A fault diagnosis strategy optimization method considering test uncertainty

Abstract

The application discloses a fault diagnosis strategy optimization method considering test uncertainty, and the specific steps are as follows: a simulation model of a measured object is established based on a Modelica language; the relationship between test points and faults is counted based on the simulation model, and a test-fault uncertainty matrix is established; on the basis of the test-fault uncertainty matrix, an error diagnosis cost caused by test uncertainty is introduced, an information entropy algorithm is improved, and a diagnosis strategy is obtained based on the improved information entropy algorithm. The application firstly proposes to establish a test-fault uncertainty matrix based on a simulation model; then, the error diagnosis cost is introduced into the information entropy algorithm, and the accuracy of the diagnosis strategy is improved. Finally, a backtracking step is added to the information entropy algorithm, and the algorithm is prevented from falling into local optimization. The research on the fault diagnosis strategy optimization based on the test uncertainty has important significance for improving the fault diagnosis capability of the measured object. The diagnosis strategy is fed back to the forward design of the system, and a reference is provided for the design of the measured object.

Description

Technical fields:

[0001] This invention relates to a fault diagnosis strategy optimization method that takes into account test uncertainties, which belongs to the field of system testability analysis and design. Background technology:

[0002] Testability refers to a system's ability to determine its state accurately and promptly and effectively isolate its internal faults. The level of testability directly affects the system's fault diagnosis capability. Traditional testability analysis is based on the assumption of perfect test reliability, with a Boolean relationship between test and fault (either 0 or 1). In reality, tests are often affected by factors including environmental interference, sensor failures, and inherent test errors. Research shows that ignoring the influence of external factors on testing can lead to artificially inflated testability metrics. To improve the testability of a system, a fault diagnosis strategy optimization method that considers test uncertainties is urgently needed.

[0003] Without considering the interference of environmental factors on the test, the relationship between the test and the failure is represented by a deterministic "test-failure" matrix:

[0004]

[0005] In equation (1), the i-th row [b i1 b i2 … b in ] indicates that when the fault mode f i When it occurs, the output results from the 1st test point to the nth test point; the jth column [b 1j b 2j … b mj ] indicates test point t j The output results for the first to the mth fault modes.

[0006] When the test points are completely reliable, the relationship between the test and the fault can be represented by equation (1). However, when the test is affected by factors such as improper test point settings, improper manual operation, environmental interference, and unreasonable threshold settings, the relationship between the test and the fault is no longer a Boolean relationship of 0 or 1. Considering test uncertainty, the following four situations may occur:

[0007] Case 1: Fault f i No occurrence, test point t j No fault was detected;

[0008] Case 2: Fault f i Occurred at test point t j A fault was detected, indicating that test point t j For fault f i Detection probability;

[0009] Case 3: fault f does not occur, test point t i occurs, test point t j does not occur, test point t j false alarm probability of fault f. i false alarm probability of fault f.

[0010] Case 4: fault f occurs, test point t i does not occur, test point t j occurs, test point t j false alarm probability of fault f. i false alarm probability of fault f.

[0011] Therefore, after considering the test uncertainty, the test points may occur false negative and false positive cases. Since the test point is assumed to be completely reliable, the testability index calculated will inevitably have the problem of false high. In addition, including greedy algorithm, information entropy algorithm, etc. are a kind of local optimal algorithm, and the obtained diagnosis strategy is not the optimal solution, and although the global optimal AO* algorithm can obtain the optimal diagnosis strategy, when the number of faults is greater than 50, the algorithm complexity is too high, which leads to too long solving time, and it is difficult to meet the requirement of rapidity of the diagnosis strategy, and even local optimal situation appears. SUMMARY

[0012] In order to overcome the above-mentioned testability analysis, the testability index is false high, the diagnosis strategy is local optimal, and the like, the present application provides a fault diagnosis strategy optimization method considering test uncertainty.

[0013] The technical scheme adopted by the present application is as follows:

[0014] (1). A simulation model of the object to be tested is established based on the Modelica language;

[0015] (2). The relationship between the test points and the faults is counted based on the simulation model, and a test-fault uncertainty matrix is established;

[0016] (3). On the basis of establishing the test-fault uncertainty matrix, the error diagnosis cost brought by the test uncertainty is introduced, the information entropy algorithm is improved, and the diagnosis strategy is obtained based on the improved information entropy algorithm.

