Stochastic parameter identification based on dynamic/static sensitivity analysis for feed-forward control of robot periodic motion

Abstract

In robotic thick plate welding, the welding strength is ensured by periodically oscillating the welding torch attached to the end-effector in the horizontal direction, which is called weaving motion. Because the vertical accuracy of the end-effector affects a quality of the welding, it is necessary to reduce vertical error of the end-effector. In this paper, we focus on periodic motions, such as the weaving motion, and consider the suppression of vertical errors by feed-forward control. To calculate the feed-forward torque, it is necessary to identify the dynamic parameters of the robot. However, since noise and un-modeled dynamics are included in the experimental data, we identify parameters that have small influence on the motion error even though the existence of the parameter errors by considering the parameters as stochastic variables and identifying their error covariance to be the reference one. The reference error covariance is calculated from the sensitivity analysis of the vertical direction of the end-effector in the periodic motion with respect to the minimum set of dynamic parameters. Sensitivity of the dynamic/static motion, and sensitivity of feed-forward control are employed to improve the stability of the iterative calculation. By experiments using the planar 3-link manipulator, we confirm that vertical errors of the weaving motion are reduced by the feed-forward torque using the proposed method compared to the Least Mean Square.

Sequential updating of minimum set of dynamics parameters by stochastic identification

“…In this section, we summarize our conventional stochastic identification method (Okada et al, 2022). The equations of motion of the robot are summarized as linear equations with respect to minimum set of dynamics parameters p as;…”

“…In this experiment, Σ r is determined based on sensitivity analysis (Okada et al, 2022). Let consider the robot dynamic equation represented by the following difference equation around the reference motion q r k .…”

“…( 1 ) The effectiveness of the stochastic identification has been experimentally evaluated in (Okada et al,2022). The results in Fig.…”

“…Therefore, the data used for identification should be acquired in the same environment and in the same motion in which the robot is operated (Otani et al, 1992;Villagrossi et al, 2013;Pedrocchi et al, 2014), however, this is difficult due to un-satisfaction of PE (Persistently Exciting) condition. The authors have proposed stochastic identification method to obtain suitable parameters for use in control system design from arbitrary motion data (Okada et al, 2022). Based on the sensitivity of the motion with respect to parameters, the important parameters are precisely, not important parameters are roughly identified, which means the parameters are identified from arbitrary data by WLS as if it is acquired from the motion in which the robot will be operated.…”