Exploring Acoustic Signatures of Different Aircraft Types and Operations Using Advanced Data Analysis

Abstract

Understanding acoustic characteristics of aircraft is critical for designing optimal fleet compositions in terms of noise and improved airport operations. This study investigates acoustic signatures across different aircraft types, engine designs, and operational conditions. A dataset consisting of 457 field acoustic measurements of commercial turbofan aircraft landing and taking-off from Amsterdam Airport Schiphol was used. To unveil meaningful patterns, we focused on dimensionality reduction techniques—Principal Component Analysis (PCA) and tdistributed Stochastic Neighbour Embedding (t-SNE)— to analyse this high-dimensional acoustic data. These methods are complemented by clustering algorithms and supervised machine learning models, such as K-Means, random forests for feature importance, and multilayer perceptrons (MLP) to classify aircraft types, engine configurations, and operations. Results reveal a strong loudness axis in the first principal component, overshadowing subtle spectral and timebased differences across aircraft families, especially for takeoffs. Nonetheless, focusing on higher-order components and alternative embeddings (t-SNE) highlights additional spectral and temporal markers. Operation classification (landing vs. takeoff) achieves 98% accuracy, but aircraft and engine family classification remain challenging, with accuracy capped below 50% using these feature sets. These findings suggest that advanced feature selection and dimensionality reduction while considering amplitude characteristics are essential for disentangling nuanced design-based acoustic traits.