Fault Diagnosis Based on Rough Neural Network

Abstract

Considering training time of traditional BP neural network is too long and it can not solve the problem that input vector is multiple-valued, a new method based on rough BP neural network for fault diagnosis is presented. The approach is realized by applying PSO (particle swarm optimization) to discretize continuous attributes, using property of dependency of rough set to carry through attribute reduction and designing a kind of rough BP neural network according to the optimal decision system for fault diagnosis. A practical example is given to show the method is feasible and available.