Test excitation optimization method for analog circuit fault diagnosis

The COOT bird model optimization algorithm optimizes the test excitation signal frequency combination, which solves the excitation effect suppression problem caused by complex signal components, and significantly improves the accuracy and efficiency of analog circuit fault diagnosis.

CN118937975APending Publication Date: 2024-11-12HARBIN INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410921645.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-10
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The signal components of existing test excitation signals are complex, and the excitation effects of different frequency components on the fault characteristics may be suppressed by each other, affecting the excitation effect.

Method used

The COOT bird model optimization algorithm is used to analyze the amplitude-frequency characteristic curve chart and the phase-frequency characteristic curve chart, and the optimized test excitation signal frequency combination is selected to form a mixed signal to improve the excitation effect.

Benefits of technology

It significantly improves the accuracy and efficiency of analog circuit fault diagnosis, provides stronger fault characteristic excitation performance, which is better than traditional excitation signals.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118937975A_ABST
    Figure CN118937975A_ABST
Patent Text Reader

Abstract

The invention discloses a test excitation optimization method for fault diagnosis of an analog circuit, and belongs to the technical field of optimization. The invention aims to solve the problem that excitation effects of different frequency components of a pulse source on characteristics are likely to be mutually inhibited, and the excitation effects are influenced. Performing m times of excitation simulation on the same analog circuit under all fault information by using the test excitation signal, and respectively establishing an amplitude-frequency characteristic curve graph and a phase-frequency characteristic curve graph under all fault information according to all amplitude-frequency and phase-frequency response signals of output nodes of the analog circuit; the test excitation signal is a mixed signal formed by P sine waves with different frequencies, and selecting a frequency interval corresponding to an amplitude range from the amplitude-frequency characteristic curve graph as a search space; the COOT bird model optimization algorithm randomly extracts n individuals in the search space for position updating, and a mixed signal formed by sine waves corresponding to n frequencies corresponding to the updated n individuals is obtained and serves as an optimized test excitation signal. The method is used for optimizing the test excitation signal.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a test excitation signal optimization method and belongs to the technical field of optimization. Background Art

[0002] As an important means to improve the reliability of circuit equipment, analog circuit testability design can effectively reduce the impact of faults through test scheme optimization design, diagnostic strategy optimization design, testability verification and evaluation and other stages. The first and most critical part of test scheme optimization design is test stimulus optimization. Whether the test stimulus is properly selected will directly affect the final fault diagnosis efficiency of the analog circuit. Different test stimuli often have different degrees of stimulation for faulty components in the circuit. Choosing a suitable test stimulus can maximize the stimulation of the characteristics of the faulty components in the circuit, providing a good basic condition for subsequent analog circuit fault diagnosis. However, most of the current test stimuli use a single pulse source signal. The signal components of the pulse source are complex, and the stimulation effects of different frequency components on the characteristics may inhibit each other, affecting the stimulation effect. Therefore, in order to improve the stimulation effect, it is necessary to optimize the test stimulus of the circuit. Summary of the invention

[0003] The purpose of the present invention is to solve the problem that the signal components of the pulse source are complex, and the excitation effects of different frequency components on the characteristics may inhibit each other and affect the excitation effect. A test excitation optimization method for analog circuit fault diagnosis is proposed.

[0004] A test stimulus optimization method for analog circuit fault diagnosis, the method comprising the following contents:

[0005] Step 1: Use the test excitation signal to perform m excitation simulations on the same analog circuit under all fault information, and establish amplitude-frequency characteristic curves and phase-frequency characteristic curves under all fault information according to all amplitude-frequency and phase-frequency response signals of the analog circuit output nodes; the amplitude-frequency characteristic curve includes m amplitude-frequency characteristic curves under each fault information, and the phase-frequency characteristic curve includes m phase-frequency characteristic curves under each fault information; wherein the test excitation signal is a mixed signal formed by P sine waves of different frequencies;

[0006] Step 2: Select the amplitude between 0.15*U from the amplitude-frequency characteristic curve. max to 0.85*U max The frequency interval corresponding to the range is used as the search space, U max It is the maximum value in the amplitude-frequency characteristic curve;

[0007] Step 3: The COOT bird model optimization algorithm randomly selects n individuals in the search space to update their positions, and obtains n frequencies corresponding to the updated n individuals. The mixed signal formed by the sine waves corresponding to the n frequencies is used as the optimized test excitation signal, and each individual is the extracted frequency.

