Application of an Adaptive Clustering Network to Flight Control of a Fighter Aircraft. Phase 1

Abstract : An artificial neural net controller was developed for a typical fighter aircraft's longitudinal Stability and Control Augmentation System (SCAS) , using elevator and thrust-vector-angle controls, and operating at high angle of attack. The 'baseline' neurocontroller (NC) was in the feedforward loop (Kawato Type-C), with inputs from the pilot's pitch rate commands and rates and SCAS feedback error. An Adaptive Clustering Network algorithm was used to train the radial-basis-function neurons. Significant improvements in performance resulted from the NC action and these effects were analyzed by frequency domain describing functions. Thrust vector failures were handled satisfactorily, but reconfiguration of the SCAS was not possible within the simplified aircraft and NC effects. Phase II recommendations are included, such as: ways to choose signals for the neural net to more efficiently identify and separate failures of correlated control effectors; the further use of frequency domain describing functions to identity neurocontroller dynamic processes; and the development of a Neuro-controller Analysis Toolbox with diagnostic forcing functions, methods, analyses, and benchmark criteria for evaluation to a common NC standard.