Research On Fault Diagnosis For Analog Circuits With Tolerance

Fault diagnosis in analog circuits is the third branch of network theory, and it is the mainstay in automatic test technology. Since 1960's, many valuable results have been obtained. But in general, the development is slow. The development of fault diagnosis theory is detained by universal existence of tolerance of components, variety of circuit faults, etc. Especially when soft faults exist, the circuit fault statuses appear close to normal conditions and the circuit statuses under different fault conditions overlap, hence the difficulty of analog circuit fault diagnosis.With the development of micro-electronics technology especially VLSI technology, automatic test and fault diagnosis technology for analog circuits are increasingly important and imminent. Therefore fault diagnosis for analog circuit with tolerance was studied in this paper, which emphasized on fault diagnosis method based on neural network and the solutions of a series of problems which include large scale of training samples, slow diagnostic speed and lower diagnostic accuracy due to the existence of tolerance and the variety of analog circuit faults.Against the shortcomings of Back-propagation Neural Network (BPNN), which include slow learning speed of convergence, the nature which is easy to fall into local minimum value and the difficulty of determining the structure of network, PNN based diagnostic method for analog circuit with tolerance was proposed. Simulation results showed this method had fast diagnostic speed, accurate diagnostic results and great recognition ability for soft faults.In order to solve the problem of seeking optimum training samples for neural network, genetic algorithm was employed to obtain the minimal reduction of decision tables by combining its outstanding ability for overall searching with rough set theory. This method is a foundation to the compositive diagnostic method which employs RS theory and neural network.According to the high degree of complexity of neural networks due to the large scale of training samples which are employed to have analog circuits diagnosed, this paper proposed a kind of fault diagnosis program which based on Self-organized Competition Network-Rough Set-BP Network. It makes the scale of neural network inputs decreasing and thus the training time of neural network is brought down.According to the uncertainty of fault diagnosis for analog circuit with tolerance, a method was given based on BP and RBF network combined with D-S evidence theory. Simulation results showed that neural network data fusion method improved the accuracy of fault diagnosis.