[0017] Further, the specific steps of step (1) are as follows:

[0018] a. The basic structure of the object to be tested is clarified, the object to be tested is divided into modules according to the independence principle and the physical division principle, and a component-level model is established;

[0019] b. The mathematical equation set representing the physical characteristics of the component is written by the Modelica language, and the simulation model of the component is established.

[0020] c. Verify the accuracy of the simulation model, compare the eigenvalues, eigenvalue trends, and distribution rules of the key operating parameters obtained by simulation with the measured data, and verify the accuracy of the model.

[0021] Further, the specific steps of step (2) are as follows:

[0022] a. Collect the measured data of each test point, and obtain the specific distribution function of each test point by maximum likelihood estimation;

[0023] b. According to the different environmental interference objects, the interference to the test points and the interference to the measured object itself are divided, and through FMEA analysis, the corresponding relationship between each fault and the component parameter is determined;

[0024] c. Through multiple simulations, the relationship between the test points and the faults is obtained, and the relationship between the test and the fault is composed of fault detection rate and fault false alarm rate;

[0025] d. Establish a "test-fault" uncertainty matrix.

[0026] Further, the construction method of the "test-fault" uncertainty matrix is as follows:

[0027] First, according to the composition, function and historical fault information of the measured object, the fault set F = {f1, f2, … f m} and the corresponding occurrence probability P = {p1, p2, … p m}, the test point set T = {t1, t2, … t n} are determined;

[0028] Secondly, through the fault mode and effect analysis FMEA, the representation of each fault is determined, the mapping relationship between each fault and the component parameter in the simulation model is obtained, and the corresponding component parameter is modified to realize virtual fault injection;

[0029] Then, according to the measured data, the distribution function of each test point is estimated, and according to different interference objects, interference quantities are loaded at different places of the model to simulate the environmental interference existing in the actual test work;

[0030] Finally, on the basis of injecting faults f i , the simulation model is run multiple times, the detection rate of test point t j to fault f i is counted, and under the normal state of the measured object, the simulation model is run multiple times, the probability of false alarm of test point t j itself is counted, and by obtaining the fault detection rate and the false alarm rate, the test-fault uncertainty matrix is established.

[0031] Further, the calculation step of the improved information entropy algorithm in step (3) is as follows:

[0032] step1: initialize the "test-failure" uncertainty matrix D * , wherein the total number of failures is m, and the total number of test points is n;

[0033] step2: based on the information entropy algorithm, a temporary diagnosis tree of each test point t j is obtained by considering the error diagnosis cost;

[0034] step3: the average information amount of each temporary diagnosis tree is calculated, and the test point with the maximum average information amount is selected as the optimal test point, and the average information amount is represented as:

[0035]

[0036] In formula (3), step(j) represents the number of test steps generated in the diagnosis tree of the test point t j ; Cost j represents the error diagnosis cost brought by the test point t j ; and I(X;t j ) represents the information amount value after the test point t j is executed on the "test-failure" matrix (X).

[0037] The present application has the following beneficial effects: compared with the existing testability diagnosis strategy optimization method, the present application firstly proposes to establish a "test-failure" uncertainty matrix based on a simulation model; then, the error diagnosis cost is introduced into the information entropy algorithm to improve the accuracy of the diagnosis strategy; finally, a backtracking step is added to the information entropy algorithm to avoid the algorithm from falling into local optimization. The research on the fault diagnosis strategy optimization based on test uncertainty has important significance for improving the fault diagnosis capability of the measured object. By feeding back the diagnosis strategy to the forward design of the system, a reference is provided for the design of the measured object. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is a whole flowchart.

[0039] Figure 2 is a flowchart for establishing a "test-failure" uncertainty matrix.

[0040] Figure 3 is a flowchart of the diagnosis strategy optimization based on the improved information entropy algorithm.

[0041] Figure 4 is the observation value of T1 point.

