Conditional generative adversarial network driven approach for direct prediction of thermal stress based on two-phase material SEM images

Abstract

A conditional generative adversarial network (cGAN)-driven approach for the direct prediction of thermal stress is proposed. Synthetic two-phase structure images of ceramic top coat (TC) with CaO–MgO–Al2O3–SiO2 (CMAS) inclusions are established, and the TC matrix and CMAS inclusions are semantically segmented by grayscale. The thermal stresses of the two-phase structure are calculated using image-restoration finite element models (FEMs) under the isothermal process. The training database is established based on a small-scale original dataset of integral structures and their stresses. Each integral TC–CMAS structure and its corresponding stress distribution are partitioned into many local images, using which the cGAN model is trained. The model stresses are generated by the trained cGAN directly from the structure images. The deviations between the predicted stress and the FEM stress are small in most areas of the images. In terms of computing time, the proposed approach has higher stress evaluation efficiency than does the FEM.