磁気粘性流体を用いた可変慣性モーメント型振動抑制装置（強化学習による慣性接続質量切替時の地震波加振実験）

Abstract

In previous paper the authors successfully developed a vibration suppression device (VSD) that has variable moment of inertia using Magneto-Rheological (MR) fluid. The MR fluid is filled into a flywheel. Ferrite particles of the MR fluid are clustered when magnetic field is applied to the flywheel by electromagnets, series inertia mass of the VSD will be switchable. In this paper seismic tests during switching series inertia mass by reinforcement learning are focused in order to get more vibration reduction while random excitations. Firstly, agents are trained by the reinforcement learning using Deep Deterministic Policy Gradient (DDPG) algorithm while a one-degree-of-freedom system with the VSD, which has the switchable series inertia mass, is excited under earthquakes. Secondly, the VSD is improved for having 8 electromagnets, magnetic field is analyzed by FEM, resisting force characteristics and response performance of the VSD are investigated. Then, to confirm the vibration control under some earthquakes, seismic tests of one degree-of-freedom system with the VSD under the reinforcement learning using DDPG are carried out by a shaking table, and seismic responses are simulated numerically. From the experimental results, the peaks of response acceleration and displacement decrease about 1/2 in case with the VSD using DDPG, and also cut 20 % by switching series inertia mass. Finally, validity of vibration reduction during switching series inertia mass by the reinforcement learning is confirmed.