Fast stress evaluation of the top coat of thermal barrier coatings under CaO–MgO–Al2O3–SiO2 penetration based on image recognition and an artificial neural network

Abstract

An image recognition and artificial neural network (ANN)-based method for fast stress evaluation of thermal barrier coatings (TBCs) is proposed. Random two-phase images of ceramic top coats (TCs) with penetrated CaO–MgO–Al2O3–SiO2 (CMAS) are established, and the TC–CMAS image boundaries are recognized. Image-restoration finite element models are constructed based on boundaries optimized using the average coordinate method. The isothermal thermal stress in the restoration models is calculated. The structural parameters and local maximum stress are extracted to establish sample databases on which the ANN is trained. The thermal stress states in a new group of samples are predicted by the trained ANN. The root-mean-square error between the stress calculated by the new stress evaluation method and those of results calculated using the finite element method is approximately 6%. The new method also has a higher calculation efficiency and its stress evaluation speed is accelerated by a factor of approximately 77 compared with that of the finite element method.