[0008] Preferably, the COOT bird model optimization algorithm randomly selects n individuals in the search space to update their positions in the following ways:

[0009] Step 21, combining the ith randomly generated value Li and the search space, determine the initial position of the ith individual, i=1, 2...n; the value range of the randomly generated value is 0 to 1;

[0010] The range from 0 to 1 is evenly divided into 3 ranges, namely, range 1 to range 3. When updating the position of each individual in the jth round, if Li belongs to range 1, the random moving position update method is used to update the position of the i-th individual. If Li belongs to range 2, the chain position update method is used to update the position of the i-th individual. If Li belongs to range 3, the position of the i-th individual is updated based on the position of the group leader. The initial value of j is 1.

[0011] Step 22, determine whether j is equal to the preset maximum number of iterations. If so, stop updating and obtain the frequency value corresponding to the i-th individual position as the i-th frequency in the optimized test excitation signal. If not, set j=j+1 and execute step 21.

[0012] Preferably, in step 21, the initial position CootPos(i)' of the i-th individual is expressed as:

[0013] CootPos(i)'=rand(0,1)*(ub-lb)+lb Formula 1;

[0014] Where ub is the end frequency value of the search space, lb is the starting frequency value of the search space, and rand(0,1) is a randomly generated value ranging from 0 to 1.

[0015] Preferably, in step 21, the random moving position updating method is expressed as:

[0016] CootPos(i)”=CootPos(i)’+A×R2×(Q-CootPos(i)’) Formula 2;

[0017] In the formula, Iter is the maximum number of iterations, j is the number of iterations, R2 is a random number in the interval [0,1], Q is a random position value, and CootPos(i)” is the position of the i-th individual after the update.

[0018] Preferably, the random position value Q is expressed as:

[0019] Q=rand(0,1)*(ub-lb)+lb Formula 3.

[0020] Preferably,

[0021] In step 21, the chained position update method is expressed as:

[0022] CootPos(i)”=0.5×(CootPos(i+1)’+CootPos(i)’) Formula 4;

[0023] Where CootPos(i)' is the position of the i-th individual, CootPos(i+1)' is the position of the i+1-th individual, and CootPos(i)" is the updated position of the i-th individual;

[0024] Formula 4 is used to update the individual positions from the 1st to the n-1th individual positions, and the nth individual position is not updated.

[0025] Preferably, in step 21, the location update method based on the group leader is expressed as:

[0026] According to the formula:

[0027] k=1+(i MOD NL) Formula 5;

[0028] Select NL individual positions from the initial position set or the updated position set as leaders;

[0029] In the formula, NL is the number of leaders, k is the index number of the leader;

[0030] Get the position of the selected leader LeaderPos(k):

[0031]

[0032] In the formula, gBest j To select the position set corresponding to the minimum fitness value from the fitness values ​​of the position sets in each round before the current round, R3 and R4 are random numbers in the interval [0,1], and R is a random number in the interval [-1,1]. Iter is the maximum number of iterations, j is the number of iterations;

[0033] Substitute Formula 6 into Formula 7 to obtain the positions of individuals except the leader position in the initial position set or the updated position set:

[0034] CootPos(i)”=LeaderPos(i)+2×R1×cos(2Rπ)×(LeaderPos(i)-CootPos(i)”) Formula 7;

[0035] Where R1 is a random number in the interval [0,1], and CootPos(i)” is the position of the i-th individual among the other individuals after the update.

[0036] The beneficial effects of the present invention are:

[0037] In view of the problem that the current traditional test excitation of analog circuits may not be able to stimulate effective characteristics of different faults, the present invention proposes an innovative analog circuit fault diagnosis method, which optimizes the test excitation frequency based on the circuit characteristic curve and the COOT bird natural model. First, n individuals are extracted from the amplitude-frequency characteristic curve diagram to form an initial position set. Then, the COOT model optimization algorithm is used to simulate the irregular and regular movements of birds on the water surface to optimize the search space and select the optimal test excitation position set. The COOT model enhances the global search capability and convergence speed of the algorithm by simulating the collective behavior of white-bone birds, including random movement, chain movement, and adjusting position according to the leader.

[0038] When this optimized test stimulus signal is input into an analog circuit, it can stimulate the characteristics of faulty components in the circuit, thereby accurately diagnosing the type of fault in the circuit.

