On Combining Convolutional Autoencoders and Support Vector Machines for Fault Detection in Industrial Textures

Area of Science:

• Materials Science
• Computer Science
• Artificial Intelligence

Background:

• Defects in textured materials exhibit high variability, necessitating tailored detection solutions.
• Existing methods often struggle with the diverse nature of defects in textured surfaces.

Purpose of the Study:

• To develop a robust and automated defect detection system for textured materials.
• To combine Convolutional Autoencoders (CA) and Support Vector Machines (SVM) for enhanced defect identification.

Main Methods:

• Utilized two machine learning approaches: Convolutional Autoencoders (CA) for image reconstruction and one-class Support Vector Machines (SVM) for classification.
• Trained models exclusively on defect-free textured images, with automatic sample labeling for SVM.
• Implemented two image processing streams: CA for reconstruction-based defect measurement and SVM for classification using latent layer features.

Main Results:

• Achieved an average success rate of 92% in defect detection.
• The combined approach demonstrated superior performance compared to previous methods.
• Successfully integrated image reconstruction and classification measurements for improved accuracy.

Conclusions:

• The proposed hybrid CA-SVM model offers an effective and automated solution for detecting defects in textured materials.
• This approach addresses the variability challenge by leveraging unsupervised learning and feature extraction.
• The high success rate indicates significant potential for industrial applications in quality control.