Flight Maneuver Classification Using Artificial Neural Networks

Abstract

There are globally known basic flight maneuvers performed by fighter and aerobatics pilots with agile aircraft. These basic flight maneuvers have strict rules that the pilots should follow to complete the maneuver and can be used to evaluate the pilot’s or aircraft’s capabilities. Flight data is classified into flight maneuvers by aviation professionals before evaluation. The need for this classification created a research area to be filled: Automatic flight maneuver classification. This study proposes a solution to the automatic flight maneuver classification problem by exploiting artificial neural networks. It also contributes a flight maneuver classification dataset to the literature. This dataset was generated using professional flight simulation tools. The flight data attributes were evaluated and selected to give the optimum performance in terms of accuracy, precision, and recall. The types of artificial neural networks used and compared were single hidden layer neural networks, deep neural networks, and recurrent neural networks. Combinations of these types, activation functions, optimization methods, and gradient descent algorithms were tested against the problem to maximize the performance of the solution. There are no secondary studies on the subject of flight maneuver classification and this study also fills this gap by contributing a systematic literature review on ”flight maneuver classification using machine learning imethods”. The solution proposed by this study successfully classified ten distinct flight maneuvers such as basic descent and ascent, but also more complex ones such as Immelman, split-s, and lazy-eight. The results were evaluated by calculating accuracy, precision, recall, and loss parameters on test data. The best-performing artificial neural network type, which is deep neural networks, gave over 96 percent accuracy and less than 0.1 loss in the test set. All the artificial neural network types and solutions gave over 90 percent accuracy with correctly chosen attributes. This study also contributed a software program that classifies the flight maneuver in real-time while the flight maneuver is being performed in a simulation. It was seen that this software was also accurately predicting the pilots’ intended maneuvers in real-time.