A technique for inversely identifying joint stiffnesses of robot arms via two-way TubeNets

Abstract

Joint stiffnesses of robot arms play a critical role in the control of the posture and movement of the arm tip. This work develops a systematic approach for inverse real-time quantitative identification of the stiffnesses of joints for robotic arms using the TubeNet proposed by Liu. To start with, a finite element (FE) model for a six-axis tandem robot arm is established. Experiments are then conducted to measure the first few lowest natural frequencies of the robot arm to be compared with numerical results for the validation of the FE model. Using the validated FEM model, sensitivity analyses of the joint stiffnesses to the natural frequencies are carried out to ensure sufficient sensitivity for inverse analyses and a neural network data set is established. The selection of appropriate TubeNet layers and activation functions is exposited. Subsequently, the direct-weights-inversion (DWI) formulae for the TubeNet is adopted to inversely compute the joint stiffnesses explicitly in real time. The predicated joint stiffness using the currently proposed DWI formulae of the TubeNet is accurate with the maximum root-mean-square of test errors less than 0.0020 N·m/rad.