[0039] Therefore, compared with traditional excitation, the optimized excitation signal of the present invention has better excitation performance for faults and can provide strong support for subsequent diagnostic work. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of a test stimulus optimization method for analog circuit fault diagnosis;

[0041] Figure 2 This is the amplitude-frequency characteristic curve obtained by using Pspice software to perform 300 Monte Carlo simulations on a fault in a bandpass filter circuit;

[0042] Figure 3 This is a phase-frequency characteristic curve obtained by using Pspice software to perform 300 Monte Carlo simulations on a fault in a bandpass filter circuit;

[0043] Figure 4 A schematic diagram of updating individual positions using the random movement position update method;

[0044] Figure 5 A schematic diagram of updating individual positions using a chained position update method;

[0045] Figure 6 A schematic diagram of updating individual positions using a group leader-based position update method;

[0046] Figure 7 It is a graph showing the fitness value of the bandpass filter circuit decreasing against the number of iterations;

[0047] Figure 8 is the fault diagnosis confusion matrix of the optimal response signal;

[0048] Fig. 9 is the fault diagnosis confusion matrix of the impulse signal response;

[0049] Fig.10 is the fault diagnosis confusion matrix for random sine wave responses. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0052] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0053] Example:

[0054] Combination Figures 1 to 6 This embodiment describes a test stimulus optimization method for analog circuit fault diagnosis, the method comprising the following contents:

[0055] Step 1: Use the test excitation signal to perform m excitation simulations on the same analog circuit under all fault information, and establish amplitude-frequency characteristic curves and phase-frequency characteristic curves under all fault information according to all amplitude-frequency and phase-frequency response signals of the analog circuit output nodes; the amplitude-frequency characteristic curve includes m amplitude-frequency characteristic curves under each fault information, and the phase-frequency characteristic curve includes m phase-frequency characteristic curves under each fault information; wherein the test excitation signal is a mixed signal formed by P sine waves of different frequencies;

[0056] Step 2: Select the amplitude between 0.15*U from the amplitude-frequency characteristic curve. max to 0.85*U max The frequency interval corresponding to the range is used as the search space, U maxIt is the maximum value in the amplitude-frequency characteristic curve;

[0057] Step 3: The COOT bird model optimization algorithm randomly selects n individuals in the search space to update their positions, and obtains n frequencies corresponding to the updated n individuals. The mixed signal formed by the sine waves corresponding to the n frequencies is used as the optimized test excitation signal, and each individual is the extracted frequency.

[0058] Specifically, n individuals refer to n frequency values;

[0059] In step 3, the simulation circuit test incentive optimization method based on the COOT bird natural model, coots have different collective behaviors, and the goal is to simulate collective motion (regular and irregular motion on the water surface). The whole group is led to the target (food) by several coots in front of the group, which are called group leaders.

[0060] Therefore, we consider four different movements of otters on the water:

[0061] 1. Randomly move to this side and that side;

[0062] 2. Chain movement;

[0063] 3. Adjust position according to team leader;

[0064] 4. The leader leads the group to the optimal area (leader movement);

[0065] This embodiment selects one model from three types of models to continuously update the position set. The composition form of the position set is as follows: F = [f 1 ,f 2 ,...,f n ], when the number of updates reaches the preset maximum number of iterations, the n frequency values ​​in the position set obtained at this time are the optimal frequency values; therefore, as the number of random solutions and optimization steps (iterations) increases, the probability of finding the global optimal solution increases. Update each frequency value in the space of Formula 1.

[0066] The bird model optimization algorithm, also known as the COOT bird natural model optimization method, is a metaheuristic algorithm that simulates two different movement modes of birds on the water. This method simulates two movement modes of birds: in the first stage, the movement of birds is irregular, and in the second stage, the movement of birds is regular. The flock moves towards a leading flock of birds to obtain food supply; the end of the flock moves in the form of a string of coots, and each coot moves behind the coot in front. The overall convergence speed and global search ability of the algorithm are accelerated by the group movement of the leading flock and the chain movement of the tail flock. Finally, the population position is updated. In the position update method based on group leadership, NL number of coots are selected as group leaders. The selection of leaders is random.

