Press-pack Insulated Gate Bipolar Transistors (IGBTs), valued for their high input impedance and substantial current-carrying capacity, serve as core high-power devices in flexible DC transmission systems. To fulfill high-power application demands, multi-chip parallel packaging configurations are typically employed. However, coupled electro-thermal-mechanical multi-physics effects induce non-uniform parameter distributions within the device, significantly compromising operational stability and service lifetime. Direct experimental assessment of internal states remains challenging, and while multi-physics coupled finite element simulations are commonly used for thermal-mechanical state evaluation, their low computational efficiency and convergence difficulties present limitations. To address these issues, this paper proposes a deep learning-based conditional generation framework. This approach establishes a direct mapping model from external excitation parameters (pressure, current) to IGBT chip surface pressure and temperature distributions, utilizing a feedforward neural network for pressure distribution prediction and a deep convolutional neural network for temperature distribution prediction. This approach achieves efficient and precise prediction of chip surface stress and temperature fields. By circumventing numerical solutions of complex partial differential equations, the method significantly enhances modeling efficiency and prediction accuracy, offering substantial theoretical significance and engineering value for optimizing reliability design strategies of Press-pack IGBT devices.
Li et al. (2026) studied this question.