The objective of this research is to apply infrared sensing techniques, artificial intelligence, and robotics to improve the welding process by on-line identification and mitigation of weld plate surface contamination. A method of in-process detection of surface contaminants during the gas tungsten arc welding of steel plate has been developed, and rudimentary corrective actions have been implemented to demonstrate closed-loop control. The study employed on-line IR sensing techniques to dynamically monitor the thermal field in front of the molten pool during the welding process. Changes in thermal pattern and in the area under thermal scans taken perpendicular to the weld seam were then used by the controlling computer to identify undesirable surface contaminants. Appropriate corrective actions were generated and employed to displace the contaminants from the welding path. A computer routine was developed that recognized changes in the thermal patterns due to surface contaminants and implemented corrective procedures. The results of this study will aid in the elimination of weld defects due to surface contamination and hence will increase the reliability and productivity of the welding process.