비선형 유연시스템의 변위저감을 위한 인공신경망 입력성형기

Abstract

In this paper, Input Shaper Command using Artifitial Neural Network was developed to control residual deflection when hoisting motion of the system occurs. Input Shaping technology has been applied to various industrial sites and flexible systems, but it is senisitive to modeling errors and is forced to move vertically, and there are problems of increasing working time and reducing efficiency. In addition, since the safety of workers is important, residual deflection should be small even in situations where movement in the hoisting direction occurs. Therefore, it is necessary to develop an input shaping command with less residual deflection even when modeling errors occur. First, the input shaping commands with arfitifial neural network was analyzed when the actual flexble system contains non-linearity. A network using modeling dynamics for the Double-Pendulum System and a tow-mode input shaping command was propased. In addition, a network that can be applied to the operation of the Tower Crane has been developed using a nonlinear shaper. Network’s training algorithm used to global optimal PSO algorithm, and the performance index was evaluated. It uses a new input shaper and compares it with a shaper that has secured robustness over a wide frequency range. The simulation was divided into large-scale and mini-scale, and experimentally evaluated and verified with a mini-scale system.