Analog Circuit Fault Diagnosis Based On Neural Networks

With the development of modern industry, especially the rapid development of electronic technology, the complexity of the electronic system is higher and higher, and the degree of automation becomes more and more advanced. The proportion of digital circuit in electronic system is bigger and bigger, but analog circuit part in the electronic system is irreplaceable, and the faults of analog circuit are more than80%, analog circuit fault diagnosis is a difficulty in the development of circuit industry.Traditional methods of analog circuit fault diagnosis have more finiteness, the development of artificial intelligence technology provides new ideas and methods to analog circuit fault diagnosis, especially the development of neural network is a big boost for development of analog circuit fault diagnosis. This paper is mainly on neural network which applied to analog circuit fault diagnosis, combined with some new methods and introduced methods of extracting fault features, pre-processing method of fault information. Typical BP neural networks is taken as an example, achieving fault diagnosis to adopt the method of genetic algorithms(GA) optimizing neural networks’weights and bias. There’s main work concentrates on some aspects as follows:(1) Use the circuit simulation analysis software ORCAD completes schematic diagram and use PSPICE to make a Monte-Carlo analysis, obtain fault information of every fault pattern, training and testing sample set.(2) Conduct sensitivity analysis and find out components which have much influence on circuit properties, establish a fault dictionary based on analysis result.(3) Extract fault features from a mount of fault information, this paper adopt two methods, the one is wavelet analysis which is used more widely, the second is statistical method—factor analysis, two different methods extract fault features separately, pre-processing fault information, then as new training and testing sample set to import neural network, and last, we compare the two methods.(4) Choose BP neural networks, use genetic algorithms optimize the values of weights and bias, adopt fault features vectors which two different methods extracted training and simulate net. We can obtain the result:GA-BP is better than the values of weights and bias not optimized by genetic algorithms.(5) The neural networks which is trained by wavelet analysis extracting fault features is better on fault diagnosis, factor analysis is relatively poor, but statistical method applied on analog circuit fault diagnosis can be approved better and achieve the goal as expected. We also have the conclusion of wavelet analysis extracting fault features is more superior.