A Comparative Study on Air Data System Calibration Methods based on Machine Learning

Aircraft control system contains lots of complex components to ensure a safe flight. Air data system is one of them, which measures flow fields near the flying aircraft and derives vital information including Mach number and altitude. It is evident that the system contains measurement errors due to several reasons such as design limits and measurement uncertainties. Considering that the incorrect information might cause unexpected and harmful consequences, the air data system must include a calibration process to compensate for the errors. Although the mainstream models of calibrating air data system are based on lookup tables and polynomials, machine learning based approaches have been actively conducted and produced positive results. This paper focused on the machine learning techniques which satisfy the universal approximation theorem such as neural network, radial basis function neural network and support vector regression. By using actual flight test data, the machine learning-based calibration methods proved its remarkable performances.