Real-time thermographic monitoring for automated defect detection in welding

Welding quality is quite critical in manufacturing. This work provides a method to enhance defect detection in robotic welding by combining statistical approaches with real-time thermographic monitoring. A robotic arm welds, and an infrared camera records temperature data. After segmentation and morphological processing, key features are extracted and optionally smoothed (e.g., via Savitzky–Golay). Principal Component Analysis (PCA) reduces dimensionality, and an Isolation Forest then identifies anomalies that may indicate defects. The approach aims to minimize human involvement, improve homogeneity, and integrate statistical robustness, thereby enhancing the dependability and efficiency of robotic welding techniques.