Research on the Method of Diagnosing Fault and Locating Fault Sources Using Neural Network

There are situations of many results and many reasons in the fault diagnosis. The training method of former neural network is supervised by teachers only could forecast or diagnose what kind of failure has occurred to the failure that possibly occurs. As a result of equipment´s complexity, it is difficult to find the source of fault accurately and rapidly in practical application even if we know the fault belong to which kind by training to the network. From the angle of constituting closed path, this paper proposes that both the input of the former network and output together as the new input parameter. It renews a new network together by the fault source as the output and the original fault diagnosis network and under the front foundation uses the improved LMBP algorithm and the network trim´s method to realize the network architecture optimization and the enhancement of training speed. Finally, we suppose a group of samples carried on the training simulation combining the situations of many results and reasons in the fault diagnosis. The simulation result indicated that the network based on the closed path thought which made the former input and output together as the new input parameter had simultaneously two functions which contain the fault predication, forecast and the diagnoses of fault source. It carried on the optimization to the network architecture, reduced the complexity of computation and also raised the algorithm convergence rate. We also saw this thinking enhances diagnosis ability to the fault greatly.