[0067] The following further describes how the COOT bird model optimization algorithm implements the position update of n individuals in a randomly selected position set in the search space:

[0068] Step 21, combining the ith randomly generated value Li and the search space, determine the initial position of the ith individual, i=1, 2...n; the value range of the randomly generated value is 0 to 1;

[0069] The range from 0 to 1 is evenly divided into 3 ranges, namely, range 1 to range 3. When updating the position of each individual in the jth round, if Li belongs to range 1, the random moving position update method is used to update the position of the i-th individual. If Li belongs to range 2, the chain position update method is used to update the position of the i-th individual. If Li belongs to range 3, the position of the i-th individual is updated based on the position of the group leader. The initial value of j is 1.

[0070] Step 22, determine whether j is equal to the preset maximum number of iterations. If so, stop updating and obtain the frequency value corresponding to the i-th individual position as the i-th frequency in the optimized test excitation signal. If not, set j=j+1 and execute step 21.

[0071] Specifically, if the maximum number of iterations preset in step 22 is 2, after obtaining n initial frequencies or initial positions, the n initial frequencies or initial positions are updated twice; the update method is determined by a value randomly generated in the range of [0, 1]; if the value randomly generated in the range of [0, 1] is 0.1 during the first update, the first update uses a random moving position update method to update the n initial frequencies to obtain n frequencies after the first update, and then a value randomly generated in the range of [0, 1], if this value is 0.5 at this time, the chain position update method is used to update the n frequencies after the first update to obtain n frequencies after the second update, thereby completing two position updates in this way.

[0072] The following further describes that in step 21, the initial position CootPos(i)' of the i-th individual is expressed as:

[0073] CootPos(i)'=rand(0,1)*(ub-lb)+lb Formula 1;

[0074] Where ub is the end frequency value of the search space, lb is the starting frequency value of the search space, and rand(0,1) is a randomly generated value ranging from 0 to 1.

[0075] Specifically, a number is randomly selected from 0 to 1 and substituted into Formula 1 to obtain a frequency value, and then a number is randomly selected from 0 to 1 and substituted into Formula 1 to obtain a second frequency value. In this way, a total of n frequency values ​​are obtained, also called n individuals, and the n frequency values ​​are used as the initial frequency or initial position.

[0076] The following further describes that in step 21, the random moving position update method is expressed as:

[0077] CootPos(i)”=CootPos(i)’+A×R2×(Q-CootPos(i)’) Formula 2;

[0078] In the formula, Iter is the maximum number of iterations, j is the number of iterations, R2 is a random number in the interval [0,1], Q is a random position value, and CootPos(i)” is the position of the i-th individual after the update.

[0079] Specifically, the method of formula 2 is a random move. To achieve this move, a random position is considered in the search space according to Q = rand (1, d). * (ub-lb) + lb, and coot is moved to this random position. This movement explores different parts of the search space. This movement will cause the algorithm to deviate from the local optimum. The new position of coot is calculated according to formula 2. Figure 4 is the random movement of the coracoid in different directions. Local optimum means that if the 4 frequencies obtained last time are 2, 3, 4, 5, after the 4 frequency values ​​are updated this time, the frequency values ​​obtained are 2, 3, 6, 7. Because 2 frequency values ​​are the same last time and this time, it is equivalent to 2 and 3 being the optimal values, while 6 and 7 are not the optimal values. At this time, 2, 3, 6, 7 is called local optimum.

[0080] As further described below, the random position value Q is expressed as:

[0081] Q=rand(0,1)*(ub-lb)+lb Formula 3.

[0082] The following further describes that in step 21, the chain position update method is expressed as:

[0083] CootPos(i)”=0.5×(CootPos(i+1)’+CootPos(i)’) Formula 4;

[0084] Where CootPos(i)' is the position of the i-th individual, CootPos(i+1)' is the position of the i+1-th individual, and CootPos(i)" is the updated position of the i-th individual;

[0085] Formula 4 is used to update the individual positions from the 1st to the n-1th individual positions, and the nth individual position is not updated.

[0086] Specifically, the movement mode of formula 4 is chain movement, that is: The first method is that the average position of the two bone crested birds can be used to realize chain movement. Another method to realize chain movement is to first calculate the distance vector between the two stakes, and then move the stake toward the other stake by approximately half the distance vector. The first method is used to calculate the new position according to formula 4. Figure 5 shown.

