Learning Feedback Linearization Using Artificial Neural Networks

This paper introduces a novel approach to the design of the feedback linearization called the learning feedback linearization (LFL) method. This method is used to learn the feedback linearization input for a nonlinear plant. The control scheme of the LFL method, which is devised by one artificial neural network (ANN) block, is based on a nonlinear auto regressive moving average model. The LFL method training phase uses not only input-output data pairs of the nonlinear plant but also a generated sequence data obtained from the states of the nonlinear plant. After the training phase of the LFL, a conventional proportional-integral controller is chosen as a linear controller for the feedback linearized closed-loop system. The performance of the developed ANN based LFL method is tested for tracking control problems via mean square error during the training and the test phases. The developed method is applied on both a nonlinear exponential plant and a well-known flexible joint mechanism plant in real-time simulation mode.