Decentralized discrete-time neural control for a Quanser 2-DOF helicopter

• Published in: Applied soft computing 2012-08, Vol.12 (8), p.2462-2469
• Main Authors: Hernandez-Gonzalez, M., Alanis, A.Y., Hernandez-Vargas, E.A.
• Format: Article
• Language: English
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2012.02.016

[Display omitted] ► The decentralized control problem for output trajectory tracking in a Quanser 2 degree of freedom (DOF) helicopter is tackled. ► High order neural network is used to approximate nonlinearities. ► Neural networks are trained on-line with an extended Kalman filter based algorithm. ► The neural block control technique presents slightly better results than the neural backstepping algorithm. Control design for helicopters is a complicated and challenging problem due to the strong inter-couplings and nonlinear uncertainties in the system model. This paper deals with the decentralized control problem for the output trajectory tracking in a Quanser 2 degree of freedom (DOF) helicopter. High order neural network (HONN) is an important technique to approximate non-linearities in the model. Two different discrete-time schemes with a decentralized structure are used. Neural backstepping and neural sliding mode block control techniques are considered in order to control pitch and yaw positions. On one hand, backstepping control divides the whole system into two subsystems which are used to track the pitch and yaw references respectively. Real and virtual controls are approximated by HONNs. On the other hand, block control technique is applied to HONNs which can identify the system helicopter model. Each discrete-time high order neural network is trained on-line with an extended Kalman filter based algorithm. Without the previous knowledge of the plant parameters neither its model, we show via simulations the good performance of both strategies. The block control technique presents slightly better results than backstepping algorithm.