DEEP LEARNING FOR AUTOMATED DEFECT DETECTION IN MOLDED PRODUCTS

Abstract

With anything that relates to improving productivity and maintaining quality, smart factories implement a whole lot of modern technological solutions. This paper proposes an automated fault classification of products using deep learning in a smart factory environment. An AI-based method is employed to accomplish the objective of detecting defective items and removing them from the production line. At its core, a Programmable Logic Controller (PLC) acts as the central control unit and coordinates operation among all the modules interfaced thereto. On the other hand, the defects are classified by the AI-based boards using pre-trained Convolutional Neural Network (CNN) models. Transfer learning is applied so that it fine-tunes the models on a GPU server, thus greatly reducing the training time using in-house data yet achieving very high accuracy in classification, all in contrast to learning from scratch. This framework allows fast, reliable, and scalable defect detection of industrial casting processes toward much more intelligent and efficient manufacturing system.