Abstract: Objective The TIG (Tungsten inert gas) welding process is affected by factors such as tungsten electrode wear and thermal deformation, resulting in changes in the welding height during the welding process. At the same time, there is an error in the cutting size of the previous welding process, resulting in an uneven welding surface. These factors easily cause changes in the arc length during the welding process, seriously affecting the welding quality. Methods In order to improve welding stability, this paper adopts the arc voltage sensing method to collect arc length information, and proposes a TIG welding arc voltage preprocessing method considering the alternating peak base value of TIG welding. Arc voltage and arc length collection experiments are carried out to determine the arc voltage-arc length function relationship. Furthermore, in order to overcome the noise interference during the arc voltage collection process, this paper uses the Kalman filtering method to estimate the TIG welding arc length, establishes a mathematical model of the TIG welding Kalman filtering, and selects the Kalman filtering parameters through comparative experiments. On this basis, a variable step TIG welding arc length control method based on Kalman filtering is proposed, which accurately calculates the arc length adjustment step and controls the arc length during the welding process. Results The arc length-arc voltage experiment shows a significant linear relationship between arc length and arc voltage, providing a basis for arc length estimation. The arc length control experiment shows that through the proposed variable step arc length control method based on Kalman filtering, the standard deviation of arc length in the welding process is reduced from 0.48 to 0.305, with a reduction of 36.5%. Conclusion The experiment shows that the Kalman filtering method proposed in this paper can effectively estimate the arc length in the welding process, and the stability of TIG welding arc length can be effectively improved through variable step control.