Effectiveness Analysis of Rolling Bearing Fault Detectors Based On Self-Organising Kohonen Neural Network – A Case Study of PMSM Drive

Abstract

Due to their many advantages, permanent magnet synchronous motors (PMSMs) are increasingly used in not only industrial drive systems but also electric and hybrid vehicle drives, aviation and other applications. Unfortunately, PMSMs are not free from damage that occurs during their operation. It is assumed that about 40% of the damage that occurs is related to rolling bearing damage. This article focuses on the use of Kohonen neural network (KNN) for rolling bearing damage detection in a PMSM drive system. The symptoms from the fast Fourier transform (FFT) and Envelope (ENV) Analysis of the mechanical vibration acceleration signal were analysed. The signal ENV was obtained by applying the Hilbert transform (HT). Two neural network functions are discussed: a detector and a classifier. The detector detected the damage and the classifier determined the type of damage to the rolling bearing (undamaged bearing, damaged rolling element, outer or inner race). The effectiveness of the analysed networks from the point of view of the applied signal processing method, map size, type of neighbourhood radius, distance function and the influence of input data normalisation are presented. The results are presented in the form of a confusion matrix, together with 2D and 3D maps of active neurons.

Keywords

• pmsm
• rolling bearings
• electric drive diagnostics
• self-organising maps
• shallow neural network