosure of Invention

The invention aims to solve the technical problem of providing a fault diagnosis optimization method based on a correlation matrix aiming at the defects in the prior art.

The technical scheme adopted for solving the technical problems is as follows: a fault diagnosis optimization method based on a correlation matrix comprises the following steps:

1) According to a target system to be diagnosed and a test scheme, a multi-signal flow diagram model is established, and description of correlation between faults possibly occurring in each component of the system and the test is completed; the multiple signal flow graph model includes:

the multiple signal flow graph model includes the following elements:

1.1 A finite set of m+1 system faults, x= { X ₀,x₁,……,x_m }, where X ₀ represents a no fault state, X _i, 1< i < m, represents one of m possible fault states in the system;

1.2 Probability of occurrence of failure p= { p ₀,p₁……,p_m };

1.3 Test points tp= { TP ₁,TP₂,……,TP_q }, each test point containing at least one test t _i;

1.4 Test set t= { t ₁,…,t_j…,t_n }, where the cost of the corresponding test is c= { C ₁,c₂,……,c_n };

2) Generating a correlation matrix of faults and tests according to the multi-signal flow diagram model; the matrix element d _ij of the correlation matrix is a boolean variable, if a fault x _i can be detected by test t _j, then d _ij is 1, otherwise it is 0;

3) Judging whether a redundant test of faults exists or not according to the generated correlation matrix, and eliminating the redundant test if the redundant test exists; wherein the redundancy test is that there are identical indistinguishable columns in the matrix;

4) And constructing fault diagnosis strategies under different search width and depth combinations by adopting a method based on information gain, comparing average test cost under different search widths and depths, and optimally designing the fault diagnosis strategies.

According to the above scheme, the step 4) specifically comprises the following steps:

4.1 Calculating an average information increment for each test;

Selecting a current fault state ambiguity set X, sequentially selecting a test t _j from a test set t, and dividing the original fault state ambiguity set into two new OR nodes X _jp and X _jf for each test t _j, wherein the two new OR nodes are respectively corresponding to 'test pass' and 'test fail'; dividing the fuzzy fault set according to the test result until all faults are isolated or all tests are used; after test t _j, the information gain obtained by the system is:

wherein p (x) = Σ _xi∈xp(x_i); x= xjp u xjf;

Wherein IG (X; t _j) represents the information increment of the set X after the test t _j, and c _j is the cost of the test t _j; equation (4) is called an information increment heuristic function;

4.2 The search algorithm established according to the information gain heuristic function is as follows:

step 4.2.1) firstly, a new set Z is established, wherein the set only comprises root nodes S, and the S is a completely fuzzy node; creating an empty graph G;

step 4.2.2) repeating the steps listed below until the set Z is empty; then the test sequence in the graph G is used as a diagnosis tree to identify the fault source;

Step 4.2.2.1) extracting OR node q from the set Z, adding node q to the graph G; if q is not a terminal node, then node q is partitioned into pass and fail two sets using each test t _j of the available test sets: x _qjp, x _qif;x_qjp and x _qif are direct subsequent OR nodes of the node q, a new set Y and an empty graph G 'are established, and all direct subsequent OR nodes of the node q are added to the graph G'; if q is a terminal node, continuing to select the next OR node in the set Z; repeating the following steps until the test depth of the graph G' is l, wherein l is a predefined test depth parameter;

Step 4.2.2.1.1) extracting an OR node r from the set Y; if the node r is the target node, continuing to process the next OR node in the set Y; otherwise, calculating the unit cost information increment of each available test on the node according to the step 4.1);

Step 4.2.2.1.2) for the node r, selecting a test with the largest increment of test information per unit cost, selecting a test with the smallest index if a plurality of test values are the same, then adding the test into a graph G ', dividing the node r into pass and non-pass subset fuzzy nodes through the test, and respectively adding the obtained subset nodes into a set Y and the graph G';

step 4.2.2.2) calculating an average test cost for each directly subsequent OR node of node q according to the generated fault diagnosis strategy stored in graph G';

