Machine Design Automation Model for Metal Production Defect Recognition with Deep Graph Convolutional Neural Network

Abstract

Error detection has a vital function in the production stages. Computer-aided error detection applications bring significant technological innovation to the production process to control the quality of products. As a result, the control of product quality has reached an essential point because of computer-aided image processing technologies. Artificial intelligence methods, such as Convolutional Neural Network (CNN), can detect and classify product errors. However, detecting acceptable and small defects on base parts cannot be done with a high rate of accuracy. At this point, it is possible to detect such minor errors with the help of the graph convolutional network, which has emerged as a new method. In this study, the defect elements on the surfaces of metal nut parts are determined through the graph convolutional network, and quality control is ensured. First, the surface images of the metal nut parts are captured. For this, a python-based Raspberry pi card and a modified camera system were installed. Adapters with three different zoom options are used on the camera system, depending on the part to be captured. The images obtained in the second step are sent to the other computer, which is used for image processing via the local server. In the third stage, image transformations are obtained by graphically separating the obtained images in white and black color tones on the second computer, and histogram maps of these images are drawn. Value ranges of these maps are determined and classified according to the value ranges obtained from the images of the defective parts. As a result, nine different models were analyzed. According to the analysis results, the graph convolutional neural network method gives 2.9554% better results than conventional methods.