Data-driven failure identification using a motion-based simulator

Abstract

© 31st Congress of the International Council of the Aeronautical Sciences, ICAS 2018. All rights reserved.This paper presents a data-driven failure identification of flight control surfaces using neural networks. Experiments were performed in a motion-based flight simulator (SIVOR) that has been developed at Aeronautics Institute of Technology (ITA). We use a two-layer feed-forward network and we analyze the influence of the input parameters and the number of neurons in the hidden layer on the performance of the failure identification task. The evaluation of the neural network's performance is based on overall accuracy, training time, number of iterations, precision and recall. Best results were found for networks with 100 neurons in the hidden layer, presenting 97.2% of overall accuracy.

Keywords