CN104849650B - One kind is based on improved analog-circuit fault diagnosis method - Google Patents

Abstract

An improved analog circuit fault diagnosis method, which improves the traditional technology from two aspects: 1. For the improvement of the DAGSVM method, the SVM with the largest inter-class distance is used as the top node of the DAGSVM. If the root node classification result is class i , then select the SVM with the largest distance from class i as the child node of this layer. If the classification result of the root node is class j, select the SVM with the largest distance from class j as the child node of this layer; if the classification result does not belong to Class i does not belong to class j, then exclude these two classes, select the two classes of SVMs with the largest inter-class distance D _ij among the remaining classes as the nodes of this layer and continue the above two steps until the diagnosis result is obtained. Doing so can effectively avoid high-level nodes from diagnosing errors and causing errors in the final result. 2. In order to improve the diagnostic accuracy of each sub-node, the particle swarm optimization algorithm (PSO) is used to optimize the parameters of each sub-SVM to improve the diagnostic accuracy of the SVM of each node, so as to improve the diagnostic accuracy of the entire DAGSVM.

Description

An Improved Analog Circuit Fault Diagnosis Method

technical field

The invention relates to the field of analog circuit fault diagnosis, in particular to an improved analog circuit fault diagnosis method.

Background technique

With the continuous development of electronic technology and computer technology, the structure of equipment is becoming more and more complex. However, the analog circuit in the equipment is often prone to problems, resulting in high test time and test cost. With the development trend of circuit structures expanding and becoming more and more complex, the difficulty of testing has also become higher. The scale and complexity of modern circuit systems are getting larger and higher, and the requirements for reliability and fault diagnosis of circuit systems are also increasing. Theoretical analysis and practical application show that analog circuits are more prone to failure than digital circuits. Although digital circuits in electronic equipment exceed 80%, more than 80% of the failures come from analog circuits.

At present, some achievements have been made in the field of analog circuit fault diagnosis. Process the response signal through wavelet transform, wavelet packet decomposition, Hilbert-Huang transform or principal component analysis, extract fault features, and then realize it through expert system, neural network, support vector machine, fuzzy technology, rough set and fractal theory, etc. Troubleshooting. At present, the main problems of analog circuit fault diagnosis are insufficient fault samples and the problem of diagnostic knowledge discovery. Support vector machine (SVM) can solve the small sample problem very well, and use limited features to maximize the detection of hidden classification knowledge in data. . However, SVM was originally used to solve the problem of binary classification. At present, the commonly used methods of SVM to realize multi-classification include one-to-many (OVR), one-to-one (OVO) and directed acyclic graph (DAG). OVR training samples are unbalanced, resulting in low classification accuracy. Although OVO has higher accuracy than the OVR method, due to the voting mechanism, there will be cases where no samples belong to multiple classes at the same time or do not belong to any class. DAGSVM uses the exclusion method to classify the shortcomings of the OVO algorithm. Each sub-SVM classifier excludes a most unlikely class, and finally the classification result can be obtained. The DAGSVM algorithm also has certain problems: when the classification result of the previous node is wrong, the classification result of the subsequent node is also wrong, so that the correct result cannot be obtained.

Contents of the invention

In view of this, the purpose of the present invention is to propose a fault diagnosis method based on the improved analog circuit, which improves the wrong classification result if the previous node is classified incorrectly in the DAGSVM algorithm, and places the SVM with the highest classification accuracy in the DAGSVM algorithm. The top layer, and use the PSO algorithm to improve the classification accuracy of each SVM, minimize the classification error of the upper node, so that the fault recognition rate is improved.

To achieve the above object, the present invention provides the following technical solutions:

A fault diagnosis method based on an improved analog circuit is characterized in that it includes the following steps: Step 1: signal acquisition, collecting signals from specific nodes of the analog circuit; Step 2: extracting fault features, performing wavelet packet processing on the collected signals Decompose and normalize processing to obtain the fault features; Step 3: Use these fault features for child node PSOSVM to train respectively, and calculate the distance between classes; Step 4: Replace the SVM at the child node in DAGSVM with PSOSVM, and Each child node is determined in turn from top to bottom by the inter-class distance.

