Vibration Control for a Pendulum System with a Movable Weight along a Horizontal Track Using a Neural Network Type Controller
active vibration damping
specific damping capacity
A parallel-type learning controller, in which a three-layer neural network (3NN) is connected in parallel with a pole allocation (PA) controller, is presented as a method for actively controlling the vibration amplitude of a pendulum system. The controller utilizes a movable weight that is actuated along a horizontal track, the center of which is situated at the pendulum pivot. The 3NN is specially designed to have its error function Jn=<u1u1>^^^T based on the PA controller output u1, with the result that the object plant input, u=u1+u2 , becomes the 3NN output u2 with an increasing learning cycle. The proposed 3NN control system, including an N-dimensional observer , is not only capable of realizing the active vibration suppression effect maximally as specific damping capacity (SDC)?0.065 (ca. 25 times that of free vibration damping) , but also exhibits a markedly more robust vibration damping effect than the conventional PA control system.
Departmental Bulletin Paper