Optimal toolpath planning strategy prediction using machine learning technique

Abstract

Finding the best toolpath planning strategy for Computer Numerical Control (CNC) machining requires a trial-and-error approach. One needs to follow a number of steps: generating the toolpath, performing simulations/machining, measuring various parameters to quantify the quality of the toolpath, and then selecting the best toolpath. This conventional iterative approach is time-consuming and often error-prone, which is a bottleneck in the current CNC machining industry. This paper presents a novel framework to create a machine learning based system for choosing the best toolpath planning strategy for CNC machining (finishing) of complex freeform surfaces directly from the CAD model. Three tool path planning strategies are considered: Adaptive planar, Iso-scallop, and Hybrid. At first, a novel toolpath analysis module is presented to evaluate the quality of the toolpath considering three performance parameters: surface finish, toolpath length, and smoothness. This quality measurement technique is extensively tested for robustness and accuracy. It is then used to analyze and label a large number of CAD models to create a dataset for supervised learning. Finally, a Convolutional Neural Network (CNN) is designed and trained using this dataset to predict the best toolpath planning strategy. Compared to the conventional approaches, the developed system uses multiple performance parameters and chooses the best strategy directly from the CAD model. Results show that the proposed data-driven model achieved 96.8% test accuracy.