Complex Fault Diagnosis Methods, Media, and Systems for Quad-Op-Amplifier Dual Second-Order Analog Circuits

Abstract

This invention discloses a complex fault diagnosis method, medium, and system for a quad op-amp dual second-order analog circuit. The method includes: simulating and acquiring fault feature sample data from the output of the quad op-amp dual second-order analog circuit; classifying and reducing the dimensionality of the fault feature sample data using an improved KCPA algorithm; constructing a fault identification network model; training the fault identification network model to obtain a trained fault identification network model; in practice, sweeping the frequency of the circuit under test to acquire feature data from the output of the circuit under test; reducing the dimensionality of the feature data; and inputting the dimensionality-reduced feature data into the trained fault identification network model. The fault identification network model further reduces the dimensionality of the dimensionality-reduced feature data, identifies the fault, and isolates and locates it. This invention can accurately isolate and locate a specific component experiencing a combined fault when a circuit fault occurs, thereby avoiding secondary effects caused by the incorrect disassembly of normal components.

Description

Technical Field

[0001] This invention belongs to the field of analog circuit fault detection technology, specifically relating to a complex fault diagnosis method, medium, and system for a quad operational amplifier dual second-order analog circuit. Background Technology

[0002] Quad op-amp dual second-order analog circuits are widely used in modern industrial control and signal processing due to their advantages such as good convergence, strong filtering capability, and easily adjustable bandwidth. However, these circuits have complex structures and strong coupling between nodes. When a fault occurs, especially in complex patterns where multiple component faults overlap, it is difficult to effectively locate a specific component, which brings significant challenges to fault diagnosis and equipment maintenance. Especially in important research or production sites, equipment / circuit downtime can cause incalculable economic losses to users. Therefore, researching the fault diagnosis of complex quad op-amp dual second-order circuits is particularly important and has broad application prospects.

[0003] In the prior art, patent CN106597260A discloses a method for diagnosing analog circuit faults based on the singular entropy of the continuous wavelet coefficient matrix Tsallis. This method mainly includes the following steps:

[0004] (1) Obtain the time-frequency domain coefficient matrix of the fault signal through continuous wavelet transform, and then divide the coefficient matrix into 8 sub-matrices;

[0005] (2) By calculating the Tsallis singular entropy of each submatrix, an eigenvector is formed;

[0006] (3) The extracted singular entropy and energy features are classified by using an overlimit learning machine to achieve the goal of fault diagnosis.

[0007] The technical flaw of this patent is that the method only studies the feature extraction problem of a single fault in a four-op-amp dual second-order analog circuit, and does not study or consider the feature extraction problem of multiple fault modes. In addition, the feature vector extracted by this patent has an excessively high dimension, reaching 8 dimensions, which increases the complexity of the circuit diagnosis process. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method for diagnosing complex faults in a quad operational amplifier dual second-order analog circuit.

[0009] Another object of the present invention is to provide a storage medium.

[0010] Another object of the present invention is to provide a complex fault diagnosis system for a quad operational amplifier dual second-order analog circuit.

[0011] To achieve the above objectives, the present invention can be accomplished by adopting the following technical solution:

[0012] A complex fault diagnosis method for a quad op-amp dual second-order analog circuit includes the following steps:

[0013] S1. Obtain fault characteristic sample data of the output of the quad op-amp dual second-order analog circuit through simulation. The fault characteristic data is the effective value of the output voltage signal of the quad op-amp dual second-order analog circuit. The faults include single component faults and multiple component faults of the quad op-amp dual second-order analog circuit.

[0014] S2. The improved KPCA algorithm is used to classify and reduce the dimensionality of the fault feature sample data to obtain a sample dataset after preliminary dimensionality reduction.

[0015] S3. Construct a fault identification network model. Divide the sample dataset obtained in step S2 into a training set, a test set, and a validation set. Use the training set to train the fault identification network model to obtain a trained fault identification network model.

[0016] S4. In practical applications, the frequency of the quad op-amp dual second-order analog circuit under test is swept to obtain the characteristic data output by the quad op-amp dual second-order analog circuit under test. The characteristic data is reduced in dimensionality using step S2, and the reduced characteristic data is input into the fault identification network model trained in step S3. The fault identification network model further reduces the dimensionality of the reduced characteristic data, identifies faults, and isolates and locates faults, and finally completes the diagnosis of the quad op-amp dual second-order analog circuit under test and generates a diagnostic report.

