Multi-Objective Lookahead Bayesian Optimization for Process Parameter Optimization in Non-Isothermal Glass Molding

The optimization of non-isothermal glass molding (NGM) processes is crucial for attaining precise shape accuracy of optical components. Identifying optimal parameters poses a challenge due to unknown functional relationships and high-dimensional design and target spaces. Traditional design of experiments (DoE) approaches are sample inefficient in this regard. This paper presents a multi-objective lookahead Bayesian optimization framework applied to a use case from NGM, in which a glass gob is formed into a light optic. The approach leverages a multi-target Gaussian Process-based surrogate model. Through novel acquisition functions, the framework adeptly balances exploration and exploitation, resulting in a significant reduction of samples necessary. Improved peak-to-valley values for the glass optics demonstrate the improvement of the product quality with regard to the application of the approach. The developed framework offers a flexible, efficient approach, contributing to industrial process optimization.