Step 4.2.2.3) calculates the average test cost for each available test t _j∈T_q for node q using the following formula:

Wherein,

p(x_qjf)＝1-p(x_qjp)；

Where h (x _qjp) and h (x _qif) are the average test costs of the diagnostic strategies generated by the test pass and test fail subsets x _qjp and x _qif, respectively;

Step 4.2.2.4) sequencing the tests on the basis of the average test cost, and selecting n optimal tests t ^*∈T_q; if the values of the several tests are the same, then the test with the smallest index is selected to add t ^* to graph G, while the pass and fail subsets generated by the pass and fail results of t ^*, respectively, are appended to set Z and graph G, generating a diagnostic tree.

The invention has the beneficial effects that:

according to the invention, the information gain heuristic function is combined with the diagnosis strategy search algorithm, iterative updating is carried out, the search width and depth are increased, the approximate optimal diagnosis strategy is constructed, the average test cost can be reduced, the method is applicable to a complex system, and the global optimal solution can be ensured;

through result analysis, the influence trend of different search widths and depths on the average test cost can be obtained, and the search process of the test sequence is guided and optimized, so that the search efficiency and the accuracy are improved.

Drawings

The invention will be further described with reference to the accompanying drawings and examples, in which:

FIG. 1 is a flow chart of a method of an embodiment of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

As shown in fig. 1, a fault diagnosis optimization method based on a correlation matrix includes the following steps:

1) According to the related structure, function and test scheme of the system to be diagnosed, a multi-signal flow diagram model is established, wherein the content of the multi-signal flow diagram model comprises a system component, a signal set related to a fault module, a limited set of test points and a priority set of test, and the description of the related relationship between faults possibly occurring in each component of the system and the test is completed;

Establishing a multi-signal flow graph model according to a target system, wherein the multi-signal flow graph model is composed of the following elements:

A finite set of m+1 system faults, x= { X ₀,x₁,……,x_m }, where X ₀ represents a no fault state and xi (1 < i < m) represents one of m possible fault states in the system;

2. Probability of occurrence of failure p= { p ₀,p₁……,p_m };

3. test points tp= { TP ₁,TP₂,……TP_p }, each test point at least contains one test t;

4. Testing t= { t ₀,t₁,……,t_n }, wherein the cost of the test is c= { C ₁,c₂,……,c_n }, determined by the required time, manpower requirements or other economic factors;

5. The set of independent functional signals s= { S ₁,s₂,……,s_k }, used to represent the effects observed in the system, modifies one functional signal SX per system state (x _i), and each test can detect a set of functional signals ST (t _j).

2) After a multi-signal flow diagram model is established, obtaining first-order correlation of faults and tests according to whether connection relations exist between fault nodes and test nodes in the model; judging the connection relation between the fault nodes, recursively obtaining a high-order correlation by the first-order correlation, and marking a fault-test correlation matrix as:

where matrix element d _mn is typically a boolean variable, if a fault x _i can be detected by test t _j, then d _ij is 1, otherwise 0.

Judging whether a redundancy test exists according to the generated correlation matrix, if so, eliminating the redundancy test, and reserving one to optimize the matrix.

And constructing fault diagnosis strategies under different search width and depth combinations by adopting a search algorithm based on information gain for the optimized matrix.

3) Judging whether a redundant test of faults exists or not according to the generated correlation matrix, and eliminating the redundant test if the redundant test exists; wherein the redundancy test is that there are identical indistinguishable columns in the matrix;

4) And constructing fault diagnosis strategies under different search width and depth combinations by adopting a method based on information gain, comparing average test cost under different search widths and depths, and optimally designing the fault diagnosis strategies.

