EXTRACTION OF MANUFACTURING FEATURES AND ITS QUALITY TESTING USING CNN AND IMAGE PROCESSING

Abstract

Recognizing features automatically to extract manufacturing information from Computer-Aided Design (CAD) models for the automatic manufacturing of products has been an active research area. This automatic extraction of information can be directly learned by CNC machines to avoid complex G-codes, and M-codes for each and individual part separately, and also manual Quality testing can be avoided which can be done within the machine. In this research work, we have tried to use Deep Learning to classify manufacturing features. Our Proposed Convolutional Neural Network (CNN) model architecture is used to extract features from 2D images created by SOLIDWORKS 3D Models and assign them to the respective label to which it belongs with the highest accuracy achievable to the best of my knowledge using the PyTorch library. Dataset for experimentation of CNN model is used from Feature Net blog on GITHUB that is having 24000 models in total including 24 machining features. So, these models are first converted into 2D images by taking screenshots at different angles and in different lighting conditions. After extraction of features, the different COMPUTER VISION algorithms are used to detect the quality of extracted information using the OpenCV library by taking an example of part inspection on a Clutch plate in an assembly line. This Research work is an attempt to prepare a ready-to-go solution for Industry 4.0 to avoid any loss of information between the design stage to the production line using auto extraction of information from the model and auto Quality testing methodology using Artificial Intelligence.