Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
411566 | Neurocomputing | 2016 | 11 Pages |
Due to the importance of induction motors’ continuous operation, early detection of faults has become a major trend. As reported in an IEEE study, bearing failures include more than half of mechanical faults. To detect existence of this fault, methods such as (short-time) Fourier, (continuous–discrete) wavelet, and Park transforms introduced. Static modeling of fault behavior is determined to be the major deficiency of above-mentioned methods. In other words, using conventional detection techniques, fault is assumed to have deterministic behavior, in which the fault frequencies are constant. As a matter of fact, fault characteristics can be affected under loading or environmental conditions, which makes conventional standing invalid. Authors of this paper have developed their previously introduced technique, frequency-domain discrete wavelet transform (FD-DWT) into a stochastic model. This makes the detection process valid for more variety of fault conditions and leads to earlier detection of fault and less damage to motor compared to other strategies.