Airborne system PHM testability modeling and diagnostic strategy optimization method and device

Abstract

The present invention provides a method and device for optimizing PHM testability modeling and diagnostic strategy for airborne systems, the method comprising: using Latin hypercube sampling to construct simulation parameter spaces for different working states of the system, extracting data feature values ​​as test points through data mining, establishing evaluation criteria based on ROC curves to evaluate the detection capabilities of test points for different faults, constructing a 4-value correlation matrix with uncertain states, and then calculating a heuristic evaluation function with penalty terms to generate a test sequence and optimize the diagnostic strategy. The present invention solves the problem of difficult testability design caused by unreliable tests in complex systems, and proposes a solution to the problem of optimizing diagnostic strategies for uncertain multi-valued attribute systems, which can effectively improve the accuracy of fault diagnosis and the efficiency of fault isolation.

Description

Technical Field

[0001] The invention relates to a method and a device for optimizing PHM testability modeling and diagnostic strategy for an airborne system. Background Art

[0002] By carrying out testability modeling, analysis and evaluation of airborne systems, we can further clarify the optimal number of sensors, optimal sensor layout, most reasonable test sequence and most appropriate monitoring scheme required for equipment fault prediction and health management (PHM), which is very helpful for aircraft system and equipment status monitoring, fault diagnosis, life prediction and health management. The testability level of the system, also known as detectability, reflects the ability of the equipment to detect its working status in a timely and accurate manner. Therefore, design of testability (DFT) is crucial in the product design process and plays an important role in reducing the subsequent maintenance cost of equipment and improving product safety. By optimizing the testability modeling method, testability design can be carried out more efficiently and the product fault detection capability and reliability can be improved.

[0003] Existing testability methods include: multi-signal flow graph, information flow model and hybrid diagnosis model. The above methods can represent the correlation between faults and tests. The relationship between the working performance of the components in the system and the test results is described by the Boolean value between the fault and the test in the traditional correlation matrix (D-matrix). If a test can detect the fault of the component, the corresponding element in the matrix is ​​"1", if not, it is "0". However, for complex systems usually composed of mechanics, electronics and hydraulics, due to the mutual coupling between the modules and units in the system, the relationship between faults and the relationship between faults and test nodes are complex. The binary test matrix represented by 0 / 1 often cannot accurately describe the relationship between faults and faults and test nodes, and it is easy to have insufficient fault information or fuzzy fault groups. Therefore, in order to accurately describe the relationship between faults and tests and the degree of association, many scholars have used multi-valued attribute systems (MVAS) to describe complex systems and carried out research on corresponding fault diagnosis strategies. If there are more than two test results in a system, the system is defined as a multi-valued attribute system.

[0004] Current testability models are mainly based on deterministic assumptions, that is, if a sensor is associated with a fault, when a fault occurs, the sensor will detect the fault with a 100% probability. However, in complex systems, there are interferences caused by uncertain factors such as production assembly, use wear, and working environment. The probability of fault detection is closely related to the uncertainty of the system, which means that the test may have missed detections and false alarms. However, there are few studies on test sequence generation for MVAS with unreliable tests.

[0005] In summary, the existing technology lacks a testability modeling and diagnostic strategy optimization method for PHM of airborne systems. Aiming at the uncertainty of test results caused by uncertain factors generated in the actual operation of complex systems, a testability model of multi-valued attribute systems is established, test sequences are generated and diagnostic strategies are optimized. Summary of the invention

[0006] Purpose of the invention: The technical problem to be solved by the present invention is to provide a method for optimizing the PHM testability modeling and diagnostic strategy of an airborne system in view of the deficiencies in the prior art. Data feature values ​​are extracted as test points through data mining, the relationship between the test points and faults is quantitatively evaluated based on the ROC curve, a correlation matrix containing uncertain states is constructed, and a heuristic evaluation function integrating information gain is used to search for test points, generate test sequences, and optimize the fault diagnosis strategy.