The step 4) is specifically as follows:

4.1 Calculating an average information increment for each test;

The information gain heuristic function is to calculate the average information increment of each test, and use the value as a standard for judging the test quality, select the current fault state ambiguity set X to be isolated, sequentially select the test t _j from the test set t, and for each test t _j, divide the original fault state ambiguity set into two new OR nodes X _jp and X _jf, which correspond to "test pass" and "test fail", respectively; dividing the fuzzy fault set according to the test result until all faults are isolated or all tests are used; after test t _j, the information gain obtained by the system is:

wherein p (x) = Σ _xi∈xp(x_i); x= xjp u xjf;

the value of the information entropy reflects the extent of the failure mode that caused the system failure,

Wherein IG (X; t _j) represents the information increment of the set X after the test t _j, and c _j is the cost of the test t _j; equation (4) is called an information increment heuristic function;

4.2 According to the information gain heuristic function, the search algorithm steps are as follows:

step 4.2.1) firstly, a new set Z is established, wherein the set only comprises root nodes S, and the S is a completely fuzzy node; creating an empty graph G;

step 4.2.2) repeating the steps listed below until the set Z is empty; then the test sequence in the graph G is used as a diagnosis tree to identify the fault source;

Step 4.2.2.1) extracting OR node q from the set Z, adding node q to the graph G; if q is not a terminal node, then node q is partitioned into pass and fail two sets using each test t _j of the available test sets: x _qjp, x _qif;x_qjp and x _qif are direct subsequent OR nodes of the node q, a new set Y and an empty graph G 'are established, and all direct subsequent OR nodes of the node q are added to the graph G'; if q is a terminal node, continuing to select the next OR node in the set Z; repeating the following steps until the test depth of the graph G' is l, wherein l is a predefined test depth parameter;

Step 4.2.2.1.1) extracting an OR node r from the set Y; if the node r is the target node, continuing to process the next OR node in the set Y; otherwise, calculating the unit cost information increment of each available test on the node according to the step 4.1);

Step 4.2.2.1.2) for the node r, selecting a test with the largest increment of test information per unit cost, selecting a test with the smallest index if a plurality of test values are the same, then adding the test into a graph G ', dividing the node r into pass and non-pass subset fuzzy nodes through the test, and respectively adding the obtained subset nodes into a set Y and the graph G';

step 4.2.2.2) calculating an average test cost for each directly subsequent OR node of node q according to the generated fault diagnosis strategy stored in graph G';

Step 4.2.2.3) calculates the average test cost for each available test t _j∈T_q for node q using the following formula:

Wherein,

p(x_qjf)＝1-p(x_qjp)；

Where h (x _qjp) and h (x _qif) are the average test costs of the diagnostic strategies generated by the test pass and test fail subsets x _qjp and x _qif, respectively;

Step 4.2.2.4) sequencing the tests on the basis of the average test cost, and selecting n optimal tests t ^*∈T_q; if the values of the several tests are the same, then the test with the smallest index is selected to add t ^* to graph G, while the pass and fail subsets generated by the pass and fail results of t ^*, respectively, are appended to set Z and graph G, generating a diagnostic tree.

Through the steps, the diagnosis strategies with different search width and depth combined structures are compared, so that lower average test cost can be obtained, the method has guiding significance on the search process of the test sequence, and the method can improve the search efficiency and the search precision of the fault diagnosis test sequence.

It will be understood that modifications and variations will be apparent to those skilled in the art from the foregoing description, and it is intended that all such modifications and variations be included within the scope of the following claims.

Claims (1)

1. The fault diagnosis optimization method based on the correlation matrix is characterized by comprising the following steps of:

1) According to a target system to be diagnosed and a test scheme, a multi-signal flow diagram model is established, and description of correlation between faults possibly occurring in each component of the system and the test is completed; the multiple signal flow graph model includes: the multiple signal flow graph model includes the following elements:

1.1 A finite set of m+1 system faults, x= { X ₀,x₁,……,x_m }, where X ₀ represents a no fault state, X _i, 1.ltoreq.i.ltoreq.m, representing one of m possible fault states in the system;

