Deep reinforcement learning-based tool path generation for reducing non-cutting motion in multi-island cavities

Multi-island cavities are characterized by complex geometries, irregular boundaries, and discretely distributed islands. Tool paths generated by traditional zigzag cutting often suffer from frequent tool retractions, excessive non-cutting motion, and interference caused by islands. To address these issues, this paper proposes a deep reinforcement learning-based tool path generation method. First, the machining region of a complex cavity is discretized into driving lines, and continuous line groups are constructed through geometric processing and a grouping algorithm. In this way, the tool path generation problem is transformed into a multi-stage sequential decision-making problem, and a Markov Decision Process model is established. On this basis, a Deep Q-Network with prioritized experience replay is employed, enabling the agent to learn the endpoint visiting sequence through interaction with the environment and thereby reduce the total non-cutting motion distance. Compared with the local heuristic method and the classical zigzag scan algorithm, the proposed method reduces the total non-cutting motion distance and the number of tool retractions, especially for complex multi-island cavities. The results indicate that the proposed method has good adaptability and global optimization capability, providing a new solution for tool path generation in zigzag cutting of complex aero-engine cavities.