Aircraft Icing Information Processing Based On Artificial Neural Network

Although dozens of years there had been problems Aircraft Icing, and the Icing Protection System improved, large amounts of aircraft icing issues may contribute to flight accidents as ever. Therefore aircraft ice protection technique need to be enhanced substantially, in which aircraft icing detection technology is the key. Icing detection in advance is especially of great importance for pilots to take early measures as encountering icing.The thesis introduces diaphragm type sensor based on piezoelectric effect and resonance, including both of its static and dynamic characteristics. It makes the breakthrough by the identification of latent heat and carries through feasibility analysis of mathematical modeling and artificial neural network modeling. The thesis describes BP (back-propagation) network modeling process and optimization and then based on the characteristics of the temperature to achieve recognition of the sensor system on the action level of modeling. Designs using the toolbox of MATLAB neural networks, position-mode as well as increment-mode, the two sample programs are compared and then their respective advantages and disadvantages are given. In addition, a practical embedded system is designed by using method of neural network software.Simulation results show that neural network modeling program is feasible, and the generalization of the neural network capacity fault-tolerance features enable it to produce reasonable output. These two information-processing capabilities allow neural network solve some complex issues. BP neural network to the ice sensor information identification is only preliminary work. Feature extraction with unsupervised network of online learning will be the future direction of development.