Understanding human decision-making during production ramp-up using natural language processing

Ramping up a manufacturing system from being just assembled to full-volume production capacity is a time consuming and error-prone task. The full behaviour of a system is difficult to predict in advance and disruptions that need to be resolved until the required performance targets are reached occur often. Information about the experienced faults and issues might be recorded, but usually, no record of decisions concerning necessary physical and process adjustments are kept. Having these data could help to uncover significant insights into the ramp-up process that could reduce the effort needed to bring the system to its mandatory state. This paper proposes Natural Language Processing (NLP) to interpret human operator comments collected during ramp-up. Recurring patterns in their feedback could be used to gain a deeper understanding of the cause and effect relationship between the system state and the corrective action that an operator applied. A manual dispensing experiment was conducted where human assessments in form of unstructured free-form text were gathered. These data have been used as an input for initial NLP analysis and preliminary results using the NLTK library are presented. Outcomes show first insights into the topics participants considered and lead to valuable knowledge to learn from this experience for the future.