Learning the shape: streamlining data needs in 2D irregular contour parameterization

This presentation explores the intersection of machine learning (ML) and scientific discovery, particularly its role in solving complex engineering problems. While ML has proven successful in areas such as drug discovery and atmospheric predictions, its utility is context-dependent and must be justified based on the complexity of the problem and its data requirements. In this work, we present a proof of concept demonstrating the application of ML to optimize heat transfer in microscale geometries, one of the most computationally expensive problems due to the detailed meshing requirements. By applying data-driven approaches, we show how ML can streamline costly computations in thermal analysis without compromising precision. This methodology, developed for microfinned base geometries similar to those used in EV battery pack cooling systems, holds potential for broader applications in optimizing irregular geometries in engineering.