[0007] The method of the present invention specifically comprises the following steps:

[0008] Step 1. Determine the typical failure modes of complex systems, mainly analyzing the types of failures with higher failure rates. Define the probability distribution of uncertainty parameters, mainly considering the parameters that have a greater impact on the system state, including: environmental parameters, which characterize the working environment of the system; empirical parameters, which are obtained through engineering practice summaries or recommended by specifications and standards; fault parameters, which are parameters that cause system failures. Among them, empirical parameters and fault parameters are both structural parameters. For fault parameters, additional fault thresholds need to be set. When the parameter value reaches the fault threshold, the system fails in the corresponding mode. Step 1 is mainly based on expert experience and previous research foundations;

[0009] Step 2: Use Latin hypercube sampling to construct parameter spaces in healthy states and different fault states, perform experimental design (DOE), and obtain the experimental design DOE matrix. From the experimental design DOE matrix, extract one experimental sample at a time to perform simulation experiments, and obtain test data under the corresponding state of the system;

[0010] Step 3: Perform data mining on the system test data obtained in step 2, extract effective eigenvalues ​​to replace the traditional simple test value analysis, and use the eigenvalues ​​as test points to monitor the system status.

[0011] Step 4: Based on the receiver operating characteristic curve (ROC), evaluate the detection capability of each test point for different system states, quantify the potential of the test point for system fault diagnosis, and construct an uncertainty 4-value correlation matrix (D matrix), which contains 4 states: 1, -1, 0, U, where U is called the uncertain state;

[0012] Step 5: Determine whether the uncertainty 4-value D matrix is ​​completely divided. If the division is complete, the matrix will not be divided any more and go to step 7. If the division is not completed, go to step 6. The criteria for judging whether the matrix is ​​completely divided include: there is only one row left in the matrix, or the matrix is ​​all uninformative columns. An uninformative column is defined as: the column has only one state except the element "U".

[0013] Step 6: Traverse each test point, construct a heuristic evaluation function with a penalty term based on information gain, select the one with the largest evaluation function value as the optimal diagnostic test point, divide the uncertainty 4-value D matrix into different sub-matrices according to the optimal test point, and repeat step 5 for each sub-matrix until the uncertainty 4-value D matrix is ​​divided;

[0014] Step 7: Combine all the optimal diagnostic test points to obtain a test sequence, and determine the fault fuzzy group and the isolated fault.

[0015] Furthermore, step 2 includes: all possible values ​​of the p uncertain parameters form a p-dimensional vector, each component of the p-dimensional vector is divided into w parts according to the cumulative probability, w is the number of required sampling points; randomly sample from the w small intervals in each dimension to form an initial matrix M of w×p:

[0016]

[0017] Among them, R wp Represents the element in the wth row and pth column of the initial matrix M;

[0018] The elements of each column in the initial matrix M are randomly arranged to simulate the random combination of uncertain parameters, as shown below:

[0019]

[0020] The elements The subscript X 11 ,X 21 ,…,X w1It represents the random arrangement of the column vector elements of the initial matrix M, and finally obtains the Latin hypercube sampling matrix, where each row vector is a p-dimensional test sample. According to the obtained DOE matrix, a p-dimensional test sample is taken each time, and simulation experiments are carried out through the model to simulate different working conditions of the system in the actual working environment, and obtain test data under the corresponding state of the system respectively.

[0021] Furthermore, in step 2, the Latin hypercube sampling process needs to be run multiple times when conducting the experimental design. When constructing the healthy state parameter space, the distributions of all uncertain parameters need to be sampled, but when constructing the fault state parameter space, the corresponding fault parameters are adjusted to the fault thresholds, and the remaining parameter distributions are sampled.

[0022] Step 4 includes: uncertainty 4 value D matrix D m×n As shown below:

[0023]

[0024] In the uncertainty 4-value D matrix, one row represents a state of the system, and each column represents a test point. The matrix element D(f i ,t j ) = d ij represents the jth test point t j The detection capability of the i-th fault mode. According to the uncertainty propagation theory, due to the existence of uncertainty parameters, even in the same working state or fault mode, the parameters output by the test point will be random, which means that the test values ​​extracted based on the output parameters will appear in the form of distribution rather than specific values. Therefore, based on the receiver operating characteristic (ROC) curve, the detection capability of the test point for faults (here D(f i ,t j ) and d ij The ROC curve is a two-dimensional description of the performance of the classifier. Therefore, in order to compare the performance of the classifier, the area under the curve (AUC) is an important indicator as a single scalar value. The larger the AUC, the better the average performance of the classifier. When the AUC is less than 0.7, the classifier can be considered ineffective.

