Neuro-WPT Based Diagnosis and Protection of Three-Phase IPM Motors

This paper presents the practical implementation of a novel fault diagnosis scheme for an interior permanent magnet (IPM) motor protection using the neural network (NN). The preprocessing of line currents of different faulted and normal unfaulted conditions of an IPM motor are carried out by the wavelet packet transform (WPT) in order to minimize the structure and the training time of the proposed neural network based protection technique. It has been shown that the wavelet packet transformed second level high frequency coefficients (dd²) of line currents are able to differentiate between the healthy and faulted conditions. These are used as the input sets of a three-layer feed-forward neural network. The scheme is evaluated and tested on-line on a laboratory 1-hp IPM motor using the ds1102 digital signal processor board. Three types of faults such as single line to ground (L-G) fault, line-to-line (L-L) fault, and single phasing fault are investigated. In all the tests carried out, the types of fault are classified and identified promptly and properly, and the tripping action is initiated within one cycle of the fault occurrence based on a 60 Hz system.