[0017] Preferably, the specific steps of step S2 are as follows:

[0018] S21. Calculate the kernel matrix of the fault feature sample data, wherein the kernel function of the kernel matrix is ​​a Gaussian kernel function; the Gaussian kernel function is expressed as follows:

[0019]

[0020] Where σ is the bandwidth parameter of the Gaussian kernel function, which controls the width of the Gaussian kernel function, k(x n ,x m ) is the Gaussian kernel function, x n x m For feature sample data, ||x n -x m ||For x n ,x m The Euclidean distance between them;

[0021] S22. Perform a centering operation on the kernel matrix obtained in step S21, and convert the centered kernel matrix into an orthogonal matrix;

[0022] S23. Calculate the eigenvalues ​​and eigenvectors of the orthogonal matrix, and filter the eigenvectors based on the contribution rate of the eigenvalues ​​to obtain a set of eigenvectors;

[0023] S24. Obtain the set of coordinates in the reduced high-dimensional space;

[0024] S25. Set the hyperparameter values ​​of the kernel function in step S21 to different values, and repeat steps S21-S24 to obtain different series of high-dimensional space coordinate sets.

[0025] S26. Optimize the hyperparameters of the kernel function by performing optimization calculations on the different series of high-dimensional space coordinate sets obtained in step S25 to obtain a better high-dimensional space coordinate set.

[0026] S27. Construct a preliminary dimensionality-reduced sample dataset based on the improved high-dimensional space coordinate set.

[0027] Preferably, the specific process of step S23 is as follows:

[0028] To ensure that the transformed data maintains consistent importance in fault diagnosis and to facilitate the selection of sample data dimensions, the eigenvectors α of the orthogonal matrix described in step S22 are... j Normalization is represented as follows:

[0029]

[0030] Where, α j Let λ be the eigenvector. j For eigenvalues;

[0031] Then, based on the condition that the eigenvalue contribution rate is greater than 99.9%, the seven α values ​​with the largest eigenvalue contribution rates are selected. j As the feature vector set α test ;

[0032] The contribution rate of the eigenvalue is expressed as follows:

[0033]

[0034] Where, η opt The contribution rate of the eigenvalues, N train To calculate the number of training set samples for the kernel matrix K, the eigenvalues ​​are arranged in descending order, N opt N represents the number of eigenvalues ​​selected. opt ≤N train .

[0035] Preferably, the expression for the optimization calculation in step S26 is as follows:

[0036]

[0037] Among them, Num i_fault Num All_fault N represents the number of fault samples for each type and the total number of fault types. opt The number of selected feature values, i.e., the dimension of the sample data; i, j, k, respectively, represent the number of fault samples, the number of fault types, and the number of sample dimensions; S b ,S w These represent the distance between data means and the standard deviation of the data, respectively, f(S) b ,S w ) is related to σ 2 The relevant objective optimization function.

[0038] Preferably, the fault identification network model adopts a BP neural network, wherein the input layer neurons are set to 7, the hidden layer neurons are set to 10, and the output layer neurons are set to 10.

[0039] Preferably, the BP neural network uses the tan-sigmoid function as the neuron activation function and the Softmax function as the pattern classification function.

[0040] Preferably, during the training, testing, and verification of the fault identification model, the dimensionality of the input data is gradually reduced, and sample data with large standard deviations are removed first, provided that the fault identification model can correctly identify the fault, until the fault identification model can no longer identify the fault correctly.

[0041] Preferably, the simulation settings for the quad operational amplifier dual second-order analog circuit in step S1 are that the resistance value deviates from the nominal value by less than 5% and the capacitance value deviates from the nominal value by less than 10%.

[0042] Preferably, the process of obtaining the output feature data in step S4 is as follows:

[0043] First, test points are set at the total input and total output points of the quad op-amp dual second-order analog circuit under test. The circuit is scanned in a frequency range of 1Hz-10MHz, with the frequency increasing by 10 times. An AC sinusoidal signal is applied to the total input point, and the effective value of the output signal at the total output point is measured. The effective value of the output signal is the characteristic data.

