System Identification of an elastomeric series elastic actuator using black-box models

Flexible manipulators are the core technology for the development of collaborative robotic systems, which is a trend in Industrial Robotics. Hence, the system and parameter identification of these systems are important to develop more accurate strategies of control based on the dynamic behavior of these joints. The purpose of this paper is to build linear and nonlinear auto-regressive models with exogenous inputs based on experimental data. The data set used was collected with an experimental setup using a multi-sine torque input signal applied to the motor. The output was the angular velocity of the link. The models employed were the ARX, ARMAX, NARMAX, and NARX-NN. Futhermore, their performance was measured using the root-mean-squared error and the R
²
score. The results show that all the models presented good performance, with an acceptable R
²
score and an RMSE value close to each other.