A Novel Junction Temperature Estimation Method Independent of Bond Wire Degradation for IGBT

Abstract

The insulated-gate bipolar transistor (IGBT) junction temperature is crucial for condition monitoring, reliability assessment, and health management. However, the existing IGBT temperature monitoring methods are affected by the aging of IGBT bond wires. A novel junction temperature estimation method independent of bond wire degradation for IGBT is proposed in this paper. On-state voltage drop and turn-on gate voltage overshoot are analyzed for dependence on the junction temperature and bond wire degradation. These two temperature and bond wire monitoring electrical parameters are combined by the Artificial Neural Network (ANN), to eliminate the effect of bond wire degradation. A double pulse test platform is built to verify the influence of temperature and bond wire degradation on on-state voltage drop and turn-on gate voltage overshoot, which agrees with the theoretical analysis. Lots of experimental data at various collector-emitter voltages, load currents, IGBT junction temperature, and bond wire degradation are tested. A BP neural network is trained for junction temperature monitoring and its optimal number of hidden layers is studied. The experimental results show that the junction temperature estimation method based on the neural network can effectively eliminate the effect of bond wire degradation. The estimation error is around 4.219°C. The proposed method has the advantages of high accuracy and a wide application range for junction temperature monitoring.