Enhanced diagnosis and automatic monitoring system for a faulty two-level voltage inverter-fed induction motor using a combined signal processing and machine learning

Abstract

This paper presents a methodology for diagnosing and monitoring open-circuit faults in IGBT switches within inverter–induction motor systems using current signal analysis. The method integrates Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) for signal processing and the Random Forest (RF) algorithm for machine learning-based fault detection and classification. The process begins with applying ICEEMDAN to decompose current signals into Intrinsic Mode Functions (IMFs) that may capture fault-related harmonics. Next, statistical correlation coefficient (Corr) analysis is conducted to identify which IMFs indicate open-circuit faults; IMFs with Corr values ≥ 0.1 (IMF4, IMF5, IMF6, and IMF7) are selected. These IMFs are used to reconstruct the signal, and spectral envelope analysis is performed on the reconstructed signal to extract fault signatures. The extracted features include key frequency components 𝑓𝑐 and 𝑓𝑜𝑐, and the phase angle 𝜑( 𝑓𝑜𝑐/ 𝑓𝑐 ) . These features are used to train the RF model for both detection and localization of faults. The RF algorithm achieves a fault detection accuracy of 99.98% and a classification accuracy of 99.99%. Experimental validation confirms the effectiveness of the proposed ICEEMDAN–RF approach, demonstrating a strong potential for accurate diagnosis and continuous monitoring of open-circuit IGBT faults in inverter–induction motor systems.