[0044] A storage medium for storing non-transitory computer instructions, which, when executed, perform the complex fault diagnosis method of the quad op-amp dual second-order analog circuit.

[0045] A complex fault diagnosis system for a quad op-amp dual second-order analog circuit is provided to implement a complex fault diagnosis method for the aforementioned quad op-amp dual second-order analog circuit. The system includes:

[0046] The feature acquisition module is used to sweep the frequency of the quad op-amp dual second-order analog circuit under test, then acquire the feature data output by the quad op-amp dual second-order analog circuit under test, and then input the feature data into the feature dimensionality reduction module.

[0047] The feature dimensionality reduction module is used to classify and initially reduce the dimensionality of the input feature data using an improved KPCA algorithm, and to construct an input sample set based on the feature data;

[0048] The fault identification module is used to further reduce the dimensionality of the input sample set, identify faults, and isolate and locate faults using a fault identification network model, and finally outputs a diagnostic report of the quad operational amplifier dual second-order analog circuit under test.

[0049] The present invention has the following advantages over the prior art:

[0050] (1) The complex fault diagnosis method for the quad op-amp dual second-order analog circuit of the present invention acquires fault feature sample data of single fault and multiple faults of the analog circuit through simulation, and effectively classifies and reduces the dimensionality of the fault feature sample data through the improved KPCA dimensionality reduction classification algorithm to obtain a preliminary dimensionality reduction sample dataset. The preliminary dimensionality reduction sample dataset is used to train the constructed fault identification network model to obtain the trained fault identification network model. In practical application, the quad op-amp dual second-order analog circuit under test is first swept to obtain the feature data output by the quad op-amp dual second-order analog circuit under test. Then, the improved KPCA dimensionality reduction classification algorithm is used to reduce the dimensionality of the feature data. The trained fault identification network model is used to further reduce the dimensionality of the feature data, identify the fault, and isolate and locate the fault, and finally completes the diagnosis of the quad op-amp dual second-order analog circuit under test. This solves the problem in the prior art that there is no feature data extraction and fault diagnosis and location for the quad op-amp dual second-order analog circuit under multiple fault modes.

[0051] (2) The complex fault diagnosis method of the quad operational amplifier dual second-order analog circuit of the present invention adopts the KPCA dimensionality reduction classification algorithm, which can effectively reduce the original fault feature sample data from 71 dimensions to 7 dimensions of fault feature sample data. Then, the fault identification network filters and reduces the 7-dimensional fault feature sample data to 1-dimensional fault feature sample data, thereby reducing the size of the data space and improving the accuracy of classification.

[0052] (3) The complex fault diagnosis system of the quad operational amplifier dual second-order analog circuit of the present invention can accurately isolate and locate a specific component that has a combined fault when the circuit fails, avoid the secondary effects caused by the incorrect disassembly of normal components, and facilitate the precise repair of the analog circuit board. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the complex fault diagnosis method for the four operational amplifier dual second-order analog circuit of the present invention.

[0054] Figure 2 This is a schematic diagram of the circuit structure of the complex fault diagnosis method for the four operational amplifier dual second-order analog circuit of the present invention.

[0055] Figure 3 This is a schematic diagram of fault types represented by 1D fault feature samples, illustrating the complex fault diagnosis method for the quad operational amplifier dual second-order analog circuit of the present invention.

[0056] Figure 4 This is a schematic diagram of the complex fault diagnosis system of the four operational amplifier dual second-order analog circuit of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0059] Example 1

[0060] A complex fault diagnosis method for a quad op-amp dual second-order analog circuit includes the following steps:

[0061] S1. Obtain fault characteristic sample data of the output of the quad op-amp dual second-order analog circuit through simulation. The fault characteristic data is the effective value of the output voltage signal of the quad op-amp dual second-order analog circuit. The fault includes single component faults and multiple component faults of the quad op-amp dual second-order analog circuit.

[0062] like Figure 1As shown, a complex fault occurs when two or more components in a quad op-amp dual second-order analog circuit fail. Below are 10 fault modes, including single faults, multiple faults, and normal operation modes, for components R1, R2, R3, and R4. The multiple fault mode considers the case where two components fail simultaneously. The fault set is represented as follows:

[0063] All Fault ={normal,R1↑,R1↓,R2↑,R2↓,R3↑,R3↓,R4↑,R4↓,R1↑∩R4↓}

[0064] It should be noted that the component failures described in this invention refer to the failures of resistors or capacitors in the circuit, and amplifiers or other components are not considered at this time.

