Numerical Data Analysis with Machine Learning for Optical Lens Polishing Conditions

Abstract

In the manufacturing industry, operations such as polishing and welding, wherein sensation and experience account for much of the work, rely on the skills of skilled technicians. However, automation has been sought in such processes. We proposed a method for the evaluation of the polishing conditions of an Oscar-type polishing system, which is used for high-mix low-volume production of lenses with various curvatures, by determining the surface pressure distribution of the lens through the preliminary examination of the lens polishing conditions. In particular, (1) a judgment index focusing on the number of Newtonian rings, which decreases as polishing progresses, is adopted. (2) Extraction of important types of polishing conditions is conducted through standard analysis. (3) By using the numerical values of the polishing conditions, the number of Newton rings is predicted with high accuracy. Through the evaluation based on the above-mentioned factors, the prediction based on the newly introduced surface-pressure-distribution-based method contributed to the improvement of accuracy, and the predicted values were sufficient to cover the ability of the operator’s evaluation.