Surrogate model-assisted machine learning optimization of aerospace coating deposition processes for enhanced performance and uniformity

Purpose Optimizing coating deposition processes is essential for aerospace applications, particularly for thermal barrier coatings, environmental barrier coatings and protective films on turbine engine components. This study aims to develop a machine learning based framework to optimize two key chemical vapor deposition performance metrics: film deposition rate and thickness uniformity. Design/methodology/approach Building on previous work that benchmarked several machine learning models against computational fluid dynamics (CFD)–generated data, the XGBoost algorithm was identified as the most accurate predictor of deposition characteristics. In this study, XGBoost outputs were used to construct polynomial surrogate models for deposition rate and uniformity. These models enabled rapid optimization using the sequential least squares programming (SLSQP) algorithm under realistic process constraints. Three optimization cases were examined: (i) maximizing deposition rate subject to a uniformity requirement, (ii) minimizing nonuniformity with a minimum deposition constraint and (iii) maximizing the deposition-to-uniformity ratio. Optimal values of susceptor temperature and inlet gas velocity were obtained for each case. Findings The machine learning–guided optimization framework produced solutions that were both more accurate and computationally efficient than conventional optimization methods. Across all cases, the ML-based surrogate models achieved less than 5% deviation from CFD reference values, confirming their fidelity. Compared with traditional response surface–based optimization, the proposed framework reduced prediction errors by up to 40% and computational cost by approximately 60%. Originality/value This study introduces a novel hybrid methodology that integrates high-fidelity CFD simulation, advanced machine learning and constrained optimization. The approach reduces computational effort while retaining predictive fidelity, making it applicable for real-time control, digital twins and smart manufacturing of aerospace coatings. Furthermore, the methodology is extendable to other thin-film deposition technologies and multiphysics manufacturing processes.