Workforce rostering for decentrally controlled production systems: A simulation-based optimization framework using a genetic algorithm

Decentral production control plays a crucial role within the paradigm of Industry 4.0. Due to the fast and flexible decisions on allocation and sequencing required by this type of control, there is no baseline production schedule in advance. This creates a dilemma for efficient staff deployment - typically worker deployment times must be planned at least a few days ahead. To solve this dilemma, we present a simulation-based genetic algorithm, which creates a roster with flexible deployment intervals without a rigid shift pattern based on the production system and job load. In accordance with the zeitgeist and Industry 5.0, we include flexible working time and desired working hours of production workers. For evaluation of the method, we consider worker attendance costs, job delay costs and a cost penalty of work scheduled outside of desired working hours. We forecast the decisions of the decentralized production system by solving a job shop scheduling problem (JSP) extended by manual operations. Our algorithm iteratively uses reasonable solutions of the JSP as basis for roster optimization. With this integrated approach, it is possible to balance job delay costs against worker attendance costs as well as cost for deviation from desired working hours. To ensure compliance with working time legislation, we include appropriate repair operators in the genetic algorithm. We demonstrate the efficiency of our heuristic approach by comparison to rigid shift systems and the best of a large number of randomly created rosters.