[0087] The following further describes that in step 21, the position update method based on the group leader is expressed as:

[0088] According to the formula:

[0089] k=1+(i MOD NL) Formula 5;

[0090] Select NL individual positions from the initial position set or the updated position set as leaders;

[0091] In the formula, NL is the number of leaders, k is the index number of the leader;

[0092] Get the position of the selected leader LeaderPos(k):

[0093]

[0094] In the formula, gBest j To select the position set corresponding to the minimum fitness value from the fitness values ​​of the position sets in each round before the current round, R3 and R4 are random numbers in the interval [0,1], and R is a random number in the interval [-1,1].

[0095] Iter is the maximum number of iterations, j is the number of iterations;

[0096] Substitute Formula 6 into Formula 7 to obtain the positions of individuals except the leader position in the initial position set or the updated position set:

[0097] CootPos(i)”=LeaderPos(i)+2×R1×cos(2Rπ)×(LeaderPos(i)-CootPos(i)”) Formula 7;

[0098] Where R1 is a random number in the interval [0,1], and CootPos(i)” is the position of the i-th individual among the other individuals after the update.

[0099] Specifically, the group of coots is led by the first few coots, and the rest of the coots must adjust their positions according to the leader of the group and move towards them. A question that may be asked is that each coot will adjust its position according to the leader. The average position of the leader can be considered, and their positions can be updated according to this average position. Considering the average position will lead to premature convergence. In order to achieve this movement, the mechanism of formula 5 is used to select the leader. coot(i) must update its position according to the leader's k. Formula 7 calculates the next position of the coot according to the selected leader. The team must move towards a goal (optimal area), so the leader needs to be updated towards the optimal target area, and the leader's position is updated using formula 6. This formula looks for a better position around the current local advantage. Sometimes, the leader has to leave the current best position and look for a better position. This formula provides a good way to get close to the best position and away from the best position. Through the bird flock algorithm, the position is continuously updated, so that the fitness becomes smaller and better. By calculating the updated fitness after each update, it is found that it is indeed getting better and better using this application, verifying the superiority of this application.

[0100] Experimental verification:

[0101] The method proposed in this embodiment is verified by using the Sallen-Key bandpass filter circuit. First, the simulation software Pspice is used to perform 300 Monte Carlo simulations on various single faults, where various faults are shown in the following table 1:

[0102] Table 1 Single fault code, fault category, nominal value, fault value of Sallen-Key bandpass filter circuit

[0103]

[0104]

[0105] Each simulation obtains an amplitude-frequency characteristic curve and a phase-frequency characteristic curve. 300 simulations obtain 300 amplitude-frequency characteristic curves and 300 phase-frequency characteristic curves, such as Figure 2 and 3As shown in the figure, 300 amplitude-frequency characteristic curves under all fault categories are plotted on a curve chart, and then the frequency points are optimized using the simulated circuit test excitation optimization method based on the COOT bird model. The sweep frequency is selected as 1Hz~100kHz, the circuit amplitude is 5v, and there are 2000 frequency points in total. The transition band is 300Hz~13.323kHz, corresponding to 661 frequency points, corresponding to 80~741 of the upper and lower bounds. The maximum number of iterations is set to 1000, and the population size is set to 100. In order to select test stimuli that can stimulate highly discriminative characteristics, the frequency of the test signal needs to be optimized. Under different fault conditions, the amplitude and phase of the response signal tend to be different. Using this characteristic, the amplitude and phase values ​​corresponding to each discrete frequency point in the amplitude-frequency and phase-frequency curves of the circuit under different fault conditions are taken as individuals in the population. If a circuit requires n excitations, that is, n frequencies, then n frequencies need to be selected from the amplitude-frequency characteristic curve. In order to select a signal range with the most drastic frequency change for frequency point selection, n frequency values ​​are selected from the search space. The n frequency values ​​are F = [f 1 ,f 2 ,...,f n ]. The n amplitudes corresponding to the n frequency values ​​on an amplitude-frequency curve and the n phases corresponding to the n frequency values ​​on a phase-frequency curve form a sample, and each type of fault is regarded as a population.

