Wavelet Scattering Transform Based Induction Motor Current Signature Analysisa

Abstract

In this paper wavelet scattering transform (WST) algorithm is proposed to analyze motor current signatures and extract faults features. The algorithm overcomes the short comings of the classical techniques, which are noise sensitivity and computational complexity in features extraction. The discrete cosine transform (DCT) is also used to analyze motor current and compared with WST. The DCT is found to simplify the computational complexity without improving the detection accuracy. The WST is found to have better detection accuracy than other common fault detection algorithms. Moreover, the WST has the advantage of separating the envelope frequency in the broken bar case. The separation of the envelope frequency could be achieved in the first or second level of scattering. The data were obtained from an experimental set up consisting of an induction motor fed by a SiC-based inverter and coupled with a dc generator as a load. The dc generator was controlled by a commercial ABB-DCS800 motor drive, which is able to control torque pulsations to emulate faults.