Probabilistic Simulation and Determination of Sojourn Time Distribution in Manufacturing Processes

The ongoing transformation from automated to smart manufacturing offers new capabilities to fulfill the multidimensional challenges and demands of a globalized market. Smart in this context covers methods of data mining and machine learning to generate high-value process information. As modern manufacturing industry is characterized by a big amount of heterogeneous data from different sources, the temporal relations between the data is generally unknown. Token-based solutions (such as RFID chips) offer a highly reliable tracking but are not suitable in every scenario; possible reasons are environmental conditions or fluid products. In this paper, we introduce a directed graph model of a stochastic manufacturing process. This can be used for process simulation to generate data from different plant topologies and transition conditions. It is validated by means of a plant model representing an automated handling process containing three industrial robots. As the sojourn time distributions of the work pieces are not only dependent on the work times but also on the transition conditions like capacity and workload of the workstations, it has to be identified considering both. The sojourn time distributions can be transferred to a new process graph model, suitable for work piece tracking.