Detection of Defective Deep Drawn Sheet Metal Parts by Using Machine Learning Methods for Image Classification

The progressing digitization in the manufacturing industry and the enhancements in the field of machine learning (ML) offer new approaches for monitoring manufacturing processes. In this context, the image classification using ML methods represents a well-solved task that can be leveraged for evaluating processes and products. From this perspective, this paper presents an application of ML methods in order to classify defective deep drawn parts. It proposes a methodical approach which includes not only the data acquisition using a low-cost equipment, but also the data pre-processing as well as the setup and training of the ML model. The described method is applied on an academic part of rectangular shape. After collecting a dataset of nearly 1000 images, a convolutional neural network was trained to identify three different quality classes. The trained model was then evaluated on a test dataset containing 30% of the samples and shows an accuracy of 96,77%.