Abstract: This paper use BP network to model the flush airdata sensing system instead of using aerodynamic model. In connection with the characteristics of FADS system, a neural network architecture with two hidden layers and its fast algorithm are designed. Not only the method of producing training patterns but also the technique of compiling training data are discussed in detail. As a result,the absolute mean errors of static pressure and impact pressure are all less than 130 Pa in both the trained area and the nearby area, which can meet the requirement for the design of the FADS system.

Key words: airdata sensing system, neural network, fast algorithm, training pattern