AI-assisted vibration-based fault diagnosis of IGBT modules in electric vehicles

Early detection of structural degradation in insulated gate bipolar transistor (IGBT) power electronic modules (PEMs) is important for improving the reliability of electric vehicles operating under dynamic road loading. This study presents a physics-informed, AI-assisted vibration-based diagnostic framework for PEM condition assessment under realistic excitation conditions. A three-dimensional finite element model was used to identify resonance-sensitive frequency regions, with the dominant dynamic amplification observed near 1345 Hz. Acceleration responses generated under ISO 8608 Class K road excitation were processed to extract time-domain and frequency-domain features, including root mean square (RMS), kurtosis, and band power around the dominant resonance. These features were then used to classify healthy, mildly degraded, and severely degraded solder interconnect conditions using supervised machine learning models, namely support vector machine (SVM), k-nearest neighbors (kNN), and Random Forest. Among the evaluated models, Random Forest achieved the highest classification accuracy of 98.1 %. The novelty of the study lies in treating the 1345 Hz structural resonance not only as a mechanical response characteristic but also as a diagnostic-sensitive frequency band for feature engineering and fault discrimination. The results demonstrate the feasibility of resonance-informed vibration diagnostics for simulation-based early-stage condition assessment of PEM structures and provide a foundation for future experimental validation and embedded monitoring applications.