Non-intrusive reduced order modeling for a coupled electro-thermo-mechanical problem

  • Schuler, Louis (Laboratoire de Mécanique Paris-Saclay)
  • Chamoin, Ludovic (Laboratoire de Mécanique Paris-Saclay)
  • Khatir, Zoubir (Université Gustave Eiffel, SATIE)
  • Berkani, Mounira (Universté Paris EST Créteil, SATIE)
  • Ouhab, Merouane (Mitsubishi Electric R&D Centre Europe)

Please login to view abstract download link

Power electronic modules are devices converting electrical power and are critical components of numerous electrical systems. A power module is made of an assembly of components to achieve its electrical function, heat dissipation, and mechanical resistance. The main components are the semiconductor chips, connected to other components by wire bonds. The high electrical current passing through the chips creates a dissipated heat power by Joule’s effect, leading to temper- ature elevation. The variation of coefficients of thermal expansion of the materials used in the assembly creates thermo-mechanical stresses. Those stresses can initiate and propagate cracks, mainly at the interface between wire bonds and chip metallization. The electrical, thermal, and mechanical behaviors are coupled, as material properties are temperature dependent. Further- more, electrical and thermal resistances increase when cracks propagate, coupling mechanical and electro-thermal behaviors. In this work, we develop a reduced order model based on the Proper Generalized Decomposition (PGD) for modeling power electronic modules. The PGD is an invasive method requiring the computation of equivalent operators. We investigate the implementation of the reduced model in a non-intrusive manner with Ansys. Multiple material and design parameters are considered. First, we propose a strategy to decouple the mechanical problem from the electro- thermal problem. Then, the PGD is applied to efficiently solve the coupled electro-thermal problem and the nonlinear mechanical problem in which the crack propagation is modeled with a cohesive zone model. Once the reduced order model has been obtained in an offline stage, the model can be used in an online stage for uncertainty quantification studies. Numerical results will be presented during the conference, showing performance in offline and online stages.