Industrial production is facing growing challenges: varying product configurations, ever-shorter innovation cycles, and unstable supply chains are continuously driving up the complexity of planning and control processes. Classic, rule-based planning systems reach their limits as soon as manufacturing processes have to respond flexibly to changing components, customer requirements, or supply bottlenecks.
In particular, (re-)sequencing, i.e., the dynamic reordering of manufacturing steps and component sequences, often fails due to the limited transferability of existing solutions: previous approaches are usually tailored to a specific use case, embedded in custom software, and cannot readily be transferred to new plants, products, or customers. The PreAIrranging project (Preparing an AI-Based Rearranging Hub) addresses this bottleneck and lays the foundation for an open, scalable (re-)sequencing hub in Saarland that optimizes production and other sequential decision-making processes using data and AI.
At the core of the project is the development of a modular cloud platform that connects hybrid (re-)sequencing algorithms with digital twins of real production systems, thereby enabling both simulation-supported AI training and high-performance live operation. Previously tested but so far isolated solution approaches are being brought up to the current state of the art, developed further from individual solutions into generally applicable building blocks, and integrated into the platform. These approaches are complemented by new methods of deep reinforcement learning researched within the project, which are being systematically applied to industrial (re-)sequencing for the first time. To ensure that AI-made decisions remain comprehensible to users, explainability (XAI) is considered from the start: a UX dashboard makes both the current state and the AI's decision logic transparent, thereby creating the basis for acceptance, trust, and effective human oversight.
DFKI coordinates the project through its Neuro-Mechanistic Modeling research department (Prof. Dr. Verena Wolf, Saarbrücken). Project partners are abat+ GmbH from St. Ingbert, which contributes many years of experience in digital high-availability solutions for the manufacturing industry as well as existing platforms, cloud architectures, and dashboards, and the Chair of Modeling and Simulation (Prof. Dr. Verena Wolf) at Saarland University, which contributes expertise in deep reinforcement learning methods.
By the end of the 12-month project period, the concept for the (re-)sequencing hub is to be fully developed, the technological infrastructure built, the preliminary work of all partners updated to the latest standard and transferred to the platform, and initial independently developed solution approaches from reinforcement learning available.
Partners
- abat+ GmbH
- Universität des Saarlandes
