Gain tuning using neural network for contact force control of flexible arm

Abstract

This paper presents contact force control of a one link flexible arm consisting of a simple boundary feedback of bending moment at the base of the flexible arm. Gain tuning control system using neural network was developed and its control performance examined and compared with fixed gains by numerical simulation and experiment. In this work, feedback gain was tuned to correspond to the coupling coefficient of the neural network, and stabilized the learning by giving the initial value to the coupling coefficient of the neural network, thereby shortening the learning time. To adjust the gain value in real time, a sequential correction type technique(online learning) that repeats learning at every sampling was adopted as the learning scheme of the neural network. As a result, it was confirmed that by using the neural network, the value of the feedback gain was adaptively changed and the target contact force converged after 0.35 seconds. Comparing the performance with that obtained with fixed gain, it was found that neural network tuned controller took a shorter time to converge to the target value by 0.8 seconds confirming that the proposed controller is more effective for the contact force control of the flexible arm.