[0025] Given a test point t j and failure mode f i Receiver operating characteristic (ROC) curve The corresponding area under the curve (AUC) is recorded as According to the distribution of test values ​​in healthy state and distribution under fault conditions Draw the receiver operating characteristic ROC curve. For each pair of test points and corresponding failure modes, d ij The evaluation criteria are as follows:

[0026] if Above the calibration point and the test value is in the failure mode f i The state is larger than the healthy state, d ij =1;

[0027] if Above the calibration point and the test value is in the failure mode f i In the state, it is smaller than in the healthy state, d ij =-1; the ROC curve is above the calibration point, which means that the two distributions are very different, so as long as the means of the two distributions are compared, it can be determined whether they are too large or too small;

[0028] if d ij =0;

[0029] if Below the calibration point, and d ij =U.

[0030] The coordinates of the calibration point are (0.05, 0.9), which means that when a single test point is used to detect faults, the true positive rate (TPR) needs to be greater than 0.9 and the false positive rate (FPR) needs to be less than 0.05 to meet the detection requirements. The calibration point coordinates can control the accuracy and false alarm rate of fault diagnosis. Selecting the calibration point coordinates (0.05, 0.9) can detect all faults with as few diagnostic steps as possible while ensuring that the diagnostic accuracy and false alarm rate meet the requirements. In addition, the area under the curve AUC can be used to determine the distribution of test values ​​in a healthy state. and distribution under fault conditions Is the difference significant enough?

[0031] d ij = ±1 means the test point can detect the fault mode; d ij =0 represents and There was no significant difference between ij =U stands for and There are obvious differences between them, but the detection requirements are not met; therefore, U may even interfere with the fault isolation process, which is referred to as an "uncertain state" in the present invention.

[0032] Step 6 includes:

[0033] According to the typical failure mode, the original fault fuzzy set of the system is determined as S = {f0, f1, f2, ..., f m}, f0 represents the health state, f m Indicates the mth fault state, calculate the test point t j When the evaluation function is used, the uncertain state U will not produce effective gain for the diagnosis process, so the corresponding fault mode will be removed from the original fault fuzzy set of the system to form a new set, which is recorded as the identifiable fault fuzzy set Computing identifiable fault fuzzy sets Information gain

[0034]

[0035] in Indicates d ij = the sum of the occurrence probabilities of the failure modes of a, where a represents the three types of detection capabilities of the test point except the uncertain state 'U', i.e. a = 1, 0, -1; Represents the identifiable fault fuzzy set The sum of the occurrence probabilities of all failure modes in ;

[0036] To make the calculation meaningful, define but

[0037] The uncertain state U will interfere with the diagnosis process, so when selecting the test point, a penalty item will be calculated based on the number of U. The penalty item Defined as:

[0038]

[0039] Where N U represents the number of failure modes corresponding to the uncertain state in the original fault fuzzy set S, N S Indicates the number of failure modes;

[0040] Final test point t j The heuristic evaluation function h(S; t j ) is defined as follows:

[0041]

[0042] Optimal test point t b Select by maximizing the heuristic function value:

[0043]

[0044] Furthermore, step 6 also includes: dividing the D matrix into three sub-matrices according to the three detection capabilities {-1, 0, 1} of the optimal test point for different faults The corresponding three fault fuzzy subsets are: q represents the number of times the matrix is ​​divided. Due to the uncertainty of state 'U', the fault modes corresponding to state U are not classified separately, but are divided into three other matrices. This is because the currently selected optimal test point cannot effectively classify such fault modes, so their classification results are not concerned in this diagnosis process. Assuming that they will be classified into any fault fuzzy subset, better test points will be found in subsequent steps to diagnose them.

[0045] Step 7 includes: if the uncertainty 4-value D matrix has more than one row, the corresponding fault is judged as a fault fuzzy group; if the uncertainty 4-value D matrix has only one row and the corresponding fault does not appear in other fault fuzzy groups, the fault mode is judged as an isolated fault.

