AI-based monitoring system for real-time defect detection in wire arc additive manufacturing

The high flexibility of additive manufacturing (AM) enables the repair of components with complex geometries, contributing to the sustainability of the economy by reducing waste and increasing product lifetime. Wire arc additive manufacturing (WAAM) is well-suited for repair due to the high deposition rates compared to other AM technologies. Moreover, the automation potential of WAAM offers a promising opportunity for increasing productivity. However, the automated repair presents new challenges for quality assurance. During the deposition of successive layers, defects, such as a lack of fusion between adjacent weld beads, may be concealed within the parts. Due to their impact on the mechanical properties, such discontinuities constitute non-conformities and require part rejection. Therefore, a real-time monitoring system is required to detect lack of fusion defects and to ensure the reliable performance of the repaired components. Weld pool imaging provides detailed insights into process anomalies. Nonetheless, the harsh welding environment degrades the data quality and requires advanced imaging algorithms to extract features for a reliable analysis. The artificial intelligence (AI) architecture “You Only Look Once” (YOLO) allows for a robust detection performance and real-time capability with its one-stage approach for object detection. In this work, a monitoring system using two sequentially coupled YOLO-based models was developed. First, a detection model identifies the lack of fusion defects within the weld pool images, producing bounding boxes around the detected areas. These bounding boxes are then used as an input for a segmentation model, which provides a more precise delineation of the defects within the identified regions. The models were evaluated on unseen data, achieving a recall of over 90 % while maintaining real-time capability. This result showed the high potential of AI-based monitoring systems for real-time defect detection in WAAM to ensure the quality of the repaired components.