Industrial Product Surface Defect Detection Using CNN: A Deep Learning Approach

Abstract

Defect identification on surfaces of industrial products is a thought-provoking issue that has garnered significant attention. Defect identification of industrial products plays a primordial role in ensuring quality and reliability of products at all levels of the manufacturing process. The traditional approach relies fundamentally on the human inspection, which was unreliable and inefficient. Moreover, human inspection cannot meet the high standards for real-time detection required in industrial production situations. While methods used in image processing can solve difficult modules, they are not meant to deal with complicated ambient textures, noise, or lighting fluctuations. Innovative techniques have so been used by researchers and practitioners to improve and expedite the defect diagnosis process. This study classifies the product as "defect or not defect." and provides an intelligent approach for surface defect identification using convolutional neural network (CNN). The model demonstrates strong performance in identifying many types of industrial product defects, such as "crack, patch, inclusion, rolled, pitting, and scratching," by utilizing deep learning approaches. This involved training the model with several datasets featuring surface textures. The results were indicative of better performance, where the proposed method achieved outstanding precision in fault detection.