A Fault Prediction Method for Quadrotor UAVs Integrating Residual Detection and Transformer Models

Unmanned aerial vehicles (UAVs) operating in complex environments are susceptible to various faults in their flight control systems, potentially leading to instability or loss of control. Given these risks, fault prediction is critical to enable early intervention and ensure flight safety. To achieve earlier and more robust fault prediction, this paper proposes a hybrid model for quadrotor UAV flight control systems. Fault detection is performed using a residual-based detection approach, followed by residual prediction via a modified Transformer model. The prediction process is dependent on detection outcomes and is suitable for systems characterized by gradual fault progression or a clear degradation phase between anomaly emergence and failure. This model improves computational efficiency by constructing a lightweight Swin-iTransformer architecture, which incorporates a local attention mechanism. Experimental results demonstrate that the proposed method can effectively predict faults at an early stage. The integration of model-driven detection with data-driven prediction allows for early and reliable fault identification in UAV flight control systems, which helps improve UAV operational safety under complex conditions.