[0065] Simulations were performed on the above 10 types of faults, and then the fault characteristic sample data of the output were obtained. The process is as follows:

[0066] To simulate thermal noise and circuit interference during actual circuit operation, the simulation settings were set such that the resistance value deviated from the nominal value by less than 5% and the capacitance value deviated from the nominal value by less than 10%. The fault characteristic sample data of the 10 faults were obtained through Monte Carlo simulation, with 400 samples for each type of fault, for a total of 4000 samples. The final simulation results of some fault characteristic sample data under different faults are shown in Table 1 below.

[0067]

[0068]

[0069] Table 1 Sample data of partial fault characteristics under different faults

[0070] S2. The improved KPCA algorithm is used to classify and reduce the dimensionality of the fault feature sample data to obtain a preliminary dimensionality-reduced sample dataset; specifically as follows:

[0071] S21. Calculate the kernel matrix of the fault feature sample data; specifically, taking 400 samples of each of the 10 types of faults obtained above as an example, select 100 data points as the training set for kernel matrix calculation, and the kernel matrix is ​​represented as follows:

[0072] (K) nm =k(x n ,x m n, m = 1, 2, ..., N train N train =100

[0073] The kernel function of the kernel matrix is ​​a Gaussian kernel function; the Gaussian kernel function is expressed as follows:

[0074]

[0075] Where σ is the bandwidth parameter of the Gaussian kernel function, which controls the width of the Gaussian kernel function, k(x n ,x m ) is the Gaussian kernel function, x n x m For feature sample data, ||x n -x m ||For x n ,x m The Euclidean distance between them.

[0076] 2σ 2 =100, as the initial value, calculate the kernel matrix K for each sample, where the size of the kernel matrix K is 100x100.

[0077] S22. Perform a centering operation on the kernel matrix K obtained in step S21, and transform the centered kernel matrix into an orthogonal matrix K. C ;

[0078] S23. Calculate the orthogonal matrix K. C eigenvalues ​​λ j and eigenvector α j The feature vectors are then filtered based on their contribution rate to obtain a feature vector set; the specific process is as follows:

[0079] To ensure that the transformed data maintains consistent importance in fault diagnosis and to facilitate the selection of sample data dimensions, the eigenvectors α of the orthogonal matrix described in step S22 are... j Normalization, eigenvector α j It is expressed as follows:

[0080]

[0081] Where, α j Let λ be the eigenvector. j For eigenvalues;

[0082] Then, based on the condition that the eigenvalue contribution rate is greater than 99.9%, the seven α values ​​with the largest eigenvalue contribution rates are selected. j As the feature vector set α test ;

[0083] The contribution rate of the eigenvalue is expressed as follows:

[0084]

[0085] Where, η opt The contribution rate of the eigenvalues, N trainTo calculate the number of training set samples for the kernel matrix K, the eigenvalues ​​are arranged in descending order, N opt N represents the number of eigenvalues ​​selected. opt ≤ N train .

[0086] Due to N opt The larger the value, the greater η. opt To maximize the acquisition of projected coordinates in the high-dimensional feature space of the sample data, this embodiment sets η to be higher. opt ≥99.9%.

[0087] S24. Obtain the set of coordinates in the reduced high-dimensional space; details are as follows:

[0088] First, take 2σ 2 =100, perform kernel matrix calculation on each of the remaining 300 samples in the sample set from step S21 and the 100 samples in the training set, denoted as K. test The size of this set is 100×300, resulting in the dimensionality-reduced high-dimensional coordinate set, as shown below:

[0089] z = α test T K test

[0090] Where Z is the high-dimensional coordinate set; α test T For the feature vector set α test The transpose of the matrix, and the size of the reduced high-dimensional space coordinate set is 7×300.

[0091] S25. Set the hyperparameter values ​​of the kernel function in step S21 to different values, and repeat steps S21-S24 to obtain different series of high-dimensional space coordinate sets.

