Article
Deep Feature Fusion for Hierarchical Defect Classification in Industrial Inspection Systems
Defect classification in industrial components is critical to ensuring product quality, operational safety, and cost efficiency, as manufacturing industries face increasing defect rates due to highspeed production and complex processes. Studies indicate that a significant percentage of industrial failures originate from undetected surface and structural defects, leading to rework, downtime, and financial loss. This research is motivated by the need for accurate and automated defect detection in machinery parts, painted surfaces, and welded joints, including subtypes such as cracks, corrosion, paint blisters, scratches, porosity, lack of fusion, and weld spatter. The expected outcome is a robust classification system capable of reliably identifying defect categories and subcategories across these domains. Traditional defect inspection relies heavily on manual visual inspection, which is time-consuming, subjective, and prone to human error. Such manual systems struggle with consistency, scalability, and real-time deployment in modern industrial environments. In this work, RGB images of industrial components are utilized, followed by image preprocessing techniques including resizing and normalization to enhance feature consistency and model performance. Existing machine learning approaches such as K-Nearest Neighbours (KNN) and Decision Tree Classifier (DTC) are reviewed as baseline models for defect classification. The proposed approach employs Convolutional Neural Network (CNN)-based feature extraction combined with Logistic Regression (LR) for efficient and accurate classification. The system outputs precise defect classification results for machinery, paint, and welding components, including their respective defect subtypes, demonstrating improved accuracy and reliability over traditional methods.
Full Text Attachment
