Detection and classification of aircraft icing using neural networks

This work introduces a neural network system that detects and classifies aircraft ice accretion in order to improve flight performance and safety. Neural networks are developed for use within an ice management system that monitors icing and its effects upon performance, stability and control. The ice management system would be used to automatically operate ice protection systems, to provide aircraft envelope protection and, when necessary, to adapt flight controls. A simplified longitudinal model of the DH-6 Twin Otter is used to study tailplane ice accumulation. An H^∞ parameter identification technique is applied to this model and relevant parameter estimates are used to characterize icing during a maneuver. Parameter estimates are used as inputs to a neural network system that both detects the presence of icing and classifies its severity. Extensive simulation results are presented that indicate the accuracy of the neural network detection and classification, even for modest levels of input.