A central task in production engineering is the parameterization of manufacturing processes and machinery. The parameterization has a significant impact on product quality, process efficiency, and profitability of the production. Bayesian optimization (BO) - an adaptive black-box optimization algorithm for efficient and performance-optimal parameterization - has emerged in recent years as a promising alternative to conventional experimental design methods such as design of experiments, one factor at a time, or trial and error. Because optimization of manufacturing processes falls under the responsibility of human process experts, close collaboration between BO and human experts is key to successful optimization. Although first approaches to collaborative BO exist, intuitive dashboards that communicate and explain parameter suggestions and optimization progress to process experts are missing. In this paper, we propose a three-phase pipeline for collaborative BO, motivate the need for a collaborative Bayesian process optimization dashboard and define a total of 15 requirements for the dashboard design. Based on this, we propose a design concept for the BO-dashboard comprising multiple metrics and visualizations to explain parameter suggestions, create transparency in the optimization process, and promote the accumulation of process knowledge. We showcase the implementation of the dashboard at the example of optimizing an ultra-short pulsed laser ablation process. By enhancing human-BO collaboration, we aim to promote the adoption of BO within the conservative industry of production engineering.
Leyendecker et al. (Thu,) studied this question.