Ultrasonic prediction of thermal barrier coating porosity through multiscale-characteristic-based Gaussian process regression algorithm

Abstract

Excellent performance of thermal barrier coatings (TBCs) significantly depends on their internal pore microstructure. Accurate characterization of TBCs porosity is difficult through conventional ultrasonic techniques due to the complex pore morphology. In this paper, a multiscale-characteristic-based Gaussian process regression algorithm, termed M-GPR, is presented to predict the TBCs porosity based on ultrasonic pulse-echo technology. The acoustical characteristics containing porosity information are decoupled through Morlet wavelet decomposition combined with a constructed ultrasonic pressure reflection amplitude spectrum. The extracted acoustical characteristics and porosities are utilized to train the M-GPR algorithm optimized by genetic algorithm and cross-validation for predicting the unknown TBCs porosity. A random pore model (RPM) revealing the complex microstructure of TBCs is developed. The proposed M-GPR method is demonstrated through a series of finite element simulations, which are implemented on the RPMs of a plasma sprayed ZrO2 coating with porosities of 1%, 3%, 5%, 7%, and 9%. Simulation results indicated that the relative errors of the predicted porosity of ZrO2 specimens are 0.69%, 4.44%, 2.78%, 0.60%, and 6.28%, respectively, which has 16% accuracy higher than that predicted by BP neural network. It is verified that the proposed M-GPR algorithm can accurately estimate the porosity of TBCs with complex pore morphology.