Application of artificial neural network and OpenCL in spectral and wavelet analysis of phase current of LSPMS machine

Abstract

The paper presents a parallel computing algorithm with its implementation in software for diagnostic of line start permanent magnet synchronous motor (LSPMSM). The software based on the developed algorithm, allows for analysis using a discrete Fourier transform (DFT) or a discrete wavelet transform (DWT). The elaborated software was tested using the phase current of the LSPMSM. In the case of wavelet analysis, the input signal refers to start-up of the motor supplied with symmetrical voltage, without external load, while steady-state waveforms were used for the DFT analysis. Moreover, the mentioned software has an implemented multi-layer perceptron neural network which can be used as decision element of the diagnostic system. In addition, the article brought closer the issues related to the structure and learning algorithms of artificial neural networks and OpenCL framework.