Machine Learning Aided Design and Optimization of Conformal Porous Structures

Abstract

The porous cooling system has been proved to have significant advantages over traditional 2D conformal cooling channels due to its rapid cooling performance during the injection molding process. Compared to conventional porous systems, the conformal porous structures (CPS) have been proven to have even more uniform cooling performance and a reduced temperature variance of the part. For the part with unevenly distributed thickness values however, the temperature variance problem remains unsolved. In addition, there is a lack of modeling and optimization efforts on developing an optimal CPS structure with varying cooling cell sizes to achieve better cooling performances. To solve this problem, a machine learning approach is applied to predict the part surface temperature based on identified CPS design parameters. With this surrogate temperature prediction model, the optimization is performed to generate a machine learning aided design of CPS. The simulation results of a swimming pedal case study indicate that the machine learning aided CPS is able to achieve a 76% reduction in temperature variance compared to conventional CPS.