Application of Artificial Intelligence in Predicting Weld Quality and Defects

Automated weld defect detection is critical for maintaining structural integrity and operational safety in industrial manufacturing. Traditional inspection methods often lack consistency, scalability, and precision, especially when applied to complex defect morphologies. This study introduces a novel hybrid deep learning framework that combines the strengths of Convolutional Neural Networks (CNNs) and Transformer modules to address both classification and segmentation challenges in weld defect analysis. The proposed model is evaluated on both the GDXray benchmark dataset and a custom-developed dataset consisting of various real-world weld defect types, including cracks, porosity, slag inclusion, and lack of fusion. The hybrid architecture integrates attention mechanisms and transformer-based contextual feature extraction to improve spatial awareness and semantic representation. Experimental results reveal that the proposed model outperforms conventional approaches such as Support Vector Machines (SVM), classical CNNs, and ResNet-50 across multiple performance metrics, including accuracy, precision, recall, F1-score, and inference latency. Pixel-wise segmentation metrics—specifically Intersection over Union (IoU) and Dice coefficient—demonstrate a substantial reduction in undersegmentation and oversegmentation errors, indicating high fidelity in defect localization. Additionally, ablation studies confirm the significant contributions of each architectural component, while statistical hypothesis testing validates the robustness and generalizability of the model across datasets. Real-time inference evaluation highlights the model’s feasibility for deployment in latency-sensitive industrial applications, with reduced computational overhead achieved through pruning and quantization. The proposed CNN–Transformer hybrid framework offers a scalable, accurate, and interpretable solution for automated weld defect inspection, paving the way for next-generation intelligent quality control systems in manufacturing environments.