AN IMPROVED NEURAL-NETWORK REALIZATION FOR RELIABILITY-ANALYSIS

In this paper we present an improved neural network training algorithm and architecture for reliability analysis of a simplex system and a T MR system which includes the effects of permanent fault and intermitte nt fault. A fully-connected three-layer neural network represents a di screte-time ii-state reliability Markov model of a fault-tolerant syst em. The desired reliability of the system is fed into the neural netwo rk, and when the neural network converges, the design parameters are r etrieved from the weights of the neural network. Finally, the simulati on results show that the proposed method converges faster than other m ethods, especially in the case of the state number oz the Markov model , which increases. This technique is also suitable for any system. (C) 1998 Elsevier Science Ltd. All rights reserved.