L. Schnitman (Brazil), J.A.M. Felippe de Souza (Portugal), and T. Yoneyama (Brazil)
Constrained control, Nonlinear control, Neural control
This work presents the design and implementation of a controller for nonlinear, unstable and constrained systems. For instance, a magnetic levitation system is selected to highlights the controller properties especially with respect to stability and constraints satisfaction. The control action is based on the reference governor (RG) approach that uses a Lyapunov's concepts of energy to prevent constraints violation both on the state and on the manipulated variable even during large changes in the reference signal. To design the RG an inner loop controller is proposed. The RG receives the system states and the desired reference to compute and supply the reference signal to the inner loop controller. This procedure guarantees the stability while avoiding constraints violation. The paper describes in full detail the inner loop controller and RG design, but in spite of the successful results, the RG approach is not able to treat uncertainties. Here the authors propose to replace the whole linearization and control action for a single Artificial Neural Network (ANN). The ANN is trained using the system states and desired reference as input and the nominal control action as the output. Training data is generated thought simulation based on the action of the designed inner loop controller with the RG. The major objective in the use of the ANN is may also be able to treat uncertainties and allows straightforward implementation of training techniques to further provide adaptation capabilities.
Important Links:
Go Back
