Machine learning based image analysis for surface defect detection

Abstract

The progressive and intelligent advancement of the manufacturing industry demands precise quality control to ensure product excellence. The surface defects that arise during the manufacturing processes pose significant concern as they can lead to quality issues and compromise production integrity. The traditional surface defect detection methods, reliant upon human-driven visual inspection, are limited by accuracy, speed, and adaptability across diverse defect categories. To address these challenges, this project introduces an innovative approach that utilizes the application of advanced machine vision techniques, known for enhancing the efficiency, performance, and reliability of defect detection. Currently, the machine vision-based defect detection methodologies often rely on conventional image processing algorithms. However, these methods prove inadequate in achieving optimal results and the existing literature on automated detection in this area is limited. Therefore, this project proposes a novel methodology that leverages Convolutional Neural Networks (CNNs) to automate the process of detecting surface defects. The primary focus of this project lies in the formulation and execution of a CNN-based image analysis framework specifically tailored for accurate surface defect detection and identification.