Abstract
In the aerospace and automotive industries, structural integrity depends on it, and it is extremely important to automate weld defect detection because hand X-ray inspections are generally error-ridden or inefficient. The proposed work is presented as an advanced deep learning framework employing convolutional neural networks (CNNs) and traditional data augmentation techniques to mitigate the poorly distinguishable defects between X-ray and scanning electron microscope (SEM) images. The GDXray Welds Dataset was used to fine-tune a ResNet-50 model pretrained on ImageNet to tackle data scarcity. Considering these experimental results, the proposed approach achieved an F1-score of 0.93 and a mean average precision (mAP) of 0.90, which was significantly better than the baseline models, including the vanilla ResNet-50 (F1 0.80) and SVM-based classes (F1 0.63). The system demonstrated high efficiency but had problems stemming from session constraints in cloud-based environments and reduced sensitivity to sub-millimeter defects. The findings demonstrate the practicality of applying AI in the practice of quality assurance in the industrial field, particularly for small-scale operations. This study fills the gap between industrial needs and the academic development of AI-based manufacturing automation systems to become scalable and sustainable.
Keywords: Convolutional Neural Networks, Data Augmentation, Industrial Automation, Nondestructive Testing, Weld Defect Detection.