Expectation-Maximization Algorithm for Identification of Mesh-Based Compartment Thermal Model of Power Modules

Abstract

Accurate prediction of temperatures in power modules is crucial for proper thermal management. Lumped parameter thermal models are preferred in this application for their low computational cost. The estimation procedure of the parameters of these models requires measurements of temperatures of all active elements. This requirement is relaxed in this contribution. Specifically, the previously used dark gray-box compartment model is replaced by a structured compartment model utilizing a mesh-based discretization of the physical layout of the module. Compartments are categorized into several types with common parameters for each type. The parameters are identified from the data of the measured elements using the Expectation-Maximization algorithm. The algorithm internally predicts the temperatures of the unmeasured elements. The sensitivity of the estimation to regularization of the process covariance matrix is also studied. The implied high-dimensionality of the state-space increases the computational cost of the conventional estimation procedure, therefore, a simplified procedure with a much lower computation cost is proposed. The performance of the proposed approach is tested on simulated data.