Font Size: 

This paper presents a method for detecting the weld pool in images of weld beads produced during the cladding process. By applying deep learning algorithms, it was possible to accurately localize the weld pool and perform automated geometric analysis of the extracted region. The study utilized static images captured after the cladding process, which were manually annotated in terms of weld pool position, size, and shape. Detection and segmentation models (including YOLOv8-seg) were tested and trained on a dedicated dataset. Subsequently, a set of geometric features such as area, ellipticity, symmetry, and offset from the weld axis, was extracted from the identified weld pools. In the next stage, the results of weld quality classification based on these features were compared with expert evaluation. The proposed method is a part of a broader framework aiming to support quality assessment and technological parameter estimation in automated or robotic cladding processes, including 3D cladding using the WAAM method.