Prediction of Processing Optical Elements Results Using Machine Learning

Abstract: Improving the quality of processing optical elements in a shorter production cycle is an actual problem. The complexity of the task increases with the production of complex rare products. On the one hand, it is necessary to produce high-quality products, on the other hand, to optimize the processing process, minimizing its cycle. The paper proposes machine learning models that predict, based on historical data, the results of polishing optical elements of experimental-design production. Classification and regression models are constructed. The best results for the classification were obtained using the Xgboost and LightGBM algorithms, for the regression using the CatBoost algorithm. The achieved quality levels of models on test datasets allowed us to identify the influence of machine settings on the result of polishing using SHAP method. The obtained results were agreed with the production technologists of the Research Institute of Optical and Electronic Instrumentation.

Keywords: optical elements, experimental-design production, polishing, machine learning, variables importance