[0046] The present invention also provides airborne system PHM testability modeling and diagnostic strategy optimization, including:

[0047] The parameter determination module is used to determine the typical failure mode of the complex system, define the probability distribution of the uncertainty parameters, determine the parameters, including: environmental parameters, experience parameters, and failure parameters, where both experience parameters and failure parameters are structural parameters, set the failure threshold of the failure parameter, and when the failure parameter value reaches the failure threshold, the system will fail in the corresponding mode;

[0048] The experimental design module is used to construct the parameter space under the healthy state and different fault states respectively through Latin hypercube sampling, conduct experimental design, obtain the experimental design DOE matrix, and sequentially extract an experimental sample from the experimental design DOE matrix to conduct simulation experiments and obtain the test data under the corresponding state of the system;

[0049] A data mining module is used to perform data mining on the test data obtained in step 2, extract effective characteristic values ​​to replace the test value analysis, and use the characteristic values ​​as test points to monitor the system status;

[0050] The fault diagnosis and optimization module is used to evaluate the detection capability of each test point for different system states based on the receiver operating characteristic curve, quantify the potential of the test point for system fault diagnosis, and construct an uncertainty 4-value correlation matrix (D matrix), which contains 4 states: 1, -1, 0, U, where U is called the uncertain state; determine whether the uncertainty 4-value D matrix is ​​completely divided. If the division is completed, combine all the optimal diagnostic test points to obtain the test sequence, and determine the fault fuzzy group and the isolated fault; if the division is not completed, traverse each test point, construct a heuristic evaluation function with a penalty term based on information gain, select the one with the largest evaluation function value as the optimal diagnostic test point, divide the uncertainty 4-value D matrix into different sub-matrices according to the optimal test point, and for each sub-matrix, determine whether the uncertainty 4-value D matrix is ​​completely divided until the uncertainty 4-value D matrix is ​​completely divided; the criteria for judging whether the matrix division is completed include: only one row is left in the matrix, or all the columns in the matrix are non-information columns.

[0051] The present invention also provides a storage medium storing a computer program or instruction. When the computer program or instruction is executed, the airborne system PHM testability modeling and diagnostic strategy optimization method is implemented.

[0052] The beneficial effect of the present invention is that a testability modeling and fault diagnosis strategy optimization method for PHM of aircraft airborne systems and equipment is developed. Taking into account various uncertain factors in manufacturing and operation, the simulation parameter space of the system under different operating states is constructed by Latin hypercube sampling. Extracting effective eigenvalues ​​instead of traditional simple test value analysis, using eigenvalues ​​to monitor the system state for test points, solves the problem of insufficient description of fault information caused by the limitation of the number of sensors. In view of the randomness of test values ​​caused by uncertainty propagation, a new 4-value attribute D matrix is ​​constructed to describe the relationship between test points and faults. The receiver operating characteristic (ROC) curve is introduced to evaluate each element of the D matrix, and an evaluation standard is established according to the calibration point and the area under the curve AUC, which can effectively evaluate the diagnostic ability of the test point for different faults. In order to optimize the diagnostic strategy of a multi-value attribute system (MVAS) with uncertain states, due to the interference caused by the uncertain state to the fault diagnosis process, a heuristic evaluation function with a penalty term based on information gain is proposed to search for the optimal test point, and a new submatrix partitioning method is proposed for the uncertain state, and finally the test sequence with the highest diagnostic efficiency is obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0054] Figure 1 It is a flow chart of the present invention.

[0055] Figure 2 Schematic diagram of system characteristic values ​​in an embodiment of the present invention.

[0056] Figure 3 A schematic diagram of ROC curves of three different test results at the 16th test point in an embodiment of the present invention.

[0057] Figure 4 Schematic diagram of distribution of three different test values ​​of the 16th test point in an embodiment of the present invention.

