RTFnet: A multi-modal physiological data fusion approach for fatigue detection in remote tower controllers

Remote towers, as a groundbreaking air traffic control technology, significantly alter the workflows and task loads of controllers, which may exacerbate mental fatigue. Unlike traditional towers, remote tower controllers (RTCos) operate with less direct interaction with aircraft and more reliance on remote monitoring systems, creating unique challenges in detecting and managing their fatigue. Existing methods are insufficient for accurately monitoring RTCos’ status, making it difficult to prevent human errors that could compromise aviation safety. To address these challenges, we propose a multi-modal physiological data fusion framework, RTFnet, which integrates a dynamic graph attention mechanism (DGAT) with gated recurrent units (GRU) to enhance the detection of controller fatigue. Specifically, RTFnet explicitly models the hierarchical relationships among features, fuses key features with appropriate weights, and utilizes GRU to capture the temporal evolution of fatigue states. Experiments on a real-world dataset show that the framework achieves high real-time detection accuracy (93.95% within a 10-second time step, with precision = 93.56%, recall = 95.39%, F1 = 94.32%, Specificity = 93.14%). The model’s visualization analysis reveals that a well-designed graph structure plays a critical role in processing non-Euclidean data, guiding efficient information flow among relevant task nodes. This method offers a robust solution for real-time fatigue monitoring in remote tower operations, reducing human error and enhancing aviation safety.