Multitool Drilling Path Optimization by Multiagent Reinforcement Learning Approach

In hole machining, drilling path optimization (DPO) plays a critical role in enhancing production efficiency by minimizing tool path length through the strategic sequencing of drilling operations. However, existing DPO model has largely neglected the impact of variability in pilot hole sizes, leading to excessive tool changes and unnecessary tool movement, which limits their effectiveness, preventing them from fully maximizing efficiency in practical production settings. To tackle this challenge, we establish a practical model: Multitool drilling path optimization with decision (MTdDPO), which allows pilot hole sizes to be selectable rather than fixed. To solve MTdDPO, we develop a novel multiagent reinforcement learning (MARL) method, named MARL for MTdDPO (MM). In MM, we creatively treat the holes requiring pilot predrilling as agents and integrate a specially designed social attribute (SA) mechanism that equips the agents with the ability to perceive the geometric relationships with surrounding agents, thereby enhancing their decision-making capabilities. Furthermore, the convergence of MM is analyzed to demonstrate its theoretical feasibility. Finally, the effectiveness, superiority, and practicality of the MM approach, along with the soundness of the SA mechanism, are validated through performance comparisons, iteration, and learning curve analyses against several state-of-the-art methods on eight real-world workpieces of varying scales.