INTRODUCTION: This work focuses on describing the use of a machine learning algorithm for anomaly detection on an electrical flight control system of an Airbus aircraft. Data used in this work was obtained from the industrial benchmark problem proposed at the IFAC 2020 conference. The Flight Control System (FCS) is a group of machine-driven and electronic equipment which allows an aircraft to fly with remarkable accuracy and constancy, it is one of the most critical systems inside an aircraft. The FCS controls the individual cockpit surfaces, connecting linkages, and the mechanisms to coordinate an aircraft’s direction on the wing, the control ranges, and even on the aircraft’s engine, as it can modify the aircraft speed. In recent years, with the rapid advancement of aviation technology, the FCS is driven by an electrical and computerized system called fly-by-wire, significantly providing enhancement to the performance of the aircraft. But as the technological improvements assist with the safety of the aircraft, it also involves failure that is very likely to provoke harm to pilots and passengers. Therefore, an effective fault detection method must be established to provide safety and reliability for the aircraft. – OBJECTIVES: It is proposed the implementation of machine learning techniques for the aim of anomaly detection in aerial systems, in order to improve the performance of the overall control system. – MATERIALS AND METHOD: This paper works with Naive Bayes, Decision Tree, K-Nearest Neighbors, and Random Forests classification algorithm. The problem approached in this paper is to identify a malfunction of an aircraft operation system. In order to do that efficiently, the system must be precise, fast, and have good efficiency. Also, as required by the competition, it must be simple to implement. It will be shown a variety of ways to combine different machine learning techniques to successfully deal with anomaly detection on an electrical flight control system of an Airbus aircraft. – RESULTS: Comparing all the results, it is possible to see that the best accuracy score reached is 70%, obtained mainly using the Decision Tree and Random Forest models. The result had a small variance reaching an average of 67% using the k-Nearest Neighbors, and the Naïve Bayes model was able to reach a result of an average of 55%. Additionally, we were able to see that the normalization technique used has a great impact on the script run-time. – FINAL CONSIDERATIONS: This project implemented various machine learning techniques to reach a good result on an aircraft anomaly detection problem. The outcome shows all possible result variations, including the average accuracy score, precision, recall, and F1, together with the result considering different numbers of features and the script run-time of each one. This project can be used on different kinds of aircraft, providing a relatable score of malfunction detection and, furthermore, providing a fast and efficient method to be used on the aircraft software.
