Tags:Automation, Computer Vision, Deep Learning, Defects, Inspection, Machine Learning and Remanufacturing
Abstract:
Inspection in remanufacturing is a labour-intensive and time-consuming step that involves identifying defects on the surfaces of components that are candidates for remanufacturing. Traditional techniques have limitations in terms of cost, detecting multiple defects at a time, and inspector reliability. As a result, automated inspection techniques have garnered remanufacturers’ attention because of their potential cost advantage and improved defect detection capability. This study examines the capability of optical inspection techniques to decrease inspection costs and errors in remanufacturing. We implemented object detection methods to classify and locate defects on steel surfaces from surface images with reasonable accuracy. The YOLO (You Only Look Once) V4 algorithm was used to capture and classify the defects. The performance of the algorithm is compared with state-of-the-art approaches using recall, average precision, and mean average precision metrics. Our model demonstrates effective defect localization with a mean average precision (mAP) of 64.13%, which shows promise for the development of automated inspection technology.