Infrared (IR) quality assessment of robotized resistance spot welding based on machine learning

Resistance spot welding for industrial applications is a fully automated process which however lacks this aspect when it comes to quality assessment (QA). The current study introduces an online QA approach that utilizes machine learning methods, on data captured from an infrared (IR) camera, mounted on an RSW-robotized system. The models' development was carried out in the context of two experimental approaches which included a different number of process parameters and weld-quality criteria. Results indicated that the quality-prediction uncertainty of the models depends on the proximity of the points of the process parameter space. The maximum IR intensity and the temporal features of the IR cooldown profile offered the greatest class separability and marked the corresponding classifiers as the most successful ones. In addition, the advantage of IR monitoring was highlighted, and it was justified that in an industrial scenario where the time between welds is short, the IR data from such a monitoring system may include limited temporal information, which depending on the model used could potentially compromise the model's prediction performance.