<p>Structural integrity in robotic gas tungsten arc welding (GTAW) depends on the quality of the weld; however, traditional inspection techniques often fall short in simultaneously identifying surface and internal defects in real-time. To overcome this constraint, this study aims to develop a multi-sensor-based intelligent monitoring system that can detect a wide range of welding defects, including both surface and internal defects. A YOLOv11-based convolutional neural network has been employed to detect surface defects, including burn through, lack of fusion, misalignment, and contamination in weld images. A Long Short-Term Memory (LSTM) model has been used to detect internal defects, i.e., porosity, cracks, and lack of penetration. Time-series data of sound pressure level from the acoustic sensor and arc voltage data from the Hall sensor have been analyzed to detect internal defects. The results revealed that the mean average precision mAP@0.5 of 0.994 and mAP@0.5:0.95 of 0.836 have been attained in the case of the YOLO model for surface defects. The accuracy of the LSTM model for Hall data and acoustic data has been attained at 99.89% and 99.52%, respectively, for internal defects. The results of the developed system are further validated by thermal and X-ray analysis of weld beads. The validation results showed that the developed system is reliable and effective in real-time monitoring of both surface and internal defects during the GTAW process. The developed system provides a scalable solution for reducing rework, improving process reliability, and enhancing product quality across automotive, aerospace, construction, and shipbuilding industries.</p>