Study On Airfoil Ice Accretion Shape Prediction Based On Neural Networks

Airfoil icing does great harm to flight safety. One of the most important reflections ischanging the shape of original aircraft aerodynamics. For the reason of the decrease ofaircraft’s aerodynamic performance, the accretion influences the flight safety. Study on airfoilicing is significantly meaningful, therefore this paper mainly focuses on the prediction ofairfoil icing shape.In this paper, function fitting for the complexity ice shape curve was built. Based on thefitting function, BP neutral networks model was established to predict the airfoil icing shape.The paper comprehensively reviewed of icing research background to show the necessity ofaircraft accretion research. It outlined three stages of icing research progress (i.e. the 1920s to1978, 1978 to 1994 and 1994 so far) and research results of each phase. Further it elucidatedthe significant meaning of researching airfoil icing by a brief overview of the general weatherconditions causing aircraft icing, then analyzed the harm of icing to flight and the reason ofhazards. This paper also gave examples of flight accidents and incidents caused by aircraft iceaccretion. By extensive reading and studying, this paper summarized and analyzed the fourmethods of aircraft ice accretion research based on neutral networks at home and abroad.Then it pointed out the shortcomings existed in each method and discussed the progress andtrend of the research. By reviewing past researches, this paper outlined six key parameters ofthe ice shape and discussed the impact of each parameter on the ice shape as well as theresearch methods. Finally, according to function fitting principle, a multi-valued function(sine function superposition) was got which fitted the complexity ice shape curve well.Considering the complexity and nonlinearity of the ice shape, a prediction of airfoil ice shapebased on BP neutral networks was carried out. The reliability of the prediction which is got byusing neutral networks was verified well in the way of sensitivity analysis.Results show that the multi-valued function fits the nearest degree of the complexity iceshape curve. The thickness, horn character and range of ice shape predicted by BP neutralnetworks basically agreed with the original ice shape.