Cutting Path Planning Using Reinforcement Learning with Adaptive Sequence Adjustment and Attention Mechanisms

Abstract

In industrial manufacturing, cutting path planning is very important since it directly affects cutting quality and efficiency. However, traditional methods are no longer suitable for large-scale and real-time cutting path planning due to the long computation time needed. Currently, though the deep learning method can be used for cutting path planning, the node number has to be fixed, and a large amount of labeled data is required. Another method is deep reinforcement learning, which can be used for cutting path planning. The node number also has to be fixed. As a result, these two potential methods are unsuitable for practical industrial cutting problems. To solve the above problem, this study provides a new reinforcement learning approach that integrates adaptive sequence adjustment and attention mechanisms. Compared to traditional methods, which took hours or days to finish a complicated cutting path plan, this approach significantly improves processing efficiency, completing tasks in seconds, and also minimizes non-cutting travel paths. And compared to deep learning and deep reinforcement learning provided by others, a variable number of nodes can be processed with this method. Therefore, it is more suitable for practical industrial cutting problems.