[0092] S26. Optimize the hyperparameters of the kernel function by performing optimization calculations on the different series of high-dimensional space coordinate sets obtained in step S25 to obtain a better high-dimensional space coordinate set; specifically as follows:

[0093] The expression for the optimization calculation is as follows:

[0094]

[0095] Among them, Num i_fault Num All_fault N represents the number of fault samples for each type and the total number of fault types. opt The number of selected feature values, i.e., the dimension of the sample data; i, j, k, respectively, represent the number of fault samples, the number of fault types, and the number of sample dimensions; S b ,S wThese represent the distance between data means and the standard deviation of the data, respectively, f(S) b ,S w ) is related to σ 2 The relevant objective optimization function.

[0096] f(S b ,S w The expression indicates that the objective optimization selection considers not only the distance between the data and the mean, but also the data dispersion. This is because, in actual circuit fault mode classification, there may be large distances between samples of different types, but the precision of samples of the same type is not good enough, resulting in a large standard deviation and leading to classification errors. Calculations show that when the hyperparameter is approximately 2σ... 2 The target value reaches its maximum when the hyperparameter is 3000. Table 2 shows the dimensionality reduction results for different types of sample data when the hyperparameter is 3000.

[0097]

[0098] Table 2. High-dimensional spatial coordinate set after dimensionality reduction by KPCA algorithm

[0099] S27. Construct a preliminary dimensionality-reduced sample dataset based on the improved high-dimensional space coordinate set.

[0100] S3. Construct a fault identification network model. Divide the sample dataset obtained in step S2 into a training set, a test set, and a validation set. Use the training set to train the fault identification network model to obtain a trained fault identification network model.

[0101] Specifically, the sample dataset after initial dimensionality reduction is divided into three parts: 70% as the training set, 15% as the test set, and 15% as the validation set.

[0102] The fault identification network model adopts a BP neural network, with 7 neurons in the input layer, 10 neurons in the hidden layer, and 10 neurons in the output layer.

[0103] Specifically, the fault identification network outputs 10 elements, each corresponding to a working mode. The element value can be 0 or 1, where 1 represents true and 0 represents false. For example, an output value of {1,0,0,0,0,0,0,0,0,0} indicates that the circuit is working normally and there are no other faults.

[0104] The BP neural network uses the tan-sigmoid function as the neuron activation function and the Softmax function as the pattern classification function. The goal is to achieve a minimum error between the classification result and a predetermined value. The network is trained, tested, and validated through multiple iterations, ending around 70 iterations to find the optimal connection weights and biases for each layer. Specifically, the trained fault identification model can correctly identify each type of fault pattern. In this embodiment, a Softmax function output value higher than 99.8% is considered a correct classification.

[0105] During training, testing, and verification, the fault identification model gradually reduces the dimensionality of the input data and prioritizes removing sample data with large standard deviations, provided that it can correctly identify faults, until the fault identification model can no longer identify them correctly.

[0106] Specifically, when the fault identification model is working, it optimizes the seven dimensionality-reduced input samples, selecting the sample combination with the smallest cross-entropy value. This value is then used to classify the ten different data samples. During the optimization selection, a one-dimensional sample data dimension is sufficient. The one-dimensional data for the ten different fault types is as follows: Figure 3 As shown in the figure. At this point, the number of dimensions m after dimensionality reduction of the sample data is 1, which is also the optimal number of dimensions. This setting can reduce the size of the data space and improve the accuracy of classification.

[0107] Table 3 below compares the results obtained by classifying the input sample test set and circuit fault mode set using the improved KPCA algorithm adopted in this embodiment with those obtained by using multiple methods such as principal component analysis (PCA), linear discriminant analysis (LDA), and PCA+LDA combining principal component analysis and linear discriminant analysis.

[0108]

[0109] Table 3. Comparison of classification results between the improved KPCA algorithm and other algorithms.

[0110] The following conclusions can be drawn from Table 3:

[0111] The improved KPCA algorithm used in this invention achieves the best dimensionality reduction classification effect. The minimum dimensionality requirements for accurate classification for the four dimensionality reduction algorithms listed in Table 3 are: PCA-LDA 7 dimensions, LDA 33 dimensions, PCA 6 dimensions, and the algorithm of this invention 1 dimension. Simultaneously, this invention also directly inputs the original sample data into the neural network for fault diagnosis. The results show that this method can also achieve classification, but the classification accuracy is not high. This indicates that some data in the original dataset can introduce errors into the classification, thus affecting the accuracy. This also demonstrates the importance of dimensionality reduction for the original data.