[0106] The n frequency points initially selected are used as the initial position set, and the values ​​in the position set are continuously updated. The fitness value of the position set is calculated after each update. Therefore, in the present embodiment, the two characteristic curves are combined to obtain the optimal test frequency set for the analog circuit, and this optimal test frequency set is used as the main frequency of the final test excitation sinusoidal wave signal. Then the bird model optimization algorithm is used to optimize the amplitude and phase value of the curve in the frequency set, so that the spacing between every two adjacent curves after optimization is maximized; when the population is updated a preset number of times, the spacing between every two adjacent individuals in the obtained population position is the largest and best, so the fault type can be better judged at this time. After each updated position set is obtained, the fitness function can be used to verify whether the obtained position frequency is the best. The fitness function is introduced below:

[0107] Fitness function: The purpose of analog circuit test excitation optimization is to diagnose different faults more accurately. Therefore, the best optimization goal is to make the amplitude vector U of the fault data set vector =(u 0 ,u 1 ,...u n ) has the largest inter-class distance and the smallest intra-class distance, where n represents the number of frequencies in the test excitation frequency concentration. Therefore, this topic uses the Wilkes statistic as the optimization target, and the definition of the Wilkes statistic is shown in the following formula 10:

[0108]

[0109] Among them, Λ (m) represents the Wilkes statistic corresponding to m types of faults, W represents the intra-class deviation sum of squares matrix, B refers to the inter-class deviation sum of squares matrix, and T refers to the total deviation sum of squares matrix (the intra-class deviation sum of squares matrix: the sum of the intra-class deviation sum of squares matrix and the inter-class deviation sum of squares matrix). The calculation formulas of W and B are shown in the following equations (11) and (12):

[0110]

[0111] In the formula, g represents the total number of sample categories, n represents the number of samples in the i-th category, and x ij refers to the jth sample in the i-th class. represents the mean of the samples of the i-th category, then Represents the mean of the sample population.

[0112] According to the definition of Wilkes statistic, when Λ (m) belongs to [0,1), when Λ (m) When it is equal to 1, it means that the intra-class deviation square sum matrix is ​​equal to the total deviation square sum matrix, which means that the classification ability of these multi-class faults is very poor. The smaller the value of the Wilkes statistic, the greater the proportion of the inter-class deviation square sum to the total deviation square sum. It also means that the classification ability of the corresponding multi-class faults is strong.

[0113] Its fitness decrease curve is as follows Figure 7 As shown, from Figure 7 It can be seen that the curve has been decreasing and now no longer decreases, indicating that the frequency values ​​in the updated position set no longer change. For example, after the last update, the frequencies in the position set were 4.286kHz, 6.639kHz, 10.462kHz, and 20.635kHz. After the next update, they are still 4.286kHz, 6.639kHz, 10.462kHz, and 20.635kHz, indicating that the position set no longer changes and is the optimal position set, which is: 4.286kHz, 6.639kHz, 10.462kHz, and 20.635kHz.

[0114] Then, the circuit was simulated 300 times using Pspice software. Different from the first simulation, the time domain response signal curve of the circuit was obtained this time. Then, the linear kernel support vector machine was used to classify 13 types of faults. The diagnostic results were compared with the target classification effect with an accuracy of 96.32%. Figure 8 shown.

[0115] In order to verify the superiority of the excitation obtained by the present invention and other excitation signals, a pulse signal and a single frequency sine wave signal are selected for comparison. The diagnostic accuracies are 75.6% and 68.3% respectively. The diagnostic confusion matrices are shown as follows: Fig. 9 and Fig.10 As shown, Figure 8 The accuracy is higher. From the accuracy, it can be seen that the superiority of the method proposed in this embodiment.

[0116] The Wilkes statistic is used as the fitness function to measure the distance within and between classes and optimize the population. Finally, the linear kernel support vector machine (SVM) is used to verify the fault diagnosis. The experimental results show that this method significantly improves the accuracy of fault diagnosis and is better than the traditional single pulse signal and single frequency sine wave signal.

[0117] The innovation of the present invention lies in the combination of circuit characteristic curve and bionic optimization algorithm to achieve the optimization of test excitation, which significantly improves the efficiency and accuracy of analog circuit fault diagnosis. It has been verified on the Sallen-Key bandpass filter circuit and can achieve better results than other commonly used excitations.

[0118] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.

