Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : May 10, 2017 - May 13, 2017
In a conventional balance control method, the target object is modelled and the control parameters are determined. However, if there are changes in the model parameters, such as weight or length, after determining the control parameters, the control may become unstable. To solve this problem, we propose a method to adaptively adjust the control parameters during operation even if the quantitative model of the control target is not given. As a typical example, we chose the case of stabilization an inverted pendulum using a flywheel, which is a classic problem of balance control. The time series data of the state of the pendulum of different sizes were obtained and used to train the neural networks to obtain optimum control parameters. The usefulness of the proposed method through simulations.
