Interpretable Anomalous Event Detection with Uniform Dynamic Alignment of Time Series in Aircraft Aviation Safety

Interpretable anomalous event detection plays a crucial role in proactive aviation safety by identifying anomalous patterns and explaining the occurrence of such events. However, varied-length time series in certain flight phases pose challenges for traditional modeling methods that require equal-length in-puts. To address this, we propose a uniform dynamic time warping with a uniform Scaling alignment method. This method combines global and local information, resolving the variable-length issue while minimizing information loss. Additionally, we perform error analysis on specific samples tot est t he detection model's capacity to identify different adverse samples with varied anomalous behaviors. Results in real flight data verify the proposed alignment method enhances the model performance and also shows a slight edge over other alignment methods.