Novel GP-Based Gate Voltage V GE Unsupervised Precise Monitoring Strategy for IGBT Module Over High-Speed Switching Transients

This article proposes a novel Gaussian process (GP)-based gate voltageVGEunsupervised precise monitoring strategy for insulate-gate bipolar transistor (IGBT) modules in high-speed switching transients. First, a GP with constrained variational inference model is proposed to reconstruct low sample rateVGEback to extreme high sample rate in real time without pretraining. Second, aVGE-specific physics-guided hybrid kernel function is designed to accurately capture the features ofVGEwaveforms. Unlike traditional compressed-sensing-based monitoring method, the proposed method is not only applicable to signals with significant sparsity but also more suitable forVGE, and it avoids the severe distortion of theVGEduring sudden switching changes. Finally, real experiments are conducted on an IGBT double pulse test platform under various operating conditions, including different bus voltages, junction temperatures, load currents, and gate resistances, which validate the robustness and superiority of this technology. The method applicability and impact of multiple potential factors such as loop stray inductance are also analyzed.