[0042] Figure 5 is the diagnosis strategy based on the improved information entropy algorithm.

[0043] Figure 6 This is a diagnostic strategy based on the information entropy algorithm. Detailed implementation method:

[0044] The invention will now be further described with reference to the accompanying drawings.

[0045] This invention considers an optimization method for fault diagnosis strategies that address testing uncertainties. The specific steps are as follows:

[0046] (1) Establish a simulation model of the object under test based on the Modelica language;

[0047] (2). Based on the simulation model, statistically analyze the relationship between test points and faults to establish a test-fault uncertainty matrix;

[0048] (3). Based on the establishment of the test-fault uncertainty matrix, the error diagnosis cost caused by test uncertainty is introduced, the information entropy algorithm is improved, and the diagnosis strategy is obtained based on the improved information entropy algorithm.

[0049] The specific steps of step (1) are as follows:

[0050] a. Define the basic structure of the object under test. Analyze the structure and functional implementation process of the object under test. Based on the testability requirements, divide the object under test into specified levels and model it using a top-down approach. Divide the object under test into modules according to the principles of independence and physical partitioning, and establish a component-level model.

[0051] b. Establish component models. Using the Modelica language, write a system of mathematical equations representing the physical characteristics of the components to establish a simulation model. Simultaneously, define the components as abstract types to improve model reusability through inheritance, derivation, and explicit declarations.

[0052] c. Verify the accuracy of the simulation model. Compare the characteristic values, characteristic change trends, and distribution patterns of the key operating parameters obtained from the simulation with the measured data to verify the correctness of the model.

[0053] like Figure 2 As shown, the specific steps of step (2) are as follows:

[0054] a. Estimation of the distribution function of test points. Test points are susceptible to interference from various factors, including improper test point setup, incorrect manual operation, environmental disturbances, and unreasonable threshold settings. This invention primarily studies the environmental interference affecting testing. By collecting test data from the tested object, the specific distribution function of each test point is obtained through maximum likelihood estimation. When actual test data is lacking, test data from similar systems can be referenced.

[0055] b、According to the different environmental interference objects, interference to the test points and interference to the measured object itself are divided. On the basis of FMEA analysis, virtual fault injection is realized based on the simulation model. In order to accurately simulate the occurrence of a specified fault, through FMEA analysis, the correspondence between each fault and the parameters of the component is determined. On the basis of determining the mapping relationship between the fault and the parameters, virtual fault injection can be realized by modifying the corresponding component parameters. Random interference quantities conforming to the distribution function of each test point are generated, and according to different interference objects, interference quantities are loaded at different places of the model.

[0056] c、Through multiple simulations, the relationship between the test points and the faults is obtained. The relationship between the test and the fault is composed of fault detection rate and fault false alarm rate. When injecting a fault f i , the simulation times are set to N, and the number of times that the test point t j detects the fault f i occurs is n1, then the detection rate of the test point t j to the fault f i is n1 / N. Similarly, when the simulation model is in a normal state, the simulation times are set to N, and the number of times that the test point t j detects any fault f i occurs is n2, then the probability of false alarm of the test point is n2 / N.

[0057] d、Establish a "test-fault" uncertainty matrix. In the "test-fault" certainty matrix, when the test point t j is logically related to the fault f i , b ij =1; in the "test-fault" uncertainty matrix, the element Pd ij in the ith row and jth column is n1 / N; when the test point is not logically related to the fault f i , b ij =0, and in the "test-fault" uncertainty matrix, the element pf j in the jth column is n2 / N. Therefore, the element d ij of the "test-fault" uncertainty matrix is represented as:

[0058] d ij =P(t jf |f i )=b ij ·Pd ij +(1-b ij )·pf j (2)

[0059] In formula (2), b ij represents the element in the ith row and jth column of the "test-fault" matrix.

[0060] When bij = 0, equation (2) represents the false alarm rate of test point t j ; when b ij = 1, equation (2) represents the detection rate of test point t j for fault f i .

