This paper aims to propose a novel methodology to mitigate welding torch vibration caused by sensor measurement noise during laser vision-based seam tracking, which compromises robotic welding quality.
A Kalman filter is applied to smooth the noisy sensor data for precise tracking. To avoid costly and inefficient manual tuning of the filter’s hyperparameters, Bayesian optimization is utilized to approximate optimal settings with minimal experimental iterations. Its efficiency is validated through comparative experiments with grid search and random search.
Experimental results demonstrate that Bayesian optimization identifies a configuration meeting the accuracy requirement within 3 iterations, compared to approximately 20 iterations for grid and random search. This achieves over 80% greater efficiency, significantly reducing tuning time and cost.
This paper proposes a sample-efficient method that ensures effective vibration suppression and tracking accuracy while significantly reducing optimization iterations. The approach offers a practical solution for the efficient calibration of welding robots and other high-cost control systems.
