Automated contour mill-tool grinding path generation via deep learning

This paper attempts to introduce a design architecture for the contour mill-tool grinding. Due to the unexpected three-dimensional interference known as the undercut/overcut between the grind wheel and the contour mill-tool, the corresponding design needs to go through a complicated iterative process in order to obtain the path generation parameters to compromise the flute constraints at different cross-sections. Neural networks (NNK) in recent years have proven to be of great significance in decrypting nonlinear relationships, use of which greatly reduces the computation time from hours to minutes for the tool designer. The key to success is either using a database or the NNK result to reduce the computational burden in the conventional numerical iterations. The aim of this work is to develop an automated NNK search using reinforced adaptation to learn the domain where the undercut/overcut problem is needed to be compromised. Furthermore, a new neural architecture algorithm guarantees the model convergence and result precision. The grind wheel position and orientation are then converted into NC codes for a specific five-axis grinding machine to perform either the cutting simulation or manufacturing. The experiment on machining and design comparisons to the state-of-art software have been included in this paper.