A Sliding-Window-Based Reinforcement Learning for Dynamic Assembly Flow Shop Scheduling with Multi-Product Delivery

Short summary

Researchers developed a sliding-window reinforcement learning framework to optimize real-time scheduling in hybrid manufacturing systems with dynamic order arrivals and complex kitting dependencies for multi-product assembly. Tested on actual home appliance factory data, the method consistently reduces order tardiness compared to classical dispatching rules and existing deep reinforcement learning approaches. The system dynamically adapts to variable production conditions, resource availability, and shifting bottleneck constraints.

• •RL framework handles dynamic manufacturing scheduling with multi-product kitting constraints
• •Real-world factory testing shows consistent tardiness reduction vs traditional and existing RL methods
• •Adapts to changing production conditions, resource availability, and bottleneck shifts

Generated with AI, which can make mistakes.