Automated human-in-the-loop commissioning of industrial processes: a glue dispensing case study

This paper addresses the challenge of tuning process parameters during the commissioning of industrial processes. The process industry is hindered by unique challenges, including the absence of physical models and weakly specified bounds on process parameters. Moreover, multiple experiments can be performed simultaneously in a batch, but only under some extra constraints, such as a constant temperature for every experiment in the batch. These observations motivate the need for practical commissioning tools that can assist human machine operators or technicians. While there are solutions available for automatic tuning, they often lack the simplicity required for an operator-in-the-loop to easily understand and fine-tune the automatic tuning without being a machine learning expert. Therefore, we propose a simple Bayesian optimisation (BO) method with a single hyper-parameter and a systematic approach that can be easily understood. The effectiveness of the algorithm is demonstrated for glue dispensing, a complex industrial process with diverse properties that make tuning cumbersome. We validated our method in three cases, reducing the commissioning time by an inexperienced human-in-the-loop from days to hours.