Detection of Aircraft In-flight Icing in Non-steady Atmosphere Using Artificial Neural Network

This paper attempts to research the issue of in-flight icing identification of aircraft flight dynamics. A nonlinear aircraft dynamics model is set up to simulate the wind turbulence effect on aircraft. The effect on flight dynamics by icing and wind disturbance are compared with clean one. In non-steady atmosphere, it becomes not so easily to detect. So a method using neural network and Kohonen self-organizing maps (SOM) to distinguish ice configuration form the clean model. Firstly, ANN models train on the aircraft dynamics for iced and clean aircraft in order to get the connection weights. The weights are used as input to SOM to identify the configuration as being clean or being iced.