[0058] Figure 5 A schematic diagram of a fault diagnosis tree in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The process flow chart of the present invention is as follows Figure 1 As shown, including:

[0060] S1. Determine the typical failure modes of complex systems, mainly analyzing the types of failures with high failure rates. For example, in the fuel metering unit (FMU), there are 6 common failure modes, including: demagnetization of the servo valve permanent magnet, degradation of the servo valve feedback spring stiffness, internal leakage of the servo valve slide valve, internal leakage of the metering valve, blockage of the fixed throttle hole of the servo valve, and stagnation of the servo valve slide valve. Consider the 9 uncertain parameters that have a greater impact on the system state, and define the probability distribution of the uncertainty parameters. Environmental parameters: oil temperature, oil pressure; empirical parameters: actuator cylinder diameter, actuator rod diameter; fault parameters: residual magnetic flux, feedback spring stiffness, slide valve internal leakage diameter, metering valve internal leakage diameter, fixed throttle hole diameter, slide valve static friction. According to expert experience and previous studies, given the distribution of uncertain parameters and fault parameter fault thresholds, the uncertain parameters and corresponding values ​​of the FMU system are shown in Table 1.

[0061] Table 1

[0062]

[0063] S2. Accurately construct the parameter space through Latin hypercube sampling and perform the Design of Experiment (DOE). The Latin hypercube sampling process needs to be run multiple times when performing the experimental design. When constructing the healthy state parameter space, the distribution of all uncertain parameters needs to be sampled, but when constructing the parameter space of the fault state, the corresponding fault parameters are adjusted to the fault threshold, and the remaining parameter distributions are sampled. Therefore, a total of 7 samplings were performed in the example, and the DOE matrices under 7 states of the system were obtained respectively. Latin hypercube sampling includes the following steps:

[0064] All possible values ​​of p uncertain parameters form a p-dimensional vector. Each component of the p-dimensional vector is divided into w parts according to the cumulative probability, where w is the number of required sampling points and p is the number of uncertain parameters. Random sampling is performed from the w small intervals in each dimension to form the w×p initial matrix M:

[0065]

[0066] The elements of each column in the initial matrix are randomly arranged to simulate random combinations of uncertain parameters, as follows:

[0067]

[0068] The subscript X of the element R 11 ,X 21 ,…,X w1 Represents the random arrangement of the column vector elements of the M matrix, and finally obtains the Latin hypercube sampling matrix, where each row vector is a p-dimensional test sample. According to the obtained DOE matrix, a p-dimensional test sample is taken each time to perform uncertainty simulation to simulate different working conditions of the system in the actual working environment.

[0069] S3. In order to solve the problem of insufficient fault information caused by the limited number of sensors, inspired by the servo valve flow gain curve, this embodiment combines the displacement signal of the linear variable differential transformer (LVDT) with the servo valve control current to obtain a speed gain curve to reflect the overall performance of the FMU system. This curve only needs to measure the control current I control and spool speed V spool You can get. spool The valve core position X measured by the LVDT sensor spool The speed gain curve of the whole loop is obtained by differential calculation, with the control current as the independent variable and the valve core movement speed of the metering valve as the dependent variable. Based on the speed gain curve, the FMU fault characteristics are extracted as test points. Some key features are Figure 2 Displayed in.

[0070] S4. Construct a 4-value D matrix based on a large amount of simulation data. The elements in the matrix represent the fault detection capability of the corresponding test point. 16 The evaluation criteria are explained using the fault set as an example. Figure 3 As shown, the horizontal axis represents the false positive rate and the vertical axis represents the true positive rate. Respectively represent t 16 ROC curves for detecting three failure modes f1, f2, and f5. According to the evaluation criteria, Below the calibration point and So t 16 The detection index d for f1 1,16 =U. So t 16 The detection index d for f2 2,16 =0; It is above the calibration point and the characteristic value is smaller in the fault mode f5 state than in the healthy state, so t 16 Detection index d for f5 5,16 =-1.

