Predicting processing times in high mix low volume job shops

Abstract

Production planning is essential for any manufacturing company, especially when complex and varied processes must be considered. Accurate processing times play a critical role for scheduling production runs and allocating resources effectively. In practice, the respective master data from the ERP system is often used for this purpose. However, maintaining the master data is challenging, especially with large amounts of data in flexible environments. In this context, incorrect or outdated data quickly lead to significant planning inaccuracies. This paper presents a study that uses machine learning (ML) models to accurately predict the processing times of single operations of future production runs based on historical production runs. Various ML algorithms were trained and evaluated on a real-world dataset. In comparison to the master data the root mean squared error could be reduced by 23% using ML. Thus, these estimated times can be used for optimizing future schedules and incorporating such an ML model in the production planning process eliminates the need for master data.