Functionally Graded Structure Design for Heat Conduction Problems using Machine Learning

Abstract

This study introduces a topology optimization method for the simultaneous design of macro-scale structural configuration and unit structure variation to ensure effective heat conduction. Shape changes in the unit structure depending on its location within the macro-scale structure result in microas well as macro-scale design and enable better performance than using isotropic unit structures. They result in functionally graded composite structures combining both configurations. The representative volume element (RVE) method is applied to obtain various thermal conductivity properties of the multi-material based unit structure according to its shape change. Based on the RVE analysis results, the material properties of the unit structure having a certain shape can be derived using machine learning. Macro-scale topology optimization is performed using the traditional solid isotropic material with penalization method, while the unit structures composing the macro-structure can have various shapes to improve the heat conduction performance according to the simultaneous optimization process. Numerical examples of the thermal compliance minimization issue are provided to verify the effectiveness of the proposed method.