Physics-Informed Machine Learning for Surrogate Modeling of Heat Transfer Phenomena

Tomoyuki Suzuki, Kenji Hirohata, Yasutaka Ito, Takehiro Hato, Akira Kano
J. Comput. Nonlinear Dynam. November 2023, 18(11): 111001. 
https://doi.org/10.1115/1.4063224

This paper proposes a sparse modeling method for automatically creating a surrogate model for nonlinear time-variant systems from a small number of time-series data. The proposed method is an improvement over a method for sparse identification of nonlinear dynamical systems first proposed in 2016, for application to temperature prediction simulations. The form of the thermal model is constrained by the physical model, and we use three novel machine-learning methods to efficiently estimate the model parameters. We verify the proposed method’s effectiveness using time-series data obtained by thermo-fluid analysis of a power module mounted on a comb-shaped heat sink. The proposed method has potential applications in a wide range of fields where the concept of equivalent circuits is applicable. Because the proposed method requires few data, has high extrapolation accuracy, and is easily interpreted, we expect that design parameters can be fine-tuned and actual loads considered, and that condition-based maintenance can be realized through real-time simulations.

Full Paper