Identification of Friction in the Joints of an Experimental Manipulator with 3-DOF Based on MLPnetworks

The relevant article is aimed at creating a dynamic model of friction in the joints of the investigated system with 3-DOF, which is driven by FESTO fluidic muscles, based on MLP networks. For their creation, data obtained on the basis of the measurements was used. During the measurements, the rotation angle of Joint 2 and Joint 3 was monitored during the movement of the manipulator under the influence of gravity and propulsion using fluidic muscles. For the need to simulate the movement of the manipulator, a scheme was created in the Matlab Simulink environment, which represents the behavior of the real system. After the creation of the simulation scheme, the process of estimating the parameters of the original friction model followed, based on the comparison of the measured and simulated course of the rotation angle of Joint 2 and Joint 3. The individual inputs to the MLP network subsequently represented the values of acceleration, speed, and deflection angles of Joints 2 and 3 generated by the dynamic model at the specified excitation methods and the use of the original, estimated friction model. After the creation of the candidates MLP networks, the dynamic model of the investigated system was modified, replacing the original friction model with the corresponding MLP networks, for the subsequent process of validating the complete dynamic model based on the proposed statistical indicators in order to determine the most efficient candidate MLP network.