Predicting the service life of IGBT modules is an effective way to assess their health status and reliability. Based on the IGBT aging experiment measurement, this paper constructs a two-dimensional IGBT state detection index including a saturation voltage drops (VCEs) and junction temperature. For the normalized data, this paper introduces a way of segmented processing method to remove the large index fluctuation caused by the break of the IGBT bond wire. Based on VCEs and junction temperature data, the prediction model of neural network-based IGBT module remaining service life is proposed in this paper. The accuracy of the proposed neural network model in the prediction of remaining life is analyzed for the same sample with different channels and different experimental conditions.
