Artificial neural networks (ANN) have been shown to have the capability of approximating arbitrarily complicated continuous functions. Nevertheless, one of the drawbacks of using conventional ANN technology alone to solve engineering problems is its "black-box" characteristics. In most cases, engineers cannot interpret the network's decision-making process from a heuristic point of view; they can only know that the network gives a correct input-output mapping.

In this project, the fuzzy logic technology is incorporated with the artificial neural network technology to form a neural/fuzzy system to solve motor fault detection problem. Fuzzy logic provides a mechanism by which to perform problem solving with some expert knowledge. However, if this knowledge is only partially correct, then the fuzzy logic system will not perform optimally. On the other hand, if there exists a means to correct whatever erroneous assumptions were made, then the fuzzy logic system could perform much closer to an optimal level. By structuring an ANN in a fuzzy logic format, the network could be trained to classify fault data while modifying the initial fuzzy membership functions and fuzzy rules to more realistic expert values. In turn, this would provide more accurate expert knowledge about the problem at hand. Neural network training through backpropagation provides a weight adaptation which will fine tune these fuzzy functions. This effectively adapts the fuzzy membership functions and rules to reach a correct decision.