[0112] S4. In practical applications, the frequency of the quad op-amp dual second-order analog circuit under test is swept to obtain the characteristic data output by the quad op-amp dual second-order analog circuit under test. The characteristic data is reduced in dimensionality using step S2, and the reduced characteristic data is input into the fault identification network model trained in step S3. The fault identification network model further reduces the dimensionality of the reduced characteristic data, identifies faults, and isolates and locates faults, and finally completes the diagnosis of the quad op-amp dual second-order analog circuit under test and generates a diagnostic report.

[0113] The process of obtaining the output feature data in step S4 is as follows:

[0114] Since the total output point of the quad op-amp dual second-order analog circuit has the largest number of associated components, and since an excitation signal needs to be applied to the total input point of the same circuit in order to obtain data from the total output point, the test points are initially set at the total input and total output points. If the subsequent fault isolation performance does not meet the requirements, other points in the circuit are added as test points and the test is repeated. Then, the quad op-amp dual second-order analog circuit under test is scanned in a frequency range of 1Hz-10MHz, increasing by a factor of 10, while an AC sinusoidal signal is applied to the total input point. The effective value of the output signal at the total output point is measured, and this effective value is the characteristic data.

[0115] Specifically, the isolation index is set to 100%, which is the amplitude-frequency data fault isolation rate of the total input point, that is, the ratio of the number of faults isolated by means and methods to the total number of faults detected.

[0116] Example 2

[0117] A storage medium for storing non-transitory computer instructions, which, when executed, perform a complex fault diagnosis method for a quad op-amp dual second-order analog circuit as described in Example 1.

[0118] Example 3

[0119] A diagnostic system for a quad op-amp dual second-order analog circuit is provided to implement the complex fault diagnosis method for the quad op-amp dual second-order analog circuit described in Example 1. The system includes:

[0120] The feature acquisition module is used to sweep the frequency of the quad op-amp dual second-order analog circuit under test, then acquire the feature data output by the quad op-amp dual second-order analog circuit under test, and then input the feature data into the feature dimensionality reduction module.

[0121] The feature dimensionality reduction module is used to classify and initially reduce the dimensionality of the input feature data using an improved KPCA algorithm, and to construct an input sample set based on the feature data;

[0122] The fault identification module is used to further reduce the dimensionality of the input sample set, identify faults, and isolate and locate faults using a fault identification network model, and finally outputs a diagnostic report of the quad operational amplifier dual second-order analog circuit under test.

[0123] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A complex fault diagnosis method for a four operational amplifier biquad analog circuit, characterized by, Comprise the following steps: S1, obtain the fault feature sample data of the output of the four operational amplifier double second order analog circuit through simulation, the fault feature data is the effective value of the output voltage signal of the four operational amplifier double second order analog circuit; the fault includes single component fault and multiple component fault of the four operational amplifier double second order analog circuit; S2, the improved KPCA algorithm is used to classify and reduce the dimension of the fault feature sample data, and a preliminary reduced sample data set is obtained; S21, the kernel matrix of the fault feature sample data is calculated, and the kernel function of the kernel matrix adopts Gaussian kernel function; the Gaussian kernel function is represented as follows: wherein σ is a bandwidth parameter of a Gaussian kernel function, k(x n ,x m ) is a Gaussian kernel function, x n , x m are feature sample data, and ||x n -x m || is an Euclidean distance between x n , x m . S22, the kernel matrix obtained in step S21 is subjected to centering operation, and the kernel matrix after centering operation is converted into an orthogonal matrix; S23, the eigenvalue and eigenvector of the orthogonal matrix are calculated, the eigenvectors are screened with the condition of eigenvalue contribution rate, and a set of eigenvectors is obtained; S24, a high-dimensional space coordinate set after dimension reduction is obtained; S25, the hyperparameter value of the kernel function in step S21 is set to different values, steps S21-S24 are repeated, and different series of high-dimensional space coordinate sets are obtained; S26, the hyperparameter optimization of the kernel function, the different series of high-dimensional space coordinate sets obtained in step S25 are subjected to optimization calculation, and a more optimal high-dimensional space coordinate set is obtained; The expression of the optimization calculation in step S26 is represented as follows: wherein, N um i_fault , Num All_fault is the number of each type of fault sample and the total number of fault types, N opt is the number of selected feature values, that is, the dimension of sample data; i, j, k are respectively the number of fault samples, the number of fault types, and the number of sample dimensions; S b , S w are respectively the distance between data mean values and the data standard deviation, f(S b , S w ) is a target optimization function related to σ 2 ; S27, a preliminary reduced sample data set is constructed based on the more optimal high-dimensional space coordinate set; S3, a fault recognition network model is constructed, a training set, a test set and a validation set are divided from the preliminary reduced sample data set obtained in step S2, the training set is used to train the fault recognition network model, and a trained fault recognition network model is obtained; S4, in actual application, the four operational amplifier double second order analog circuit to be tested is swept, the feature data of the output of the four operational amplifier double second order analog circuit to be tested is obtained, the feature data is reduced in step S2, and the reduced feature data is input into the trained fault recognition network model in step S3, the fault recognition network model is further reduced, fault recognition and isolation positioning are carried out on the reduced feature data, and finally the diagnosis of the four operational amplifier double second order analog circuit to be tested is completed, and a diagnosis report is generated.

