Development and evaluation of an ensemble neural network based methodology for rapid diagnosis and fault classification in unmanned aerial vehicles
摘要
As the use of unmanned aerial vehicles (UAVs) expands, real-time failure detection is essential for effective recovery and forensic analysis. Failures may arise from electronic issues, physical damage, or cyber-attacks. This study presents a neural network ensemble for diagnosing and classifying UAV failures using a hybrid optimizer combining particle swarm optimization and genetic algorithms, with hyperparameters refined through Bayesian optimization. The methodology, tested with data from Parks College of Engineering, achieved 100% accuracy in diagnosing UAV states and 98.51% in fault classification within 1 s windows, proving its effectiveness in rapidly identifying UAV faults.