[0071] Construct the correlation matrix of the test point to fault fuzzy set. The correlation matrix of the complete FMU is shown in Table 2. i ) indicates the failure mode f i probability of occurrence. Figure 4 Shows t in healthy state and in three different failure modes 16 The distribution histogram of the characteristic value. The X-axis represents the fault characteristic value, and the Y-axis represents the frequency of the characteristic value in each interval. It can be seen from Table 2 that d 1,16 =U, indicating that the difference between the characteristic values ​​of the fault mode f1 and the healthy state cannot meet the detection requirements. Figure 4 In the figure, the blue area and the yellow area represent the characteristic value distribution under the healthy state and the failure mode f1, respectively. Because of the existence of the overlapping part, only the t 16 It is difficult to detect faults effectively. The blue area and the green area almost overlap, which means that the feature t 16 There is no significant difference between the normal state and the failure mode f2, so d 2,16 = 0. Since the characteristic value of the failure mode f5 is significantly smaller than the characteristic value of the healthy state, d 5,16 = -1, indicating that only based on t 16 Fault f5 can be effectively detected (the color cannot be seen because the attached figure in the manual can only be a grayscale image, so this is specially explained).

[0072] Table 2

[0073]

[0074]

[0075] S5, judging whether the D matrix is ​​completely divided, judging whether the matrix is ​​divided, if the division is completed, the matrix will not be divided any more and jump to S6; if the division is not completed, execute S5. The judgment criteria for whether the matrix is ​​divided include: only one row is left in the matrix, or all the columns in the matrix are non-information columns.

[0076] S6. Find the optimal test point. By calculating the evaluation function value of each test point in the matrix, we find that t16 is the optimal measurement point, and Table 3 shows the calculation process of its heuristic function value. 16 The test results are divided into three sub-matrices The fault fuzzy sets corresponding to the three sub-matrices are Because t 16 The test result of f1 is U, so f1 is not divided into a sub-matrix alone, but is divided into three other matrices. Then, the sub-matrices are repeatedly executed in S5 and S6 until the division is completed.

[0077] Table 3

[0078]

[0079]

[0080] S7, the final test sequence is {t 16 ,t 19 ,t 15 ,t 12 ,t 18 ,t 14}, and there is no fault fuzzy set, so the fault can be effectively identified, the diagnosis tree is as follows Figure 5 shown.

[0081] This embodiment also provides airborne system PHM testability modeling and diagnostic strategy optimization, including:

[0082] The parameter determination module is used to determine the typical failure mode of the complex system, define the probability distribution of the uncertainty parameters, determine the parameters, including: environmental parameters, experience parameters, and failure parameters, where both experience parameters and failure parameters are structural parameters, set the failure threshold of the failure parameter, and when the failure parameter value reaches the failure threshold, the system will fail in the corresponding mode;

[0083] The experimental design module is used to construct the parameter space under the healthy state and different fault states respectively through Latin hypercube sampling, conduct experimental design, obtain the experimental design DOE matrix, and sequentially extract an experimental sample from the experimental design DOE matrix to conduct simulation experiments and obtain the test data under the corresponding state of the system;

[0084] A data mining module is used to perform data mining on the test data obtained in step 2, extract effective characteristic values ​​to replace the test value analysis, and use the characteristic values ​​as test points to monitor the system status;

[0085] The fault diagnosis and optimization module is used to evaluate the detection capability of each test point for different system states based on the receiver operating characteristic curve, quantify the potential of the test point for system fault diagnosis, and construct an uncertainty 4-value correlation matrix (D matrix), which contains 4 states: 1, -1, 0, U, where U is called the uncertain state; determine whether the uncertainty 4-value D matrix is ​​completely divided. If the division is completed, combine all the optimal diagnostic test points to obtain the test sequence, and determine the fault fuzzy group and the isolated fault; if the division is not completed, traverse each test point, construct a heuristic evaluation function with a penalty term based on information gain, select the one with the largest evaluation function value as the optimal diagnostic test point, divide the uncertainty 4-value D matrix into different sub-matrices according to the optimal test point, and for each sub-matrix, determine whether the uncertainty 4-value D matrix is ​​completely divided until the uncertainty 4-value D matrix is ​​completely divided; the criteria for judging whether the matrix division is completed include: only one row is left in the matrix, or all the columns in the matrix are non-information columns.

[0086] This embodiment also provides a storage medium storing a computer program or instruction. When the computer program or instruction is executed, the airborne system PHM testability modeling and diagnostic strategy optimization method is implemented.

[0087] As described above, the apparatus according to the embodiment of the present application can be implemented in various terminal devices, such as a server of a distributed computing system. In one example, the apparatus according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the apparatus can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the apparatus can also be one of the many hardware modules of the terminal device.

