Non-Cooperative Full-Envelope Calibration of a Supersonic Air Data System

AirData Systems require accurate calibration to compensate for errors caused by the disturbed flow field over their pressure ports. If uncorrected, these pressure errors lead to errors in the aircraft's airspeed and altitude measurements, posing a hazard to flight. This paper proposes a novel calibration method that substantially reduces the flight time required to calibrate an aircraft's air data system by increasing analytical rigor. Traditional altitude and GPS-based calibration methods require minimal post-flight analysis, instead requiring numerous in-flight experiments to determine the system calibration. More recent dynamic calibration methods reduce the required number of dedicated in-flight experiments but still require the test aircraft to fly tightly scripted profiles to satisfy their analytical assumptions. This paper eliminates the requirement for scripted in-flight profiles by combining machine learning methods such as Gaussian Process Regression with an Extended Kalman Filter to produce a full-envelope Air Data System calibration. The proposed algorithm is validated using real-world T-38C flight test data and compared against existing calibration methods.