Claims

1. A test stimulus optimization method for analog circuit fault diagnosis, characterized in that: The method comprises the following: Step 1: Use the test excitation signal to perform m excitation simulations on the same analog circuit under all fault information, and establish amplitude-frequency characteristic curves and phase-frequency characteristic curves under all fault information according to all amplitude-frequency and phase-frequency response signals of the analog circuit output nodes; the amplitude-frequency characteristic curve includes m amplitude-frequency characteristic curves under each fault information, and the phase-frequency characteristic curve includes m phase-frequency characteristic curves under each fault information; wherein the test excitation signal is a mixed signal formed by P sine waves of different frequencies; Step 2: Select the amplitude between 0.15*U from the amplitude-frequency characteristic curve. max to 0.85*U max The frequency interval corresponding to the range is used as the search space, U max It is the maximum amplitude in the amplitude-frequency characteristic curve; Step 3: The COOT bird model optimization algorithm randomly selects n individuals in the search space to update their positions. Each individual is the extracted frequency, and the n frequencies corresponding to the updated n individuals are obtained. The mixed signal formed by the sine waves corresponding to the n frequencies is used as the optimized test excitation signal.

2. A test stimulus optimization method for analog circuit fault diagnosis according to claim 1, characterized in that: The COOT bird model optimization algorithm randomly selects n individuals in the search space to update their positions in the following ways: Step 21, combining the ith randomly generated value Li and the search space, determine the initial position of the ith individual, i=1, 2...n; the value range of the randomly generated value is 0 to 1; The range from 0 to 1 is evenly divided into 3 ranges, namely, range 1 to range 3. When updating the position of each individual in the jth round, if Li belongs to range 1, the random moving position update method is used to update the position of the i-th individual. If Li belongs to range 2, the chain position update method is used to update the position of the i-th individual. If Li belongs to range 3, the position of the i-th individual is updated based on the position of the group leader. The initial value of j is 1. Step 22, determine whether j is equal to the preset maximum number of iterations. If so, stop updating and obtain the frequency value corresponding to the i-th individual position as the i-th frequency in the optimized test excitation signal. If not, set j=j+1 and execute step 21.

3. A test stimulus optimization method for analog circuit fault diagnosis according to claim 2, characterized in that: In step 21, the initial position of the i-th individual, CootPos(i)', is expressed as: CootPos(i)'=rand(0,1)*(ub-lb)+lb Formula 1; Where ub is the end frequency value of the search space, lb is the starting frequency value of the search space, and rand(0,1) is a randomly generated value ranging from 0 to 1.

4. A test stimulus optimization method for analog circuit fault diagnosis according to claim 3, characterized in that: In step 21, the random moving position update method is expressed as: CootPos(i)”=CootPos(i)’+A×R2×(Q-CootPos(i)’) Formula 2; In the formula, Iter is the maximum number of iterations, j is the number of iterations, R2 is a random number in the interval [0,1], Q is a random position value, and CootPos(i)” is the position of the i-th individual after the update.

5. A test stimulus optimization method for analog circuit fault diagnosis according to claim 4, characterized in that: The random position value Q is expressed as: Q=rand(0,1)*(ub-lb)+lb Formula 3.

6. A test stimulus optimization method for analog circuit fault diagnosis according to claim 2, characterized in that: In step 21, the chained position update method is expressed as: CootPos(i)”=0.5×(CootPos(i+1)’+CootPos(i)’) Formula 4; Where CootPos(i)' is the position of the i-th individual, CootPos(i+1)' is the position of the i+1-th individual, and CootPos(i)" is the updated position of the i-th individual; Formula 4 is used to update the individual positions from the 1st to the n-1th individual positions, and the nth individual position is not updated.

7. A test stimulus optimization method for analog circuit fault diagnosis according to claim 2, characterized in that: In step 21, the position update method based on the group leader is expressed as: According to the formula: k=1+(i MOD NL) Formula 5; Select NL individual positions from the initial position set or the updated position set as leaders; In the formula, NL is the number of leaders, k is the index number of the leader; Get the position of the selected leader LeaderPos(k): In the formula, gBest j To select the position set corresponding to the minimum fitness value from the fitness values ​​of the position sets in each round before the current round, R3 and R4 are random numbers in the interval [0,1], and R is a random number in the interval [-1,1]. Iter is the maximum number of iterations, j is the number of iterations; Substitute Formula 6 into Formula 7 to obtain the positions of individuals except the leader position in the initial position set or the updated position set: CootPos(i)”=LeaderPos(i)+2×R1×cos(2Rπ)×(LeaderPos(i)-CootPos(i)”) Formula 7; Where R1 is a random number in the interval [0,1], and CootPos(i)” is the position of the i-th individual among the other individuals after the update.