Learning-based clustering for Flight Condition Recognition

Abstract

This paper presents flight condition recognition (FCR) algorithms for rotorcraft health and usage monitoring systems (HUMS), which are developed by using the clustering techniques of machine learning. Training and validation dataset are generated by using a generic nonlinear helicopter simulator and several flight data are obtained to train the algorithm. Gaussian Mixture Model (GMM), Neural Networks (NN) and Logistical Regression (LR) algorithms are implemented to perform FCR analyses. Validation and comparison studies are performed and results are compared in terms of accuracy, execution and training time. Finally, a detailed flight report about the flight is provided with percentages of performed flight conditions, which is used to provide feedback for health and usage monitoring systems to predict the life of the aircraft components.