Further, the signal acquisition method described in step 1 specifically includes the following steps: use Pspice to simulate, perform Monte Carlo analysis on the circuit, sample the output terminal voltage signal, and collect 500 data to obtain samples;

Further, the feature extraction method described in step 2 specifically includes the following steps: 31: using the db2 wavelet in the wavelet system to perform wavelet packet decomposition on the denoised signal, and then extract all sub-bands from low frequency to high frequency in the Nth layer 32: Reconstruct the signal S _j of each sub-band range on the Nth layer according to the wavelet packet decomposition coefficient; 33: According to E _j =∫|s _j (t)| ² dt=∑ _{k= 1} ⁿ |x _ik | ² The energy E _j of each sub-band signal, where x _ik is the amplitude of the discrete signal reconstruction point, and j is from 1 to N; 34: Constructing the eigenvector X=[E1, E2,..., E _N ], and normalize it to obtain the analog circuit fault feature vector;

Further, the PSOSVM method described in step 3 specifically includes the following steps: 41: input the fault feature as training data, and randomly generate initial particles and initial parameters; 42: establish an SVM model, and use the accuracy of the SVM as the fitness of the particles , if the fitness is better than the optimal fitness of the particle, the position vector at this time is stored as the position vector of the particle, and if the fitness of the particle is better than the global optimal fitness, the position vector is stored as the global optimal. 43: Repeat the above steps until the final criterion is met or the set maximum number of iteration steps is reached, and the optimized parameters are obtained: the penalty parameter C and the kernel function parameter γ;

Further, the improved DAGSVM method described in step 4 specifically includes the following steps: 51: calculate the inter-class distance of every two types of faults by D _ij =||m _i -m _j || ² -r _i -r _j ; 52: Arrange the inter-class distances from large to small, and take the PSOSVM classifier with the largest inter-class distance D _ij as the top node classifier; 53: If the root node classification result is class i, select the class with the largest inter-class distance from class i The SVM classifier is used as a child node of this layer; if the classification result of the root node is class j, select the SVM classifier with the largest distance from class j as a child node of this layer; if the classification result neither belongs to class i nor class j , then exclude these two categories, select the two SVM classifiers with the largest inter-class distance D _ij in the remaining classes as the top node classifier and continue the above two steps until the diagnosis result is obtained;

The present invention proposes an improved analog circuit fault diagnosis method, which can effectively prevent final result diagnosis errors caused by upper node diagnosis errors and improve diagnosis reliability.

Description of drawings

In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail below in conjunction with the accompanying drawings, wherein:

Figure 1 is a schematic diagram of the DAG-SVM classification algorithm;

Figure 2 is a schematic diagram of the improved analog circuit fault diagnosis process;

Fig. 3 is a flow diagram of PSOSVM.

Detailed ways

The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

Figure 2 is a schematic diagram of the improved analog circuit fault diagnosis process. The invention provides an improved analog circuit fault diagnosis method, which mainly includes the following steps: analog circuit signal acquisition, fault feature extraction, and fault diagnosis using an improved DAGSVM. Each part of the process includes the following steps:

S1: Analog circuit signal acquisition. In this step, use Pspice simulation software to conduct MonteCarlo analysis on the circuit, sample the output voltage signal, and collect 500 data to obtain samples;

S2: Fault feature extraction. In this step, wavelet packet decomposition and normalization are used to obtain fault features of analog circuits. The specific steps are as follows:

S21: Using the db2 wavelet in the wavelet system to perform wavelet packet decomposition on the denoised signal, and then extract the signal features of each frequency component in all sub-frequency bands of the Nth layer from low frequency to high frequency;

S22: Reconstruct the signal S _j of each sub-band range on the Nth layer according to the wavelet packet decomposition coefficient;

S23: According to each E _j =∫|s _j (t)| ² dt=∑ _k=1 ⁿ |x _ik | ² sub-band signal energy E _j , where x _ik is the amplitude of the discrete signal reconstruction point, j is 1 to N;

S24: Construct a feature vector X=[E1, E2, . . . , E _N ], and normalize it to obtain an analog circuit fault feature vector.

S3: Use the improved DAGSVM for fault diagnosis. In this step, build the improved DAGDVM model to realize the fault diagnosis of analog circuits. It mainly includes the following two parts: training PSOSVM and constructing the improved DAGSVM model. The specific steps of each part are as follows:

S31: Training PSOSVM

S311: The fault features are input as training data, and initial particles and initial parameters are randomly generated;

S312: Establish an SVM model, and use the accuracy of the SVM as the fitness of the particle. If the fitness is better than the optimal fitness of the particle, the position vector at this time is stored as the position vector of the particle. If the fitness of the particle is better than the global Optimal fitness, the position vector is stored as the global optimum;

S313: Repeat the above steps until the final criterion is satisfied or the set maximum number of iteration steps is reached, and the optimized parameters are obtained: the penalty parameter C and the kernel function parameter γ.