[0061] As shown in Figure 3, the specific steps of step (3) are as follows: Figure 3

[0062] a. Define the error diagnosis cost under uncertain test conditions. Since the test point may miss detection and false alarm, etc., two kinds of error diagnosis conclusions will occur:

[0063] Diagnosis conclusion 1: Assuming that the fault occurred in the system is f i , since the missed detection of test point t j is not considered, test point t j can detect the fault. In the actual test process, due to external environmental interference, test point t j may miss detection, and no fault f i occurs, resulting in an error diagnosis conclusion f i ∈ X jp (X jp represents the fault set that test point t j cannot detect). Since the subsequent test cannot accurately isolate fault f i , the final test conclusion is no fault or other faults occur and miss detection of fault f i .

[0064] Diagnosis conclusion 2: Assuming that the fault occurred in the system is f k , since the false alarm of test point t j is not considered, test point t j will not detect the occurrence of fault f k . In the actual test process, due to external environmental interference, test point t j detects the occurrence of fault f k , and the diagnosis conclusion obtained is f k ∈ X jf (X jf represents the fault set that test point t j can detect). Since the subsequent test cannot accurately isolate fault f k , it finally leads to an error diagnosis conclusion.

[0065] ​In diagnosis conclusion 1, it will cause that the components which should be repaired are not repaired, and its loss can be described as missed diagnosis cost; in diagnosis conclusion 2, it will cause that other components which should not be repaired are replaced or repaired, and its loss can be described as misdiagnosis cost.

[0066] b. By adding a backtracking step, the information entropy algorithm is improved

[0067] Since the information entropy algorithm is a local optimization algorithm, in order to obtain the optimal diagnosis strategy, the basic idea of improving the information entropy algorithm is:

[0068] A single test point is determined by two calculations. First, the information entropy of all test points in the test point set is calculated on the initial "test-fault" matrix to determine the temporary diagnosis tree of each test point. Second, the average information amount brought by each temporary diagnosis tree is calculated, and the test point where the diagnosis tree with the largest average information amount is selected as the optimal test point. The calculation steps of the improved information entropy algorithm are as follows:

[0069] step1: initialize the "test-fault" uncertainty matrix D * , where the total number of faults is m and the total number of test points is n;

[0070] step2: based on the information entropy algorithm, the temporary diagnosis tree of each test point t j is obtained by considering the cost of misdiagnosis;

[0071] step3: calculate the average information amount of each temporary diagnosis tree, and select the test point where the largest average information amount is located as the optimal test point. The purpose of selecting the largest average information amount instead of the largest information amount is to avoid the algorithm falling into deep search. The average information amount is based on the total information amount of the diagnosis tree and considers the diagnosis steps, which avoids the algorithm excessively pursuing information amount and causing more steps to isolate each fault. The average information amount is represented as:

[0072]

[0073] In formula (3), step(j) represents the number of test steps generated in the diagnosis tree of the test point t j ; Cost j represents the misdiagnosis cost of executing the test point t j ; I(X; t j ) represents the information amount value after executing the test point t j on the "test-fault" matrix (X).

[0074] Step4: repeat step1~step3 until the test-fault uncertainty matrix cannot be divided.

[0075] The present application takes a complex system as an example to optimize the diagnosis strategy, and the specific steps are as follows:

[0076] Based on step (1), based on the Modelica language, a simulation model is established based on the function structure of the complex system.

[0077] Based on step (2), the complex system has 5 test points and 7 faults, and the "test-fault" matrix is shown in Table 1:

[0078] Table 1 "Test-fault" deterministic matrix

[0079]

[0080]

[0081] Taking test point T1 and fault F1 as an example, based on loading random disturbance in the model, the multiple simulation results are shown in Figure 5 . Figure 5 In the figure, the red dot represents that the test point does not detect the occurrence of the fault. When fault F1 occurs, test point T1 does not detect the occurrence of the fault twice. Therefore, the detection rate of test point T1 to fault F1 is 0.98. Table 2 shows the established "test-fault" uncertainty matrix.

[0082] Table 2 "Test-fault" uncertainty matrix

[0083]

[0084] Based on step (3), the diagnosis tree obtained by considering the error diagnosis cost in the information entropy algorithm and adding the backtracking step is shown in Figure 6 .

[0085] Figure 6 Indicates the diagnosis strategy obtained based on the information entropy algorithm without considering the test uncertainty.

