A modular model-order reduction approach for the solution of parametrized strongly-coupled thermo-mechanical problems

This paper deals with the simulation of parametrized strongly-coupled multiphysics problems. The proposed method is based on previous works on multiphysics problems using the LATIN algorithm and the Proper Generalized Decomposition (PGD). Unlike conventional partitioning approaches, the LATIN-PGD solver applied to multiphysics problems builds the coupled solution by successively adding global corrections to each physics within an iterative procedure. The reduced-order bases for the different physics are built independently through a greedy algorithm, ensuring accuracy up to the desired level. This flexibility is used herein to efficiently handle parametrized problems, as it allows to enrich the bases independently along the variations of the parameters. The proposed approach is exemplified on several three-dimensional numerical examples in the case of thermo-mechanical coupling. We use a standard monolithic scheme to validate its accuracy. Our results highlight the adaptability of the proposed strategy to the coupling strength. Concerning the parametrized aspects, the method’s capability is illustrated through parametric studies with uncertain material parameters, resulting in significant performance gains over the monolithic scheme. Our observations suggest that the proposed computational strategy is effective and versatile when dealing with strongly-coupled multiphysics problems.