Ultrasonic characterization of thermal barrier coatings porosity through BP neural network optimizing Gaussian process regression algorithm.

Porosity is an integral part of thermal barrier coatings (TBCs) and is required to provide thermal insulation and to accommodate operational thermal stresses. Accurate characterization of the TBCs porosity is difficult due to the complex pore morphology and ultra-thin coating thickness. In this paper, a BP neural network optimizing Gaussian process regression (GPR) algorithm, termed BP-GPR, is presented to characterize the TBCs porosity based on a constructed ultrasonic reflection coefficient amplitude spectrum (URCAS). The characteristic parameters of URCAS are optimized through the BP neural network combined with a high determination coefficient R2 rule. Then the optimized parameters are utilized to train the GPR algorithm for predicting the unknown TBCs porosity. The proposed BP-GPR method was demonstrated through a series of finite element method (FEM) simulations, which were implemented on random pore models (RPMs) of plasma spraying ZrO2 coating with a thickness of 300 μm and porosities of 1%, 3%, 5%, 7%, and 9%. Simulation results indicated the relative errors of the predicted porosity of RPMs were 6.37%, 7.62%, 1.07%, and 1.07%, respectively, which has 32% and 48% accuracy higher than that predicted only by BP neural network or GPR algorithm. It is verified that the proposed BP-GPR method can accurately characterize the porosity of TBCs with complex pore morphology.