This work applies a forecasting-based anomaly detection approach by using an optimized multi-output Temporal Fusion Transformer (TFT) model and multivariate time-series digital flight data acquired from a diverse set of on-board sensors. Nominal flight behavior forecasting capability is achieved through training a TFT model using nominal flight data only. This TFT model is then used in a forecasting-based anomaly detection setting to detect anomalous events relating to unstable approach (UA). Investigations that were conducted to support this approach included hyperparameter optimization, Root Mean Squared Error (RMSE) threshold profiling, and performance benchmarking. Flight Operational Quality Assurance (FOQA) digital flight data was used to conduct the analyses.
Forecasting-Based Anomaly Detection with Temporal Fusion Transformers Using Flight Operational Quality Assurance Data
2024-09-29
847795 byte
Conference paper
Electronic Resource
English
Analysis of Flight Operational Quality Assurance Data Using Model-Based Activity Tracking
SAE Technical Papers | 2001
|Flight Operational Quality Assurance in Corporate Aviation
British Library Conference Proceedings | 2002
|Flight Demand Forecasting with Transformers
TIBKAT | 2022
|Flight Demand Forecasting with Transformers
AIAA | 2022
|Application of Flight Operational Quality Assurance to Corporate Flight Operations
British Library Conference Proceedings | 2007
|