Generative Constructal Design of a Multiphysics Heat Sink for Managing Transient Thermal Loads

Abstract In this article, we automate the constructal design method to design a multiphysics heat-sink with phase-change material for transient thermal loads. By combining high-conductivity material (HCM) and phase change material (PCM), we enable fast thermal transfer and stabilization during power spikes. The design domain is anisotropic, and its behavior is captured using finite elements with a variable-viscosity model that interpolates thermal properties across HCM and PCM phases. The system uses Navier–Stokes equation with the Boussinesq approximation, integrated in Python via FEniCS, to simulate natural convection within the PCM. This high-fidelity model drives a generative design agent that optimizes both conduction and free convection pathways. The HCM layout is generated by a variational auto-encoder (VAE), trained on synthetic and bio-inspired geometries evolved for diverse flow systems. Transfer learning adapts the genetic content from conductive systems, refining it to meet the specific needs of this multiphysics problem. This approach, rooted in constructal theory, showcases the adaptive power of generative methods for thermal design. It demonstrates how combining digital information flows with multiphysics models can unlock highly efficient, evolutionary solutions to complex problems. The results reinforce the necessity of fine-tuned, multiphysics formulations in the design of high-performance thermal systems. The methodology can be readily extended to other multiphysics applications to automate evolutionary design.