Welding Seam Classification in the Automotive Industry using Deep Learning Algorithms

摘要

Welding seam inspection is key process in the automotive industry and should guarantee the quality required by the client. Visual inspection is achieved by the operator who checks each part manually, making the reliability highly improvable. That's why automating the visual inspection is needed in today's production process. Collecting data from inside the plant may not provide a balanced number of images between good welding seams and bad welding seams. In this article, we will compare a standard deep learning algorithm applied on raw data with data augmentation approaches. Our target is to reach an accuracy of 97 % on the defected reference parts. This target is reached on some welds, while it remains a challenge on other welds.