2. The complex fault diagnosis method of a biquad analog circuit using four operational amplifiers according to claim 1, wherein The specific process of step S23 is as follows: The eigenvector α of the orthogonal matrix described in step S22 is normalized, represented as follows: j Normalization, represented as follows: wherein α j is an eigenvector, and λ j is an eigenvalue. Then, with the condition that the eigenvalue contribution rate is greater than 99.9%, the 7 α with the largest eigenvalue contribution rate are selected j As the feature vector set α test ; The expression of the eigenvalue contribution rate is as follows: where η opt is the contribution rate of the eigenvalue, N train is the number of training set samples for calculating the kernel matrix K, N opt is the number of selected eigenvalues, N opt ≤ N train .

3. The complex fault diagnosis method of biquad analog circuit with four operational amplifiers according to claim 1, characterized in that, The fault recognition network model adopts BP neural network, the input layer neurons of the BP neural network are set to 7, the hidden layer neurons are set to 10, and the output layer neurons are set to 10.

4. The complex fault diagnosis method of a biquad analog circuit using four operational amplifiers according to claim 3, wherein The BP neural network adopts tan-sigmoid function as the neuron activation function, and adopts Softmax function as the pattern classification function.

5. The method of claim 1, wherein, The simulation setting condition of the four operational amplifier double second order analog circuit in step S1 is that the resistance value deviates from the nominal value within 5%, and the capacitance value deviates from the nominal value within 10%.

6. The complex fault diagnosis method of biquad analog circuit with four operational amplifiers according to claim 1, characterized in that, The acquisition process of the output feature data in step S4 is as follows: First, test points are set at the total input point and the total output point of the four operational amplifier biquad analog circuit to be tested; the four operational amplifier biquad analog circuit to be tested is scanned in a frequency range of 1 Hz-10 MHz with 10 times frequency increment, an AC sinusoidal signal is applied to the total input point, and the output signal effective value of the total output point is measured, which is the characteristic data.

7. A storage medium, characterized by A computer storage medium storing non-transitory computer instructions, when executed, performs the complex fault diagnosis method of the four operational amplifier biquad analog circuit according to any one of claims 1-6.

8. A complex fault diagnosis system of a four operational amplifier biquad analog circuit for implementing the complex fault diagnosis method of the four operational amplifier biquad analog circuit according to any one of claims 1 to 6, characterized by, Comprise: a characteristic acquisition module for sweeping frequency on the four operational amplifier biquad analog circuit to be tested, then acquiring the characteristic data output by the four operational amplifier biquad analog circuit to be tested, and then inputting the characteristic data into a characteristic dimension reduction module; the characteristic dimension reduction module is used for classifying and preliminarily reducing the dimension of the input characteristic data by using an improved KPCA algorithm, and constructing an input sample set based on the characteristic data; a fault identification module for further reducing the dimension, identifying the fault and isolating and positioning of the input sample set by using a fault identification network model, and finally outputting a diagnosis report of the four operational amplifier biquad analog circuit to be tested.

Images