Recognition of Aircraft Maneuvers Using Inertial Data

The report is devoted to solving the problem of maneuver type determination for aircraft flight recordings using machine learning methods. The purpose of the study was to obtain real flight data with labeled maneuvers for use in developing motion cueing algorithms. These algorithms are supposed to be used in training simulators. During the flight synchronized inertial and video data were recorded. It was decided to use machine learning methods to label these records because the manual markup process takes a lot of time, requires pilots’ consultations, and is very difficult if there is no video of the horizon overboard the aircraft. Markup during the flight also is not an option since this will either interfere with the pilot and can lead to an accident, or there must be another person in the cabin who is capable of marking in real time, which is quite expensive. Both video and inertial data were used to manually label the used data, but only preprocessed inertial data was used to train the classifiers. The article compares 10 different types of classifiers which determine the type of maneuver being performed from inertial data recorded during 10 s of flight.