Identifying misalignments and coupling unbalances in soft-started induction motors applying persistence spectrum and CNN to stray-flux signals

This work proposes a methodology to automatically detect and classify coupling unbalances and misalignments in soft-started induction motors. This method relies on the analysis of the stray-flux signals during the time of operation of the soft-starter, thus analyzing the startup transient and part of the steady-state regime. In this study, two levels both of misalignment and coupling unbalance are forced in the laboratory, obtaining experimental results. Once the stray-flux signals are captured with a coil-based sensor, their envelopes are obtained applying the Hilbert transform (HT). Then, their persistence spectra are computed and used as input in a Convolutional Neural Network (CNN), which automatically classifies the aforementioned faults. The accuracy rate for the classification achieves 96.3%. Additionally, the results when analyzing the same signals with the Short Time Fourier Transform (STFT) are shown, reaching an accuracy rate of 66.2%. That confirms the potential of the proposed methodology compared to others.