Defect Inspection on Metal Production using image processing

Abstract

Enhanced product performance and quality control are essential for advanced industrial systems. Nonetheless, various surface defects adversely affect the product's performance, as well as its aesthetics and functionality. This study investigates the detection and classification of surface defects in industrial materials using advanced deep learning techniques. This paper focuses on quantifiable targets based on baseline performance in manufacturing environments through the utilization of deep learning models applied to the NEU Surface Defect Database. Convolutional neural network (CNN)-based approaches as MobileNetV2 Model and ResNet50V2 Model are employed, leveraging image processing techniques to identify six common types of surface defects. The MobileNetV2 Model is implemented and tested to obtain high accuracy besides specifying the resulting limitations. On the other hand, the ResNet50V2 Model is enhanced by adding more layers to overcome overfitting and the resulted limitations of the MobileNetV2 Model, achieving accuracy of 95% for classifying each defect. These results demonstrate that the proposed framework achieves high accuracy in defect detection, proving that deep learning techniques can significantly improve quality control in manufacturing processes, reducing manual inspection efforts and minimizing errors.