A Scalar Field–Based Toolpath Generation Method Using Convolutional Neural Network

Finish machining is the final operation in numerical control (NC) machining and directly determines the surface integrity and dimensional accuracy of the workpiece. Its toolpath governs the motion of the cutting tool along the workpiece surface, thereby serving as a critical factor influencing both machining efficiency and quality. Conventional toolpath planning methods often suffer from non-uniform path spacing and excessive machining redundancy. In order to overcome the above limitations, a scalar field–based toolpath planning framework is developed, where self-supervised learning is employed to construct an optimal scalar field and toolpaths are extracted from its iso-level lines. The method starts with the generation of a maximum machining strip width direction field and then formulates two optimization objectives—alignment with the target direction field and preservation of uniform scallop height—which serve as the loss functions for the learning model. Furthermore, a novel convolutional module is introduced to effectively integrate the spatial geometry, curvature, and directional vector features of the surface. Based on this module, a Scalar Field Generating - Convolutional Neural Network (SG-CNN) is developed. Experimental evaluations demonstrate that the scalar fields generated by the SG-CNN model are well aligned with the desired geometric and machining criteria, and the resulting toolpaths exhibit high continuity.