Machine-Learning-Based Diagnosis of an Inverter-Fed Induction Motor

The principal objective of this paper is to detect and automatically monitor switch open-circuit faults in a two-level three-phase voltage source inverter fed induction motor from the processing of its current signals. The proposed diagnostic method uses both signal processing techniques and machine learning techniques in order to detect and localize the switch under an open-circuit fault. First, the Hilbert-Huang transform using the empirical ensemble mode decomposition is employed for each phase current signal, which leads to extracting the intrinsic mode functions. In order to optimally choose the function indicating the open-circuit fault harmonic, two factors, namely, the root mean square and the correlation coefficient are calculated out for each function. In this regard, two criteria are proposed that lead to choose the optimal function giving better information about the defected phase. The spectral envelope of the optimal function permits extra0cting the fault harmonic of the switch. Second, different machine learning techniques are applied to locate and classify the switch open-circuit faults with the hyper-parameters optimization for a better design of the different models. Finally, a comparative study of the different machine learning techniques is carried out for determining the best classifier for the open-circuit faults. The experimental results effectively demonstrate a very high classification rate of 98.98%.