Deep Learning-Based Defect Inspection in Sheet Metal Stamping Parts

Abstract

Defect inspection is a crucial step in sheet metal stamping manufacturing. However, current inspection methods largely consist of visual inspection by trained operatives but are unreliable and prone to error. Computer vision techniques have the potential advantage of utilising low cost hardware to enable accurate classification of defects particularly through using techniques such as deep learning. Currently, the use of convolutional neural networks (CNN) is one of the best methods in the field of computer vision for classification tasks. Despite the advantages, vision-based deep learning models for detecting defects in sheet material are currently limited to flat sheet materials and certain classes of surface defects such as scratches and delamination. This research proposes a practical deep learning approach for classification of cracks in realistically formed sheet metal stamping components and suggests a route towards reliable and automated inspections in sheet metal stamping. This study used ResNet18, a state-of-the-art deep learning model to classify split defects in “Nakajima” stamped components. The model was able to achieve a $99.9 \%$ accuracy on validation set, which implies that this technique could be suitable for automated defect detection on stamped metal parts.