Research On Aircraft Icing Environment Based On Electromagnetic Simulation And Machine Learning

Ice accumulation is one of the common safety problems of aircraft.It is not noticeable at the beginning of the icing but once it reaches a certain thickness it is immediately serious and difficult to remove.The purpose of this study is to use electromagnetic simulation software and machine learning methods to study the aircraft flight environment.The research contents and conclusions are as follows:1.Ice airfoil aerodynamic coefficients of the simulation show that: compared with the clean airfoil,icing airfoil stall Angle of attack and lift coefficient were reduced,drag coefficients increase.The different positions of the upper and lower surfaces have little effect on the lift coefficient,but the drag coefficient increases obviously,and the drag coefficient of ice accumulation on the upper surface is larger after the stall.Slight ice accumulation at the leading edge makes the drag coefficient significantly larger.The lift-drag coefficient of wedge ice before stall is similar to that of clean airfoil and the lift-drag ratio of double Angle ice and mixed ice is extremely low at any Angle of attack.2.The scattering characteristics of various ice pack particle swarmed are calculated by using XFDTD software.According to the scattering cross section,the seven snowflakes are divided into flat snowflakes,branching snowflakes and intermediate snowflakes.The backscattering cross sections of the particle swarm are all larger than the simple addition results of the spherical particles,but the different incidence and polarization directions of the electromagnetic wave will result in significantly different backscattering cross sections.The backscattering cross section increases as the melting snowflake transforms from solid to liquid.The backscattering cross section of supercooled water/ice droplet increases with the increase of equivalent diameter and decreases with the increase of temperature,but has no obvious law with the change of incident frequency.3.Based on the observation data of FY-2 satellite,millimeter-wave radar and microwave radiometer,the macroscopic and microscopic characteristics of cloud were used to qualitatively analyze the icing environment.The satellite data were used to preliminarily judge the main icing threats in the middle and upper troposphere.Three fuzzy logic algorithms with different weights show that the depolarization ratio has an important effect on the classification results.The neural network algorithm can accurately distinguish the existence of ice particles,and it is more accurate to judge the small-size particles,but it is difficult to judge the large-size particles.4.Based on existing icing regions in China and Icing Index,the sounding stations in different icing regions and before and after precipitation process were analyzed.The icing distribution in different regions,seasons and heights varied greatly.Before the summer rainfall in Northeast China,there was icing at the height of 4km to 6km,and the intensity might be related to the rainfall.In winter,snow on the ground in eastern China is easy to cause low altitude icing and its intensity is greater.Using a variety of machine learning and statistical method,from the perspective of qualitative and quantitative diagnosis of aircraft icing conditions,using principal component analysis(PCA)to improve equation significance and reduce the noise data,clustering analysis can be ice sample proportion increase,the diagnosis results of the two neural networks are similar,but it is difficult to calculate the Icing Index quantitatively.