Abstract: The optimal estimation algorithm for real-time welding deviation based on Kalman filtering is presented.The state equation and measurement equation for detecting the weld position is established, and the optimal estimation of the Kalman filtering recursive algorithm also is established according to the principle of minimum mean square error.Measurement noise covariance is obtained from the statistical value of measurement error, and after the process noise is supposed to derive from the changes in acceleration, the initial values of the welding center position are determined by the two-point method.During the welding process, the welding position is accurately predicted while the noise interference is eliminated by Kalman filtering.The computer simulation and experiment results show that the weld deviation signal processed by the Kalman filtering can eliminate the disturbance of causal factors and random noise, improve the tracking precision and the stability of system, and be suitable for the practical engineering applications.