Abstract: Vision and IR sensors of different types are in use for topside online monitoring of the surface geometry and temperature distribution of weld pool in GTAW and GMAW. An attempt has been made to develop a topside vision system for a welding robot in CO₂ welding process, consisting of CCD camera with IR band pass filter interfaced by video capture card. In the present robot vision system the interference of the arc radiation with that of the weld pool is inevitable and this leads to errors in the sensed parameters. The minimization of interfering radiation from the arc has been achieved by the usage of far IR band pass filter. Further to eliminate the interference of arc radiation on the weld pool it has been quantified by a constant called arc interference factor at various welding currents and plate thickness. Comparing the weld pool width measured from IR images taken online with arc and that with Genetic Algorithm-neural networks predicting the input variables for which the weld pool widths are found. The feasibility of usage of the vision system developed by using GA-NN has been studied.  The variation of arc interference factor with welding current is also studied. The predicted weld pool widths with vision system are reproducible and are in close agreement with the actual values.
Key words: Robot vision, CCD Camera, Image Analysis, Arc interference, IR band Pass filter, Arc radiations, weld pool width. Genetic Algorithm, neural Networks