A method for identifying the health state of IGBTs based on CEEMDAN and ADP

Insulated Gate Bipolar Transistors (IGBTs) are core components in electric energy conversion systems, and monitoring their health is essential for ensuring reliable operation. Turn-off stress wave signals emitted by IGBTs can be detected using acoustic emission (AE) sensors, offering a non-invasive means to assess device condition. In this study, we present a comprehensive framework that integrates complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), asymmetric dot pattern (ADP) conversion, and a residual network (ResNet) for accurate health-state identification.First, stress wave signals are decomposed into intrinsic mode functions (IMFs) via CEEMDAN. Next, IMFs with the highest energy contributions and the most distinct correlation differences are selected to emphasize fault-relevant features. Selected IMFs are then transformed into two-dimensional ADP images that capture signal dynamics and subtle variations across different health states. Finally, a ResNet classifier is trained on these images to categorize multiple IGBT health conditions.Experimental results on benchmark datasets demonstrate that our method achieves over 98 % classification accuracy—significantly outperforming traditional feature-based and shallow learning approaches. Moreover, it exhibits strong noise robustness and computational efficiency, making it suitable for real-time monitoring. This work offers a new non-destructive detection scheme for IGBT health assessment with promising industrial applications.