Modeling the viscosity and infiltration kinetics of silicate melt -thermal barrier coating systems based on deep-learning and experimental approach

Spallation degradation of thermal barrier coatings (TBCs) caused by the infiltration of silicate melts (SMs) deteriorates the safety of aviation turbines. The SMs viscosities , which determine infiltration speed and thus the degradation of TBC, are highly dependent on SMs compositions and temperature gradients within the coatings. In this study, we developed machine learning (ML) models to accurately predict the viscosity of SMs. The training dataset comprised 7188 collected experimental data and 60,000 high-quality synthetic data generated by CTGAN. Our predictors achieved superior accuracy (R 2 > 0.97) compared to previous models. Using these models and extensive datasets, we analyzed the differences between natural silicates and the synthetic silicate melt analogs named CMAS and explored the impact of composition on viscosity through interpretability techniques and first-principles calculations. Furthermore, series of infiltration experiments were conducted to quantitatively evaluate the effects of viscosity and temperature gradients on SMs infiltration kinetics in TBCs. The complete data-stream from silicate composition to infiltration kinetics was modeled in this work.