Machine Learning for Predictive Quality in Optics Production

Currently, digitization potentials are not yet fully exploited in optics production. Data is collected along the process chain of glass and plastics molding, but the collected data is often not connected and therefore not suitable for analysis. Furthermore, high manual effort is required to measure the produced lenses or to determine their product quality. Therefore, the objective of this paper is to realize predictive quality by predicting the quality of a lens using machine learning with production data. For that purpose, two use cases from a research factory are considered. These include the prediction of the center thickness in the precision glass molding process and the scrap prediction in the injection molding process of polymer lenses. A procedure was developed containing the necessary steps to realize predictive quality applications with focus on optics production. Following the procedure, two machine learning models were trained including the respective data preprocessing for each use case. Firstly, the successful realization of predictive quality within a research environment showcases the potential of machine learning for optics production and promotes the transfer into the industry. Secondly, the resulting models support the process experts to understand underlying interactions between the input and output variables and are the basis for future research regarding the optimization of process parameters.