Research On Correction Method For Static Source Error Of Air Data System Of Civil Aircraft Based On Machine Learning

At present,the civil aviation industry in China is in the state of continuous development and expansion.Considering the situation demand,it is more and more important to master the key technologies of independent research of commercial airliners.Among all the technologies,design of air data system is a key link,which is related to the aircraft flight safety.In the phase of flight,the air data system measures the total and static pressure and calculates the important flight parameters for each system,which requires the air data system to measure pressure accurately enough.However,the influence of the static source error leads to the static pressure measurement is not accurate enough.Aiming at the static source error of civil aircraft air data system,the law of the static source error variation was analyzed,the error fitting model which is used to correct the static source error was studied.Based on the B777-200 aircraft model,the variation of the static source error with altitude,Angle of attack and Mach number was analyzed using the computational fluid dynamics simulation experiment method,and the effectiveness of the model and experimental scheme was verified according to the post-processing results.On the premise of verifying the validity,the static source error values at different altitude,Angle of attack and Mach number are extracted,The error regression model was constructed by using the data of simulation experiments combined with neural network method and stepwise regression method to correct the error.The experimental data show that the error correction scheme of stepwise regression and neural network method can better correct the static source error,and the error of airspeed calibrated with the corrected static pressure can meet the airworthiness standard which requires the error of airspeed is less than 3%,Furthermore the stepwise regression analysis model performs better than the neural network model in computational accuracy and speed.The results show that stepwise regression is easier to use and has better results than neural network method in solving the problems of fitting static source error with small samples and fewer sample characteristics.