Infrared image recognition for seam tracking monitoring during fiber laser welding

Seam-tracking ability of a laser-welding system is important for welding process, and the accurate detection of deviations between the laser-beam focus and the weld seam position is prerequisite for seam-tracking control. Infrared image sensing and visual recognition techniques for real-time seam tracking monitoring during high-power fiber laser welding is researched to improve the accuracy of seam-tracking ability. Molten pool images are caught by an infrared sensitive high speed camera arranged off-axis orientation of a laser-welding head which is fixed to a robot. Through the image processing, the feature detection of a near-infrared image is used in visual tracking. The gray-value gradient of near-infrared image is calculated and the keyhole margin of a molten pool is also detected. Combining the gradient and keyhole margin of a molten pool image, the thermal gradient parameter based on the thermal distribution of a molten pool is extracted. As a visual feature in robot control system, this parameter can be used to determine the deviations between the laser-beam focus and the weld-seam center. In comparison with direct detection of the narrow gap position, this parameter can be measured easily and the delay error resulted from the forerun of the sensor can be eliminated. This provides a practical approach to detect the deviations and the possibility to adjust the laser-beam focus position in real time, which can sensibly promote seam tracking accuracy. The proposed algorithm is tested during a butt-joint laser welding of Type 304 austenitic stainless steel plates at a continuous wave fiber laser power of 6 kW and 10 kW. Its effectiveness is confirmed by the welding experiments.