Using Deep Learning to Improve Simulation-Based Decision Making by Process Lead Time Predictions

Based on digital twins, simulation is often used in companies for the regular planning and control of operational processes. However, when modeling the lead times of individual processes, mean values of process times measured in advance are often used, which can lead to errors in the planning. This work demonstrates how models of these time distributions can be created and updated within a digital twin framework using machine learning. The lead times are used in simulation to create schedules. The approach is validated using the online order workforce scheduling of a medium-sized company that assembles individual packages of office materials for its customers.