关键词: 航空发动机, 环境沉积物, 机器学习, CMAS

Abstract: High-temperature thermal barrier/environmental barrier coatings serve as critical protective materials for hot-end components in aero-engines. However, during service, they are susceptible to corrosion damage from environmental deposits (CMAS, primarily composed of sand, dust, and volcanic ash) at elevated temperatures, posing a severe threat to engine operational safety. Given that the physicochemical properties of CMAS are highly dependent on geographical origin, corrosion protection models developed for specific environments have limited applicability across China's complex topography. However, systematic research on the ambient/high-temperature characteristics of CMAS across the entire nation remains lacking. To address this gap, this study collected 90 CMAS samples from representative inland regions of China. It systematically characterised both ambient properties (including chemical composition, particle size distribution, relative crystalline content, and relative adsorbed water content) and high-temperature properties (including melting temperature and high-temperature viscosity). Results indicate significant differences in the normal/high-temperature properties of sand-gravel, dust, and volcanic ash. Sand-gravel and dust exhibit strong regional homogeneity, whereas volcanic ash shows pronounced heterogeneity. A machine learning predictive model for CMAS high-temperature viscosity was developed, with SVM, MLP, and GBDT models demonstrating significantly higher prediction accuracy (R2 ≥ 0.97) than traditional methods (GRD and Factsage). SHAP analysis revealed that sand/gravel CMAS viscosity is governed by the multi-factor coupling of chemical composition and structural characteristics. This study established China's first systematic CMAS property database. The proposed machine learning approach provides a novel pathway for high-temperature viscosity prediction, offering critical data and theoretical support for the design of CMAS-resistant coatings.

Key words: aeroengine,, environmental deposits,, machine learning,, CMAS