Online artificial neural network model-based nonlinear model predictive controller for the meridian UAS


Correspondence to: Gonzalo A. Garcia, Aerospace Engineering, University of Kansas, Lawrence, KS, 66045, USA.



A worldwide accident survey of manned and unmanned aircraft shows in-flight loss of control remains a major contributor to aircraft accidents. Operation outside the normal fight envelope is usually subject to failure of components, inappropriate crew response, and environmental conditions. The aerodynamic model of aircraft associated with hazardous weather and abnormal conditions is inherently nonlinear and unsteady. A novel adaptive nonlinear model predictive controller is proposed and conceptually proven to ensure safe control of the Meridian unmanned aerial system in off-nominal conditions. Nonlinear model predictive controllers are capable of handling input and output constraints that directly satisfy safety and performance requirements in off-nominal conditions. The performance of a nonlinear model predictive controller relies heavily on the physics-based model. Controller performance is improved by updating the physics-based model by using real-time nonlinear estimation of aerodynamic forces. Real-time nonlinear estimation of aerodynamic forces is obtained utilizing artificial neural networks with time-varying adaptive parameters. The adaptive nonlinear model predictive controller, coupled with real-time parameter identification, is shown to exhibit robust characteristics and to successfully mitigate the impact of nonlinear and unsteady aerodynamics while preventing loss of control. Copyright © 2013 John Wiley & Sons, Ltd.