Neural Network Sensitivity to Inputs and Weights and its Application to Functional Identification of Robotics Manipulators

Abstract

Neural networks are applied to the system identification problems using adaptive algorithms for either parameter or functional estimation of dynamic systems. In this paper the neural networks' sensitivity to input values and connections' weights, is studied. The Reduction-Sigmoid-Amplification (RSA) neurons are introduced and four different models of neural network architecture are proposed and analyzed. A two degree-of-freedom manipulator is considered as a case study and the functional dynamics for computed torque are identified using the proposed models. The simulation results are studied and analyzed for different models.