[0088] Alternatively, in another example, the apparatus and the terminal device may also be separate terminal devices, and the apparatus may be connected to the terminal device via a wired and / or wireless network, and transmit the interactive information in accordance with an agreed data format.

[0089] The present invention provides a method and device for optimizing PHM testability modeling and diagnostic strategy for airborne systems. There are many methods and approaches to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be considered as the scope of protection of the present invention. All components not specified in this embodiment can be implemented using existing technologies.

Claims

1. Airborne system PHM testability modeling and diagnostic strategy optimization method, characterized by: The following steps are involved: Step 1: Determine the typical failure mode of the complex system, define the probability distribution of the uncertainty parameters, determine the parameters, including: environmental parameters, experience parameters, and failure parameters, where both experience parameters and failure parameters are structural parameters, set the failure threshold of the failure parameter, and when the failure parameter value reaches the failure threshold, the system will fail in the corresponding mode; Step 2: Use Latin hypercube sampling to construct parameter spaces in healthy states and different fault states, perform experimental design, and obtain the experimental design DOE matrix. From the experimental design DOE matrix, extract one experimental sample at a time to perform simulation experiments, and obtain test data under the corresponding state of the system; Step 3: Perform data mining on the test data obtained in step 2, extract effective eigenvalues ​​to replace the test value analysis, and use the eigenvalues ​​as test points to monitor the system status; Step 4: Based on the receiver operating characteristic curve, evaluate the detection capability of each test point for different system states, quantify the potential of the test point for system fault diagnosis, and construct an uncertainty 4-value correlation matrix. The matrix contains 4 states: 1, -1, 0, U, where U is called the uncertain state. Step 5, determine whether the uncertainty 4-value D matrix is ​​completely divided. If the division is completed, jump to step 7; if the division is not completed, execute step 6; the criteria for judging whether the matrix is ​​divided include: only one row is left in the matrix, or the matrix is ​​all non-information columns; Step 6: Traverse each test point, construct a heuristic evaluation function with a penalty term based on information gain, select the one with the largest evaluation function value as the optimal diagnostic test point, divide the uncertainty 4-value D matrix into different sub-matrices according to the optimal test point, and repeat step 5 for each sub-matrix until the uncertainty 4-value D matrix is ​​divided; Step 7: Combine all the optimal diagnostic test points to obtain a test sequence, and determine the fault fuzzy group and the isolated fault; Step 2 includes: dividing each component of the p-dimensional vector into w parts according to the cumulative probability, where w is the number of required sampling points; randomly sampling from the w small intervals in each dimension to form an initial matrix M of w×p: Among them, R wp Represents the element in the wth row and pth column of the initial matrix M; The elements of each column in the initial matrix M are randomly arranged to simulate the random combination of uncertain parameters, as shown below: The elements The subscript X 11 ,X 21 ,…,X w1 It represents the random arrangement of the column vector elements of the initial matrix M, and finally obtains the Latin hypercube sampling matrix, where each row vector is a p-dimensional test sample. According to the obtained DOE matrix, a p-dimensional test sample is taken each time, and simulation experiments are carried out through the model to simulate different working conditions of the system in the actual working environment, and obtain test data under the corresponding state of the system respectively.

2. The method according to claim 1, characterized in that Step 4 includes: uncertainty 4 value D matrix D m×n As shown below: In the uncertainty 4-value D matrix, one row represents a state of the system, and each column represents a test point. The matrix element D(f i ,t j ) = d ij represents the jth test point t j The detection capability of the i-th failure mode.

3. The method according to claim 2, characterized in that Step 4 also includes: given a test point t j and failure mode f i Receiver operating characteristic (ROC) curve The corresponding area under the curve is denoted by According to the distribution of test values ​​in healthy state and distribution under fault conditions Draw the receiver operating characteristic curve; for each pair of test points and corresponding failure mode, d ij The evaluation criteria are as follows: if Above the calibration point and the test value is in the failure mode f i The state is larger than the healthy state, d ij =1; if Above the calibration point and the test value is in the failure mode f i In the state, it is smaller than in the healthy state, d ij = -1; if d ij =0; if Below the calibration point, and d ij =U; The coordinates of the calibration point are (0.05, 0.9), which means that when a single test point is used to detect a fault, the true positive rate needs to be greater than 0.9 and the false positive rate needs to be less than 0.05 to meet the detection requirements; d ij = ±1 means the test point can detect the fault mode; d ij =0 represents and There was no significant difference between ij =U stands for and There were significant differences, but they did not meet the detection requirements.

