D.I. Soloway and P.J. Haley (USA)
Reconfigurable control, Predictive control, Aircraft control, Cooper-Harper, Pilot, Failure
This paper highlights a study that is part of a continuing research effort in reconfigurable flight controls at NASA Ames. Three NASA Dryden test pilots were tasked with evaluating two methodologies for reconfiguring an aircraft's control system when failures occur in the control surfaces and engine. Two neural network based controllers are being considered, specifically the Neural Generalized Predictive Controller and a Dynamic Inverse Neural Controller. Prior to the tests, the Neural Generalized Predictive Controller was enhanced with a simple augmentation to reduce zero steady-state error which unexpectedly made the neural network predictor model redundant and unnecessary for this level of reconfiguration. Instead a nominal single point linear predictor model was used with the new augmentation. This paper shows that even without a neural network the Generalized Predictive Controller performs equally well at reconfiguration as the Dynamic Inverse Neural Network controller, requires lower rates from the actuators, and avoids the validation dilemma of using a neural network. Also shown are the pilot ratings for each controller for various failure scenarios and two samples of the required control actuation during reconfiguration. Finally, the paper concludes by stepping through the Generalized Predictive Control's reconfiguration process for an elevator failure.
Important Links:
Go Back
