Image-Based Analysis and Estimation of Welding Features for Adaptive Control

Sammanfattning

Real-time monitoring and control of Pulsed Gas Metal Arc Welding (GMAW-P) isimportant for maintaining process stability and weld quality. This thesis investigateswhether image-derived welding features can be used to support data-drivenestimation of the physical state of the welding process. High-speed imaging was usedto quantify geometric process features, which were then synchronized with electricalmeasurements to create a dataset for training and evaluating soft-sensing models.Features related to arc behavior, droplet motion, weld pool position, and wire geometrywere extracted from high-speed video. Two imaging configurations wereevaluated: laser-lit and back-lit imaging. Laser illumination improved the visibilityof fine structures but introduced strong reflections and artifacts that reduced featureextraction reliability. In contrast, back-lit imaging produced high-contrast silhouetteswith cleaner object boundaries, making it more suitable for robust geometricfeature extraction.A computer vision pipeline was used to extract geometric features, including wireedges, contact tip, wire tip, and weld pool position. Additionally, droplets and spatterswere detected and tracked using blob detection and simple motion rules. Theseimage-based features were then synchronized with electrical signals and process parametersto build a time-series dataset.Several statistical, machine learning, and deep learning models were evaluated forpredicting image-derived features from electrical inputs, primarily voltage, current,and wire feed speed, along with additional engineered features.Overall, the results show that electrical signals contain information related to key geometricwelding features, but the estimation accuracy remains limited. The modelsgenerally capture the cyclic patterns and overall trends in the signals, but strugglewith amplitude variations between sequences. The results support the feasibilityof soft sensing for GMAW-P monitoring, but not yet the accuracy and robustnessnecessary for reliable adaptive control.