Intelligent Junction Temperature Estimation of an IGBT Using Machine Learning and Vce Measurement

The reliability of semiconductors in power converters depends mainly on the junction temperature (T-j), and it is important to monitor T-j for reliable operation. This paper proposes a predictive model for the T-j estimation of an IGBT using advanced machine learning (ML) techniques, incorporating on-stage voltage drop (V-ce), which is one of the temperature-sensitive electrical parameters (TSEPs). Datasets collected from experiments are refined through extrapolation methods, and then a more precise T-j estimation is performed through ML, based on deep neural networks (DNNs). This approach is effective in handling problems caused by non-linearity, such as inflection points while estimating T-j. In addition, a novel ensemble learning model is proposed to enhance estimation accuracy, and comprehensive analysis and comparison of ML algorithms are provided.