RUL Detection of IGBT Based on VMD-ROLLING-BI-GRU

With the rapid development of railway transportation, switch-mode power supplies (SMPS) in trains have become crucial for stable operation. However, IGBT modules in these SMPS often fail due to thermal and electrical stress, causing disruptions and increased maintenance costs. This paper proposes an RUL prediction method for IGBT modules in train SMPS using variational mode decomposition (VMD), rolling prediction, and bidirectional gated recurrent units (BI-GRU). Based on an accelerated aging dataset of IGBT modules under simulated train operating conditions, transient collector-emitter voltage spikes are selected as failure features. Experiments show that the proposed model achieves a mean absolute percentage error (MAPE) of 0.00375 and a coefficient of determination R2 of 0.952. The results demonstrate that the model can effectively predict IGBT module lifetimes, enhancing train SMPS reliability and reducing maintenance costs.