4. The method according to claim 3, characterized in that Step 6 includes: According to the typical failure mode, the original fault fuzzy set of the system is determined as S = {f0, f1, f2, ..., f m }, f0 represents the health state, f m Indicates the mth fault state, calculate the test point t j When the evaluation function is used, the fault mode corresponding to the uncertain state U will be removed from the original fault fuzzy set of the system to form a new set, which is recorded as the identifiable fault fuzzy set Computing identifiable fault fuzzy sets Information gain in Indicates d ij = the sum of the occurrence probabilities of the failure modes of a, where a represents the three types of detection capabilities of the test point except the uncertain state U, i.e. a = 1, 0, -1; Represents the identifiable fault fuzzy set The sum of the occurrence probabilities of all failure modes in .

5. The method according to claim 4, characterized in that Step 6 also includes: Define if but Penalty Defined as: Where N U represents the number of failure modes corresponding to the uncertain state in the original fault fuzzy set S, N S Indicates the number of failure modes; Final test point t j The heuristic evaluation function h(S; t j ) is defined as follows: Optimal test point t b Select by maximizing the heuristic function value:

6. The method according to claim 5, characterized in that Step 6 also includes: According to the three detection capabilities {-1, 0, 1} of the optimal test point for different faults, the D matrix is ​​divided into three sub-matrices The corresponding three fault fuzzy subsets are: q represents the number of times the matrix is ​​partitioned.

7. The method according to claim 6, characterized in that Step 7 includes: if the uncertainty 4-value D matrix has more than one row, the corresponding fault is judged as a fault fuzzy group; if the uncertainty 4-value D matrix has only one row and the corresponding fault does not appear in other fault fuzzy groups, the fault mode is judged as an isolated fault.

8. The airborne system PHM testability modeling and diagnostic strategy optimization device implemented based on the method of claim 1 is characterized in that: include: The parameter determination module is used to determine the typical failure mode of the complex system, define the probability distribution of the uncertainty parameters, determine the parameters, including: environmental parameters, experience parameters, and failure parameters, where both experience parameters and failure parameters are structural parameters, set the failure threshold of the failure parameter, and when the failure parameter value reaches the failure threshold, the system will fail in the corresponding mode; The experimental design module is used to construct the parameter space under the healthy state and different fault states respectively through Latin hypercube sampling, conduct experimental design, obtain the experimental design DOE matrix, and sequentially extract an experimental sample from the experimental design DOE matrix to conduct simulation experiments and obtain the test data under the corresponding state of the system; A data mining module is used to perform data mining on the test data obtained in step 2, extract effective characteristic values ​​to replace the test value analysis, and use the characteristic values ​​as test points to monitor the system status; The fault diagnosis and optimization module is used to evaluate the detection capability of each test point for different system states based on the receiver operating characteristic curve, quantify the potential of the test point for system fault diagnosis, and construct an uncertainty 4-value correlation matrix, which contains 4 states: 1, -1, 0, U, where U is called the uncertain state; determine whether the uncertainty 4-value D matrix is ​​completely divided. If the division is completed, combine all the optimal diagnostic test points to obtain the test sequence, and determine the fault fuzzy group and the isolated fault; if the division is not completed, traverse each test point, construct a heuristic evaluation function with a penalty term based on information gain, select the one with the largest evaluation function value as the optimal diagnostic test point, divide the uncertainty 4-value D matrix into different sub-matrices according to the optimal test point, and for each sub-matrix, determine whether the uncertainty 4-value D matrix is ​​completely divided until the uncertainty 4-value D matrix is ​​completely divided; the criteria for judging whether the matrix division is completed include: only one row is left in the matrix, or the matrix is ​​all non-information columns.

9. A storage medium, characterized in that: A computer program or instruction is stored, and when the computer program or instruction is executed, the method according to any one of claims 1 to 7 is implemented.

Images