Intelligent scheduling of aerospace precision forgings based on deep reinforcement learning and graph neural networks
To address the issues of extreme time-consuming bottlenecks such as special heat treatments in the manufacturing of aerospace precision forgings and scheduling failures caused by sudden rush orders, an intelligent scheduling model based on Graph Neural Networks (GNN) and Deep Reinforcement Learning (DRL) is proposed. The model uses disjunctive graphs to rigorously represent the topology of forgings processes and equipment status, achieving Markov Decision Process modeling for dynamic production line scheduling; a scale-independent policy network is designed to extract highdimensional features, and the Proximal Policy Optimization (PPO) algorithm is employed for autonomous training. Experiments show that under harsh conditions such as dynamic rush orders and extreme bottleneck surges, the model's solution quality and robustness are significantly superior to heuristic rules like SPT and MWKR.Under the zero - shot condition, the small - scale (15×15) trained policy can be directly generalized to ultra - large - scale (30×20) scenarios. The single - instance inference only takes 2.7 seconds, providing a new paradigm for the hard real - time scheduling of aerospace precision forgings.