Architecture and implementation of a hybrid prototype system of a digital spectral twin of thin films

Abstract

The relevance of this work is driven by the need to develop intelligent information systems for processing spectroscopic data and identifying material parameters in real time. The aim of the study is to develop a hybrid architecture prototype of a digital spectral twin for real-time identification of thin-film parameters. To achieve this aim, the tasks of integrating physical and neural network models, implementing an online self-adaptation mechanism, and evaluating the system’s performance on experimental data were addressed. A numerical scheme based on the transfer matrix method and physics-informed neural networks was implemented for modeling spectral responses and adaptive parameter estimation. The online self-adaptation mechanism, which continuously updates material parameters based on experimental spectral data under the presence of noise and slow drift, was investigated, along with the analysis of parametric identifiability using local spectral sensitivity. The results of the study is the development of an architecture that ensures stable convergence of thickness, refractive index, and absorption coefficient estimates over 100–200 iterations, and convergence of the overall spectrum up to 300 iterations. Relative errors below tree percent for thickness, hundredths of a percent for the refractive index, and tenths of a percent for the absorption coefficient were achieved. The model’s ability to track gradual parameter changes, maintain high predictive accuracy over long-term operation, and reduce spectral reconstruction errors was demonstrated. Robustness to noise and non-stationary conditions, as well as accurate reproduction of spectral maxima, relative intensities, and selective approximation, were confirmed. The developed hybrid architecture is an effective tool for intelligent spectroscopic monitoring and material identification. Further research will focus on extending the system to multilayer structures and industrial measurement applications.