Joint Optimization of Feeder Allocation and Pickup Scheduling in Pcb Assembly Via Cooperative Multi-Agent Learning

Printed circuit board assembly optimization for beam head placement machines plays a critical role in enhancing the efficiency of electronics manufacturing. To address the coupled sub-problems of feeder allocation and pickup scheduling, this paper proposes a cooperative multi-agent deep reinforcement learning framework. The optimization objective encompasses both assembly cycles and total pickup operations, prioritizing the former due to its dominant impact on assembly efficiency. The proposed framework features two functionally distinct agents utilizing attention-based encoder-decoder networks to construct their respective policies. Trained via an actor-critic reinforcement learning algorithm, these agents collaboratively minimize total pickup operations under theoretical cycle constraints. Experiments demonstrate that the proposed method yields average performance improvements of 9.77% over the mainstream approaches. Furthermore, the framework achieves a substantial reduction in computation time compared to these baselines, clearly demonstrating a significant advantage in overall computational efficiency.