1.2 Probability of occurrence of failure p= { p ₀,p₁……,p_m };

1.3 Test points tp= { TP ₁,TP₂,……,TP_q }, each test point containing at least one test t _j;

1.4 Test set t= { t ₁,t₂,……,t_n }, where the cost of the corresponding test is c= { C ₁,c₂,……,c_n }; wherein j is more than or equal to 1 and less than or equal to n;

2) Generating a correlation matrix of faults and tests according to the multi-signal flow diagram model; the matrix element d _ij of the correlation matrix is a boolean variable, if a fault x _i can be detected by test t _j, then d _ij is 1, otherwise it is 0;

3) Judging whether a redundant test of faults exists or not according to the generated correlation matrix, and eliminating the redundant test if the redundant test exists; wherein the redundancy test is that there are identical indistinguishable columns in the matrix;

4) Constructing fault diagnosis strategies under different search width and depth combinations by adopting a method based on information gain, comparing average test cost under different search widths and depths, and optimally designing the fault diagnosis strategies;

The step 4) is specifically as follows:

4.1 Calculating an average information increment for each test;

Selecting a current fault state ambiguity set X ', sequentially selecting a test t _j from the test set t, and dividing the fault state ambiguity set X' into two new OR nodes X _jp and X _jf for each test t _j, wherein the two new OR nodes respectively correspond to 'test pass' and 'test fail'; dividing the fault state fuzzy set X' according to the test result until all faults are isolated or all tests are used; after test t _j, the information gain obtained by the system is:

wherein p (x) = Σ _xi∈xp(x_i);x＝x_jp∪x_jf;

Wherein IG (X '; t _j) represents the information increment of the fault state ambiguity set X' after the test t _j, and c _j is the cost of the test t _j; equation (4) is called an information increment heuristic function;

4.2 The search algorithm established according to the information gain heuristic function is as follows:

step 4.2.1) firstly, a new set Z is established, wherein the set only comprises root nodes S, and the S is a completely fuzzy node; creating an empty graph G;

step 4.2.2) repeating steps 4.2.2.1) through 4.2.2.4) until set Z is empty; then the test sequence in the graph G is used as a diagnosis tree to identify the fault source;

Step 4.2.2.1) extracting OR node q from the set Z, adding node q to the graph G; if q is not a terminal node, then node q is partitioned into pass and fail two sets using each test t _j of the available test sets: x _qjp, x _qif;x_qjp and x _qif are direct subsequent OR nodes of the node q, a new set Y and an empty graph G 'are established, and all direct subsequent OR nodes of the node q are added to the graph G'; if q is a terminal node, continuing to select the next OR node in the set Z; repeating steps 4.2.2.1.1) and 4.2.2.1.2) until the test depth of graph G' is l, i being a predefined test depth parameter;

Step 4.2.2.1.1) extracting an OR node r from the set Y; if the node r is the target node, continuing to process the next OR node in the set Y; otherwise, calculating the unit cost information increment of each available test on the node according to the step 4.1);

Step 4.2.2.1.2) for the node r, selecting a test with the largest increment of test information per unit cost, selecting a test with the smallest index if a plurality of test values are the same, then adding the test into a graph G ', dividing the node r into pass and non-pass subset fuzzy nodes through the test, and respectively adding the obtained subset nodes into a set Y and the graph G';

step 4.2.2.2) calculating an average test cost for each directly subsequent OR node of node q according to the generated fault diagnosis strategy stored in graph G';

Step 4.2.2.3) calculates the average test cost for each available test t _j∈T_q for node q using the following formula:

Wherein,

p(x_qjf)＝1-p(x_qjp)；

Where h (x _qjp) and h (x _qif) are the average test costs of the diagnostic strategies generated by the test pass and test fail subsets x _qjp and x _qif, respectively;

Step 4.2.2.4) sequencing the tests on the basis of the average test cost, and selecting n optimal tests t ^*∈T_q; if the values of the several tests are the same, then the test with the smallest index is selected to add t ^* to graph G, while the pass and fail subsets generated by the pass and fail results of t ^*, respectively, are appended to set Z and graph G, generating a diagnostic tree.