Computer Vision in Automated Defect Detection for Manufacturing

Abstract

Defect detection in additive manufacturing (AM) is crucial for maintaining quality and reliability across industries. Traditional methods primarily rely on manual inspection or isolated machine learning models, which often fail to generalize across varying defect types. This paper introduces a hybrid framework that integrates multiple Machine Learning algorithms, including convolutional neural networks (CNNs), support vector machines (SVMs), autoencoders, and explainability techniques such as SHAP and Grad-CAM. By leveraging the complementary strengths of these methods, the framework ensures both high detection accuracy and interpretability. The experimental evaluation is conducted using the MVTec Anomaly Detection dataset, which includes diverse defect types, and the results demonstrate significant improvements in classification performance and interpretability. The proposed approach paves the way for a more reliable and transparent AI-based defect detection system applicable across various manufacturing environments.