An automated approach to reuse machining knowledge through 3D – CNN based classification of voxelized geometric features

Abstract

The enhanced digitalization in the manufacturing sector is claimed to facilitate the generation or the use of the existing process data incorporating the production variations and offers a significant increase in the productivity and efficiency of a system. Moreover, manufacturing companies possess substantial knowledge while designing a product and manufacturing procedures. The primary requirement is to link and organize all the information sources related to the operation design and production. This research is concerned with the reuse of machining knowledge for existing and new parts having similarities in geometric features and operational conditions. The proposed methodology starts by extracting each machining operation's geometric information and cutting parameters using industrial part programs in the numerical control (NC) simulator VERICUT. The removed material between two consecutive operations is obtained through mesh comparison in the simulator to analyze the feature interactions. A deep learning approach based on 3D convolutional neural networks (CNN) is applied to classify similar geometries to reuse the process design knowledge by creating a library of operations. The proposed approach is implemented on actual machining data, and the results demonstrate the effectiveness of the proposed solution. The obtained knowledge clusters in the operations library assist in making propositions related to operational parameters for similar geometric features during the process planning phase reducing the planning and designing time of operations.