Aging of power module (PM) used in power electronics applications changes the electrical and thermal parameters over the lifetime; thereby increasing the power losses and thermal impedance, and influencing the thermal model (TM) accuracy. This paper deals with PM thermal digital twin (TDT) identification from the data acquired online during PM mission. Since TM contains both slow and fast time constants, the conventional ordinary least squares (LS) minimizing one step error prediction has poor performance in identification of TDT parameters (TDTP). The proposed algorithm is an extension of LS. The ability of proposed algorithm to find optimal parameters is enhanced by adding additional terms penalizing quadratic error of multi-step prediction. The method enables tracking TM parameters through PM lifetime which enables condition monitoring and accurate die temperature estimation.
