A Baseline Modeling Method for UAV Condition Monitoring based on Multiple Flight Data

Unmanned aerial vehicles (UAVs) have been widely used. To avoid the huge losses caused by the high accident rate of UAVs, it is significant to carry out research on UAV condition monitoring (CM) to find anomalies in UAVs in time. Establishing a baseline model of UAV normal operation status from flight data provides the possibility to realize UAV CM. However, the uncertainty of multiple flights makes it difficult to enable the generalization ability of the baseline model. Therefore, this paper proposes a baseline modeling method with uncertainty representation ability based on multiple flight data for UAV CM. First, multiple normal flight data are normalized and reconstructed. Then, construct an LSTM-based baseline model to predict the monitored parameter, and Monte Carlo dropout approximate inference is introduced to represent the uncertainty of the prediction dynamically. Finally, statistical thresholds are designed based on the distribution of the predictions, and the measurements outside the thresholds will be regarded as outliers, and the real flight data is used to verify the performance of the baseline modeling method, and the experimental results proved the better F1-Score of the proposed method.