Predictive controller design with offline model learning for flexible beam control

Abstract

This paper presents the method of flexible beam control using model predictive control methodology, including discrete time-regression predictive model of the forward dynamics of the plant based on linear neural network. The model is parameterized via offline learning on the experimental data, obtained from the plant. The method of determining the tolerant prediction horizon value is proposed for tuning the controller. The control system structure for flexible beam is considered using two predictive controllers via hub angular velocity control loop and flexible beam angular tip position control loop. The control system is tested with both the computer simulation with MATLAB and controlling the laboratory model of the planar single-link flexible manipulator. The test has proved the effective vibration suppression and tip force disturbance rejection of the flexible beam.