Hybrid deep learning approaches for multiple fault diagnostics in intelligent manufacturing systems

Abstract

Early detection of surface defects in smart systems within industrial production processes is critical for automating quality control processes, reducing costs, and maintaining quality standards. Smart manufacturing systems enable the detection of different types of defects through multiple defect diagnosis. With the increase in data volume on production lines due to Industry 4.0, classical machine learning approaches have proven insufficient for defect diagnosis. This has led to a growing need for more advanced analysis methods. In this thesis, hybrid deep learning architectures are used to classify different types of defective data samples with high accuracy. This thesis examines hybrid deep learning approaches for identifying multiple defects on metal and fabric surfaces. In the proposed method, different deep learning architectures are used together to perform both object detection and segmentation. In this study, features obtained from defective images are extracted using convolutional neural networks, and then these features are analyzed with different classifiers. The deep architectures used in the classification phase are innovative approaches such as Vision Transformer (ViT) and attention-based deep neural networks. Experimental studies have shown that the proposed method achieves high performance in detecting different types of failures. This study contributes to the improvement of quality control processes in intelligent manufacturing systems.