Partially molten environmental dust spreading on an EB-PVD TBC-analog surface: experimental results enhanced by physics informed machine learning

Airborne particles such as sand, dust, and volcanic ash pose a severe threat to the reliability of thermal barrier coatings (TBCs) in jet engines. Upon ingestion into engines, these silicate-rich particles melt and form deposits that undergo complex spreading and adhesion dynamics on coating surfaces, ultimately leading to coating degradation. To simplify the reproducibility and comparability of experimental studies synthetic calcium–magnesium–alumino–silicate (CMAS) particles are commonly employed as simulants. Here, high-temperature deposition experiments have revealed that partially molten droplets (containing solid inclusions), exhibit heterogeneous spreading morphologies beyond classical viscous flow. A compound droplet model integrating heat transfer and spreading dynamics has been developed to describe quantitatively the deposition behavior of such partially molten particles. A physics-informed machine learning model, trained on 105 experimental cases was developed and has demonstrated superior predictive performance, with support vector regression achieving the highest accuracy (R² ≈ 0.71, RMSE ≈ 0.13). This combined experimental–computational strategy establishes a robust predictive paradigm linking natural threats (sand, dust, volcanic ash) with CMAS simulants, providing mechanistic insights into and mitigation strategies for CMAS-related degradation to enhance the operational reliability of aeroengines.