MCSKPCA based neural network fault diagnosis method for analog circuits

Abstract

Disclosed is an MCSKPCA based neural network fault diagnosis method for analog circuits, comprising acquiring the output voltage signal of an analog circuit to be diagnosed; performing wavelet transformation on the acquired output voltage signal; calculating the energy eigenvalues of the wavelet coefficients of the output voltage signal, obtained through the wavelet transformation; performing MCSKPCA feature extraction and dimensionality reduction on the energy eigenvalues, and obtaining an optical eigenvector; and sending the optical eigenvector to a BP neural network separator, and outputting a fault diagnosis result by the BP neural network separator. The method can be used for not only diagnosis of linear or nonlinear circuits and systems thereof, but also diagnosis of hard fault and soft fault in the linear or nonlinear circuits.

Description

technical field [0001] The invention belongs to the field of neural network and electronic circuit engineering, and relates to a fault diagnosis method for an analog circuit based on a neural network. MCSKPCA is Maximal Class Separability KPCA, which is based on the maximum class separation kernel principal component analysis. Background technique [0002] In analog circuits, faults can be divided into two categories: one is called hard fault, which refers to the open circuit and short circuit failure of components; the other is called soft fault, which means that the parameters of components exceed the predetermined tolerance range, generally they None of the equipment completely fails, for example, due to the aging of components, deterioration or changes in the use of environment, etc. caused by changes in component parameters. [0003] Since the testing and diagnosis of analog systems began to be studied in the 1960s, the progress has been relatively slow. The main reaso...

AI Technical Summary

Problems solved by technology

[0003] Since the testing and diagnosis of analog systems began to be studied in the 1960s, the progress has been relatively slow. The main reason is that the input excitation and output response of analog circuits are continuous quantities, and the parameters of each component in the network are usually continuous, that is, most faults The situation is a soft fault, so the fault model in the simulation system is more complicated, and it is difficult to make simple quantification

Since the fault parameters are also continuous, theoretically speaking, an analog component may have infinitely many faults, so testing and diagnosis are far more difficult than digital systems, so it is not yet fully mature in theory and methods, and can be paid There are few practical ones

[0004] In recent years, a lot of research has been done on the fault diagnosis of analog circuits from the system level, block level to chip level, and some methods directly intercept different test point signals as the input of the neural network classifier for fault diagnosis. The network scale is usually very large, the network training time is very long, and it is difficult to meet the real-time requirements

Some use the fault dictionary method to establish the DC or AC fault dictionary of the circuit, but this method is generally only suitable for hard faults, and the effect is not good for soft faults.

Some methods use wavelet transform to preprocess the signal, and then perform feature extraction of principal component analysis. Since only linear feature extraction and insufficient dimensionality reduction are implemented, the subsequent neural network cannot obtain effective feature data. Therefore, a relatively large number of input terminals is required, and a relatively low fault diagnosis rate is obtained

Images