Application of PMSM fault detector based on Kohonen classifier and FEM model

Abstract

Permanent magnet synchronous motors (PMSMs) are increasingly used in industrial applications. However, during the normal operation of the PMSMs, various types of damage may occur. Due to their use in many applications the detection of these damages at their initial stage constitutes an extremely important issue. Moreover, the development of the diagnostic systems in most cases means the necessity of interference in motor construction. This fact speaks for the use of mathematical models based on finite element methods (FEM) during the implementation of diagnostic tools. This article investigates the possibility of the PMSM faults classification using self-organizing Kohonen maps (SOM) based on the training with data from FEM model of PMSM. The article aims to show the possibility of detecting the PMSM faults as well as assess the type of damages: permanent magnet (PM) fault, inter-turn short-circuit (ITSC), simultaneous ITSC and PM fault. The self-organizing Kohonen neural network was trained using only the fault symptoms coming from the FEM model. Whereas the experimental verification was carried out on the real object during changing operating conditions of the drive. The results of the experimental research carried out on a specially designed PMSM show the impressive capability of the developed SOM-based neuronal classifier in the diagnostic tasks.