Neural network-based incipient fault diagnosis

摘要

This paper resents the neural network-based incipient fault diagnosis for a radwaste solar evaporation processing system. This incipient fault diagnosis is used to examine the causes of faults and abnormal conditions in the radwaste processing systems or nuclear power plants in which safety is the most important. A neural network is utilized clearly to classify faults and noise conditions, find out similar patterns from input patterns that are not in the training patterns, and rapidly process difficult problems using parallel architecture. The incipient fault diagnosis system developed can predict and diagnose fault conditions in the solar evaporation system. This system has functions of choosing 4 utilities (make-up pump, circulation pump, exhaust fan, and air monitor) to be considered quite important, constructing a neural network and classify the normal and fault conditions, training the network using the error backpropagation algorithm, and performing simulations for single and multi faults. The simulation results show that a neural network trained with normal and multi-fault patterns can achieve a higher degree of accuracy in diagnosis.