Signals Estimation of Force Sensor Attached at Manipulator End-Effector Based on Artificial Neural Network

In this chapter, the multilayer feedforward neural network (MLFFNN) is used for the estimation of the force sensor signals attached to the end-effector of a 2-DOF planar robotic manipulator and the external torque of the manipulator’s joints. The MLFFNN is designed by depending on the only signals of the position sensors of the manipulator’s joints as its inputs, its outputs are the force signal, and the external torques of both robot joints. The experimental work is executed by commanding the robotic manipulator to perform the sinusoidal motion. During this motion, the human hand carries out some random collisions on the force sensor. Data is collected from these experiments and used for the training stage then the test and the verification stages of the developed MLFFNN. The MLFFNN training happens by the algorithm of Levenberg-Marquardt, and its best performance is obtained considering very small mean squared error (MSE) and training errors. The results reveal that the trained NN efficiently estimates the force sensor signals and the external joint torques, correctly.