Enhancing Flight Condition Recognition Performance Through Functional Similarity

Abstract

Flight condition recognition (FCR) is pivotal for aviation safety, enabling the recognition of the real usage spectrum, and thus, adapting and optimizing the maintenance schedule defined at design time. However, as the spectrum of flight regimes expands, recognition accuracy becomes complex. With increasing regimes, classifier representations overlap due to synthesized indicators summarizing multivariate time-series behaviors. Neglecting temporal dependencies and dynamic behaviors impairs classifier performance, particularly for specific regimes, limiting the expansion of the recognized set. In this article, we propose a functional similarity index to address this challenge, aiding analysts in identifying similar regimes and improving classification performance. To define such index, we rely on functional data analysis to capture regime dynamics, enhancing both performance and behavior analysis. The designed similarity index offers dual benefits: it investigates supervised FCR classifier outcomes and guides a closed-loop feature selection procedure to enhance recognition performance. This process eliminates the need for direct supervised classifier retraining, using the similarity index as a gauge to estimate the classifier's capability to distinguish critical regimes given a specific features set. When applied to extensive data collected during load survey flights performed by two Leonardo helicopter models, our method provides insights into results assessed by a state-of-the-art FCR supervised classifier. It identifies critical regimes and identifies informative features to include in the set to enhance recognition. This versatile framework has the potential to significantly enhance operational efficiency and aviation system safety, thereby fortifying overall operational capabilities.