Position control of ultrasonic motor using support vector regression

The ultrasonic motor (USM) has excellent performance and many useful features that electromagnetic type motors do not have. It has been used in many practical applications. A characteristic of the USM that is affected by friction is strong nonlinearity, which makes it difficult to control. This paper proposes a position control method for the USM using Support Vector Regression (SVR), which is a regression method for Support Vector Machines. It is a newly proposed method of machine learning that does not have the disadvantages of a Neural Network such as a large number of learning times, local-minima, overfitting and so on. The proposed method uses an SVR controller combined with a PI controller. The SVR controller performs nonlinear input-output mapping of the USM. The learning of the SVR controller uses training data obtained from experiments. The effectiveness of the proposed control method is confirmed by experiments.