S32: Build an improved DAGSVM model

S321: Calculate the inter-class distance of every two types of faults by D _ij =||m _i -m _j || ² -r _i -r _j ;

S322: arrange the inter-class distances from large to small, and take the PSOSVM classifier with the largest inter-class distance D _ij as the uppermost node classifier;

S323:) If the classification result of the root node is class i, select the SVM classifier with the largest distance from class i as the child node of this layer; if the classification result of the root node is class j, select the classifier with the largest distance from class j The SVM classifier is used as a child node of this layer; if the classification result neither belongs to class i nor class j, these two classes are excluded, and the two classes of SVM classifiers with the largest inter-class distance D _ij are selected as the final class among the remaining classes. The upper node classifier continues the above two steps until the diagnosis result is obtained.

Through the above steps, the fault diagnosis of the analog circuit can be realized.

Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that it can be described in terms of form and Various changes may be made in the details without departing from the scope of the invention defined by the claims.

Claims (4)

1. A fault diagnosis method based on an improved analog circuit, is characterized in that comprising the steps: Step 1: signal collection, collecting signals from specific nodes of the analog circuit; Step 2: Fault feature extraction, performing wavelet packet decomposition and normalization processing on the collected signals to obtain fault features; Step 3: Use these fault features to train the sub-node PSOSVM respectively, and calculate the distance between classes; Step 4: Replace the SVM at the sub-nodes in DAGSVM with PSOSVM, and determine each sub-node sequentially from top to bottom through the inter-class distance; In the PSOSVM method described in step three, the specific steps are as follows: 41: the fault feature is input as training data, and initial particles and initial parameters are randomly generated; 42: the SVM model is established, and the accuracy of the SVM is used as the fitness of the particles. The fitness is better than the optimal fitness of the particle, and the position vector at this time is stored as the position vector of the particle. If the fitness of the particle is better than the global optimal fitness, the position vector is stored as the global optimal; 43: Repeat the above steps Until the final criterion is met or the set maximum number of iteration steps is reached, the optimized parameters are obtained: the penalty parameter C and the kernel function parameter γ. 2. a kind of fault diagnosis method based on improved analog circuit according to claim 1, is characterized in that: the signal acquisition method described in the step 1, concrete steps are as follows: Use Pspice to simulate, conduct Monte Carlo analysis on the circuit, sample the output voltage signal, and collect 500 data to obtain samples. 3. a kind of fault diagnosis method based on improved analog circuit according to claim 1, it is characterized in that: in the feature extraction method described in step 2, concrete steps are as follows: 31: after adopting db2 wavelet in the wavelet system to denoise Decompose the signal by wavelet packet, and then extract the signal features of each frequency component in all sub-bands of the Nth layer from low frequency to high frequency; 32: Reconstruct the signal S _j of each sub-band range on the Nth layer according to the wavelet packet decomposition coefficient ;33: According to The energy E _j of each sub-band signal, where x _1k is the amplitude of the discrete signal reconstruction point, and j is 1 to N; 34: Construct the feature vector X=[E ₁ ,E ₂ ,…,E _N ], and Normalize it to obtain the fault feature vector of the analog circuit. 4. a kind of fault diagnosis method based on improved analog circuit according to claim 1, is characterized in that: the improved DAGSVM method described in step 4, concrete steps are as follows: 51: by D _ij =||m _i -m _j || ² -r _i -r _j Calculate the inter-class distance of each two types of faults, where m _i represents the center point of the features extracted from the i-th class, and m _j represents the center point of the features extracted from the j-th class , r _i represents the radius of class i, and r _j represents the radius of class j; 52: Arrange the inter-class distances from large to small, and take the PSOSVM classifier with the largest inter-class distance D _ij as the top node classifier; 53: if the root If the node classification result is class i, select the SVM classifier with the largest distance between class i and class i as the child node of this layer; if the classification result neither belongs to class i nor class j, then exclude these two classes, Select the two SVM classifiers with the largest inter-class distance D _ij in the class as the top node classifier and continue the above two steps until the diagnosis result is obtained.