Bridging Theory and Practice: Development and Evaluation of Production Scheduling Algorithms for Real-World Production Environments
Doctoral thesis
Date of Examination:2025-12-16
Date of issue:2025-12-19
Advisor:Prof. Dr. Matthias Schumann
Referee:Prof. Dr. Matthias Schumann
Referee:Prof. Dr. Jutta Geldermann
Referee:Prof. Dr. Lutz M. Kolbe
Files in this item
Name:Dissertation_v005-Veröffentlichung.pdf
Size:2.68Mb
Format:PDF
Abstract
English
Production scheduling is becoming increasingly complex due to customization, variability, and real-time disturbances, while Industry 4.0 technologies offer real-time data and opportunities for reactive control. This dissertation addresses the gap between simplified research on production scheduling and complex real-world production environments with additional constraints, by targeting two areas: the development of real-world-capable scheduling approaches and the evaluation of such approaches under realistic conditions. In the first area, this dissertation lays the foundation for the development of real-world scheduling algorithms. First, it synthesizes real-world complexities into a taxonomy, then proposes a general problem representation, and designs a reference architecture for scheduling applications that expose these complexities to algorithms. Building on these foundations, it develops and iterates reinforcement learning–based schedulers for real-time decision making, demonstrating that state-of-the-art learning methods can integrate constraints such as sequence-dependent changeovers, alternative resources, and dynamic events while remaining computationally viable. In the second area, this dissertation enables the evaluation of such real-world scheduling algorithms. It introduces Simfia, an evaluation environment that manages individually deployable scheduling services, orchestrates experiments, and reports performance against customizable criteria. Furthermore, this dissertation proposes a benchmark that incorporates real-world complexity usable for future research to compare the performance of real-world production scheduling algorithms, thereby addressing common shortcomings of existing evaluations. Overall, the dissertation presents a comprehensive view of real-world scheduling requirements, a reusable problem representation and reference architecture, reinforcement learning schedulers tailored to practice-relevant constraints, and an evaluation stack comprising a platform and a benchmark. Together, these contributions provide actionable building blocks for researchers and practitioners, narrowing the gap between academic advances and factory-floor needs and advancing the relevance and reproducibility of production scheduling research.
Keywords: Production Scheduling; Scheduling; Reinforcement Learning; Machine Learning; Industry 4.0; Advanced Manufacturing Systems
