A Neuro-Detector Based On The Cybernetic Concepts For Fault Detection In Electric Motors

Abstract

In this study, an auto-associative neural network (AANN) is designed as a fault detector using the cybernetic concepts. In this sense, an artificial neural network structure is connected with a finite state system or a finite automata and an AANN topology is described as a virtual detector. In terms of the practical application, vibration signals, which are taken from an induction motor of 5 HP for both the healthy and faulty motor cases, are considered in the spectral domain. The vibration signal presented in the healthy motor case is separated into 4 blocks and the spectral set of these blocks is used as input and target pattern sets during the training of the AANN. After the training process, a new vibration spectrum, which is defined in the faulty motor case is applied to this trained network and the faulty case is determined by the error variation at output nodes of the AANN. In this application, the error signal shows huge amplitudes between 2 and 4 kHz as an indicator of the bearing damage.