Flight anomaly detection plays a critical role in ensuring aviation safety. However, conventional anomaly detection methods often suffer from limited accuracy due to their reliance on single-sensor observations or fixed-weight multi-sensor fusion strategies. To address these limitations, this paper proposes an adaptive weight fusion framework for flight-test anomaly detection. The proposed framework integrates a signal-to-noise-ratio-based adaptive weighting strategy, an improved Kalman filtering algorithm, and an ensemble anomaly detection model within a hierarchical fusion architecture to achieve robust multi-sensor data fusion and anomaly identification.
Experiments were conducted using the NASA DASHlink Tail_666_9 flight dataset. Compared with conventional methods, the proposed framework achieved improvements in overall accuracy, precision, recall, and F1-score. Specifically, the proposed method achieved an F1-score of 0.486 while reducing the average anomaly detection delay by more than 2 s. These results indicate that the proposed framework can improve the effectiveness and timeliness of flight anomaly detection, demonstrating its potential for real-time flight safety monitoring.