Parameter identification methods for real redundant manipulators

Abstract

This work presents the development, assessment and comparison of four techniques for identifying dynamic parameters in an industrial redundant manipulator robot with 5 degrees of freedom. Based on the Lagrange–Euler formulation, a linear model of the robot with unknown parameters is obtained. Then, these parameters are identified using the following techniques: least squares, artificial Adaline neural networks, artificial Hopfield neural networks and extended Kalman filter. The parameters identified are validated by using them for computationally simulating the performance of the redundant manipulator robot, to which are imposed reference trajectories different from the ones used in the estimation. To relate the trajectories performed by the redundant manipulator robot with the estimated parameters, the following error indexes are calculated: Residual Mean Square, Residual Standard Deviation and Agreement Index. Finally, to determine the sensitivity of the model identified – due to the variations of the estimated parameters – a new simulation is conducted on the robot, considering that its parameters vary in a restricted range. In addition, the RMS error index of the trajectories performed is determined. After this step, the parameters of the redundant manipulator robot were successfully identified and, thus, its mathematical model was known.