Unsupervised Maneuver-Aware Acoustic Fault Detection for Autonomous Drones

This project presents a research-backed framework for unsupervised, maneuver-aware acoustic fault detection in autonomous unmanned aerial vehicles (UAVs). Recognizing the safety risks and practical limitations of collecting labeled faulty-flight datasets, the methodology adopts an unsupervised learning paradigm trained exclusively on nominal flight recordings. The proposed architecture addresses the challenges of stochastic environmental noise and structured, maneuver-dependent aerodynamic variations through a two-stage deep learning pipeline. In the first stage, a Noise2Noise-inspired denoising model is utilized to attenuate random acoustic noise while preserving critical spectral-temporal features without requiring clean reference signals. In the second stage, a maneuver-Conditioned Convolutional AutoEncoder (maneuver-CCAE) integrates maneuver-related parameters, such as drone type and flight direction, to model

nominal acoustic behavior under diverse operating conditions. Fault detection is executed by evaluating the reconstruction error as an anomaly score. This framework provides comprehensive implementation guidance for developing robust diagnostic systems capable of distinguishing between normal maneuver-induced acoustic variations and actual structural or mechanical anomalies, thereby enhancing the operational safety and reliability of autonomous aerial platforms.

Message: Hi FE, I need help with “Unsupervised Maneuver-Aware Acoustic Fault Detection for Autonomous Drones” in “Electrical Engineering”