Modelling and open-loop control of a single-link flexible manipulator with neural networks

This paper presents an investigation into the development of open-loop control strategies using shaped command inputs with neural networks for vibration suppression in flexible manipulator systems. The method requires, at the first stage, that the natural vibration frequencies of the system be determined. A non-linear AutoRegressive process with eXogenous inputs (NARX) model is used with multi-layered perceptron (MLP) and radial basis function (RBF) neural networks for this purpose. Shaped command inputs including low-pass and band-stop filtered torque input functions are then developed based on the identified modes of vibration and applied to a single-link manipulator in an open-loop configuration to suppress the system vibrations. The performance of die control strategy in suppressing the system vibrations is assessed in comparison to a bang-bang torque input. Finally, a comparative performance assessment of the low-pass and band-stop filtering approaches in suppressing system vibrations is presented.