Generative Design of Heat Exchangers via Variable-fidelity Networks

Cornerstone Research Group (CRG) will develop an approach for computational design of multi-functional components featuring optimized cellular flow path geometries for enhanced heat exchange and structural performance which are readily producible using emerging advanced manufacturing techniques. This design capability will take the form of a generative methodology automating material layout throughout arbitrarily shaped design domains subject to mechanical, fluid flow, and thermal performance metrics to enable low mass utilization for high performance systems. Computation of analysis driving responses will be carried out by standing up a physics informed neural network trained using multiphysics simulations of variable fidelity. This machine learning model will be used as a surrogate to accelerate geometry revisions by using a genetic algorithm to adjust localized sizing parameters for optimal heat exchange while also meeting structural requirements. During Phase I, CRG will apply the implemented process to a selected design space for a cold plate heat exchanger followed by prototype fabrication and experimental evaluation for comparison with traditionally designed counterparts.