[0086] Table 3 gives the testability index values of different diagnosis strategies. According to Table 3, the diagnosis strategy generated based on the improved information entropy algorithm has higher accuracy.

[0087] Table 3 Comparison of diagnosis strategies

[0088] Diagnostic strategy Fault detection rate Fault isolation rate False alarm rate Diagnostic strategy of information entropy algorithm 99.4852％ 89.8863％ 10.1137％ Improved diagnostic strategy of information entropy algorithm 99.5048％ 98.6522％ 1.3478％

[0089] The above only describes the preferred embodiments of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements can be made, and these improvements should also be considered as the protection scope of the present application.

Claims

1. A fault diagnosis strategy optimization method considering test uncertainty, characterized in that: The specific steps are as follows: (1). Based on the Modelica language, a simulation model of the object under test is established; (2). Based on the simulation model, the relationship between test points and faults is counted, and a test-fault uncertainty matrix is established; (3). On the basis of establishing the test-fault uncertainty matrix, the error diagnosis cost caused by test uncertainty is introduced, the information entropy algorithm is improved, and the diagnosis strategy is obtained based on the improved information entropy algorithm; The specific steps of step (1) are as follows: a. Clearly define the basic structure of the object under test, and perform module division on the object under test according to the independence principle and physical division principle to establish a component-level model; b. Write mathematical equation sets representing the physical properties of components through the Modelica language to establish a simulation model of the components; c. Verify the accuracy of the simulation model, compare and analyze the similarity of the characteristic values, characteristic change trends and distribution rules of the key operating parameters obtained by simulation with the measured data, and verify the accuracy of the model; The calculation steps of the improved information entropy algorithm in step (3) are as follows: Step 1: initialize the "test-fault" uncertainty matrix D * where the total number of faults is m and the total number of test points is n; step2: Based on information entropy algorithm, the temporary diagnosis tree of each test point t is obtained on the basis of considering the cost of error diagnosis. j ; Step 3: Calculate the average information amount of each temporary diagnosis tree, select the test point with the maximum average information amount as the optimal test point, and the average information amount is represented as: In formula (3), step (j) represents performing test point t j The number of test steps in the generated diagnosis tree; Cost j represents performing test point t j The error diagnosis cost brought by; I(X; t j represents the information value after performing test point t j on the "test-fault" matrix X.

2. The method for optimizing a fault diagnostic strategy taking into account test uncertainty according to claim 1, characterized in that: The specific steps of step (2) are as follows: a. Collect the measured data of each test point, and obtain the specific distribution function of each test point through maximum likelihood estimation; b. According to the different environmental interference objects, the interference on the test points and the interference on the object under test are divided, and through FMEA analysis, the corresponding relationship between each fault and the component parameter is clarified; c. Through multiple simulations, the relationship between test points and faults is obtained, and the relationship between test and fault is composed of fault detection rate and fault false alarm rate; d. Establish a "test-fault" uncertainty matrix.

3. The method for optimizing a fault diagnostic strategy taking into account test uncertainty according to claim 2, characterized in that: The construction method of establishing the "test-fault" uncertainty matrix is as follows: First, based on the composition, function, and historical fault information of the object under test, determine its fault set F = {f1, f2, ... f}. m } and the corresponding probability of occurrence P = {p1, p2, ... p m The test point set T = {t1, t2, ..., t} n }; Secondly, through fault mode and effect analysis FMEA, the characteristics of each fault are clarified, the mapping relationship between each fault and the component parameter in the simulation model is obtained, and the corresponding component parameter is modified to realize virtual fault injection; Then, according to the measured data, the distribution function of each test point is estimated, and according to different interference objects, interference quantities are loaded at different places of the model to simulate the environmental interference existing in actual test work. Finally, in the injection of fault f i Based on this, the simulation model was run multiple times, and the test point t was statistically analyzed. j For fault f i The detection rate was determined by running the simulation model multiple times under normal conditions and statistically analyzing the test points t. j The probability of a false alarm occurring is determined by obtaining the fault detection rate and the false alarm rate, and a test-fault uncertainty matrix is ​​established.

Images