Linear discriminant analysis for an automatic detection of stator faults in induction motor drives

This paper presents a method for an automatic detection of a stator short circuit and for the estimation of its severity in inverter-fed induction motors, in different conditions of speed and load. The method is based on the Linear Discriminant Analysis and on the spectral analysis of three signals (current of one phase and external stray flux measured with two different sensors). In order to obtain a relatively simple instrument for applications in industrial cases, only the first odd harmonics multiple of the fundamental have been chosen as features for the classifiers. All the considered signals have provided interesting diagnostic results, even if the best performance is given by the current.