Abstract
An adaptive control method using a self-organizing map (SOM) which is a kind of neural networks is proposed to suppress vibration of a cantilever plate. The SOM learns characteristics of given data without supervising and categorizes high dimensional data into low dimensional maps with keeping complex relationships among data. The present method employs the SOM to estimate states of the controlled object and makes a lookup table of control input. Each node of the state estimation SOM consists of vectors including current and past controlled responses and control inputs. The SOM and corresponding lookup table learn and are updated only when the effective control input is applied to the controlled object where the evaluation function of control system is defined by differences between the desired and current state. The present method just requires information about the control response and input to build the statistical model on site, resulting in implementation of the vibration control without numerical models in advance for controlled object. Numerical and experimental results are given for the cantilever plate fabricated by an aluminum plate and a piezoelectric actuator, and the effectiveness of the present method is successfully confirmed from both results.
