Use of neuro-fuzzy system to time domain electronic circuits fault diagnosis

This paper presents a new concept to analog fault diagnosis. Problem of distinguishing between healthy or faulty analog circuit has always been very complicated. The most common approach based on pattern recognition, especially on mean square error measure, can not distinguish all faulty circuits from the healthy one. Normally, the dictionary has to include thousands of patterns and even then, the level of fault detection is not satisfactory. A neural network classifier has been proposed to solve the problem. Its generalization ability allows to reduce the dictionary size significantly. This paper shows how to create a neural dictionary for fault location. Moreover, at the first stage of classification, the fuzzy logic is utilized to transform a measurement vector into a zero-one range. The information from the circuit under test (CUT) has to be as high as it is possible but at the same time the stimuli has to be as simple as possible. The most common AC and DC tests don't give the best solution. Therefore, the time domain testing with pulse stimuli has been utilized. This paper presents a new concept to analog fault diagnosis