A Parallel Multi-Classifiers System for Mechanical Fault Recognition

This paper addresses the problem of detection and gnosis mechanical faults using vibration analysis. In this work, a novel multiple classifier fusion system is proposed. The diagnosis process is performed in two steps. Firstly, the acquired signals are classified using a neural network in parallel with a genetic fuzzy classifier. And then, the outputs of the two classifiers are combined and fed into another neural network to produce a final decision. The proposed scheme is tested on real signals of different mechanical faults. The experimental results show that the system outperforms individual classifiers and able to achieve classification success rate of about 97.6%.