Abstract: Based on researches on scanning signal of existing arc sensors,a kind of oscillating rotating arc sensor was put forward.After analyzing the signal characteristics of the new arc sensor,the mathematical relationship between spatial attitude of welding torch and the change of arc length was deduced,and a mathematical model was established for the arc length.Considering the characteristics and complexity of welding arc signal,the Gabor wavelet filter was used to eliminate the noise interference, and the feature vector was extracted from the sample data after filtering,and the dimension of data was reduced.The classifier of welding torch attitude according to the principle of Euclidean distance,real-time identification of the spatial attitude of welding torch was conducted with the steepest descent method of model parameters.The results show that the algorithm for recognition of spatial attitude of welding torch with swing of rotating arc sensor was simple and provided theoretical basis for improving the precision of welding seam tracking.