Towards Machine Learning for Quality-Driven Production Optimization of Optical Thin-Film Coatings

Machine learning (ML) is increasingly applied across various industrial domains; however, its adoption in the field of industrial production optimization of optical thin-film coatings remains limited. Existing research primarily focuses on the design of optical multilayer systems or the modeling of material properties to enhance application-specific performance. In contrast, the application of ML for quality-driven production process optimization and control remains largely unexplored due to complex interactions among coating and machining parameters. This paper addresses this gap by outlining a roadmap for implementing ML-based quality optimization in optical multilayer thin-film coating processes, using ion beam-assisted physical vapor deposition (IBAD-PVD) as a representative case. The investigated use case involves the coating of optical lenses, where ML models are trained on time-series data of machine parameters to predict spectral performance metrics relevant to quality assessment. The paper first presents experimental results from initial studies which, although preliminary, show promising trends. The highest achieved coefficient of determination (R²) in these experiments is ~0.5. Based on a feature importance analysis and other empirical findings, the paper proposes new conceptual approaches and outlines future research directions aimed at advancing ML integration into industrial thin-film production workflows.