CAFUC2

=============================================================================A High-Fidelity Multi-Model Benchmark Dataset for General Aviation============================================================================= Authors: Jing Lu, Yali Fang, Ziyi Huang, Yuan Zhang, Songhan Fan, Xuan WuAffiliation: Nanjing University of Aeronautics and Astronautics (NUAA), Civil Aviation Flight University of China (CAFUC)Contact: [e-mail] (Yali Fang)License: Creative Commons Attribution 4.0 International (CC BY 4.0) 1. OVERVIEW-----------------------------------------------------------------------------The CAFUC2 dataset is a high-fidelity multivariate time-series benchmark designed for General Aviation (GA) anomaly detection. It contains approx. 1 million data points (sampled at 1 Hz) derived from 120 real flight sorties.The fleet includes four heterogeneous aircraft models: Cessna 172R, Cessna 172S, Cirrus SR20, and Cirrus SR20 G6. To provide robust ground-truth labels for safety-critical events, we applied a Physics-Based Fault Injection framework to generate four types of anomalies, ensuring thermodynamic lag and kinematic consistency. 2. DIRECTORY STRUCTURE-----------------------------------------------------------------------------The dataset is organized under the root directory `CAFUC2/` as follows: CAFUC2/├── normal_data/ # Cleaned baseline flight logs│ ├── C172R/ # Categorized by aircraft model│ ├── C172S/│ ├── SR20/│ └── SR20G6/ │ └── cleaned_XXXX_filled.csv # XXXX: 4-digit flight ID│├── abnormal_data/ # Flight logs with synthetic faults│ ├── accelerator_operation/ # Throttle Surge (RPM instability & thermal lag)│ ├── course_deviation/ # Flight Path Deviation (Kinematic lateral drift)│ ├── engine_power_loss/ # Cylinder Misfire (Partial Power Loss)│ └── pitch_attitude/ # Pitch Excursion (Aerodynamic G-force coupling)│ └── clean_data_[Model]/│ └── X.csv # X: Extracted integer ID corresponding to baseline│ 3. FILE CORRESPONDENCE (COUNTERFACTUAL MAPPING)-----------------------------------------------------------------------------To facilitate counterfactual analysis (i.e., comparing a normal flight segment against its anomalous counterpart), a strict one-to-one mapping is maintained. Example: - Baseline file: `normal_data/C172R/cleaned_0001_filled.csv`- Anomalous file: `abnormal_data/pitch_attitude/clean_data_C172R/1.csv`Both files represent the exact same flight sortie, but the latter contains the injected "Pitch Excursion" anomaly at specific timestamps. 4. DATA DICTIONARY (COLUMN DESCRIPTIONS)-----------------------------------------------------------------------------Each CSV file contains multivariate time-series flight parameters. Missing values in the raw data were interpolated, and sensor noise was smoothed using a rolling IQR method. To optimize memory footprint and ensure precision, geospatial coordinates (Latitude/Longitude) are rounded to 7 decimal places, while all other sensor readings are rounded to 3 decimal places. Variables included:* label : Ground truth label (0 = Normal, 1 = Anomaly). Note: Baseline files implicitly have label=0.* Latitude : Aircraft latitude coordinate (Decimal Degrees).* Longitude : Aircraft longitude coordinate (Decimal Degrees).* AltMSL : Altitude above Mean Sea Level (Feet).* VSpd : Vertical Speed / Climb rate (Feet per minute).* Pitch : Pitch attitude angle (Degrees, positive = nose up).* Roll : Roll attitude angle (Degrees, positive = right bank).* LatAc : Lateral Acceleration (G).* NormAc : Normal Acceleration / G-force (G).* HDG : Magnetic Heading (Degrees).* TRK : Ground Track angle (Degrees).* E1 FFlow : Engine 1 Fuel Flow (Gallons per hour).* E1 OilT : Engine 1 Oil Temperature (Degrees Fahrenheit).* E1 RPM : Engine 1 Speed (Revolutions per minute).* E1 CHT1-4 : Engine 1 Cylinder Head Temperature for Cylinders 1 to 4 (Degrees Fahrenheit).* E1 EGT1-4 : Engine 1 Exhaust Gas Temperature for Cylinders 1 to 4 (Degrees Fahrenheit).* AltGPS : GPS-derived Altitude (Feet). 5. ANOMALY DESCRIPTIONS-----------------------------------------------------------------------------1) Throttle Surge (accelerator_operation): Simulates uncommanded engine RPM fluctuations. Fuel flow reacts instantly, while EGT/CHT exhibit a delayed response (thermal lag).2) Flight Path Deviation (course_deviation): Simulates navigation drift. Geospatial coordinates (Lat/Lon) are recalculated based on deviated track angles using spherical trigonometry.3) Cylinder Misfire (engine_power_loss): Simulates cooling system degradation. Causes non-linear CHT temperature rises, coupled with subtle performance drops in Vertical Speed.4) Pitch Excursion (pitch_attitude): Simulates abrupt attitude upsets. Pitch changes are kinematically coupled with corresponding fluctuations in Normal Acceleration (G-force). 6. CITATION-----------------------------------------------------------------------------If you use this dataset in your research, please cite the corresponding Data Descriptor published in Scientific Data (Nature Portfolio). @article{CAFUC2_2026, title={A High-Fidelity Multi-Model Benchmark Dataset for General Aviation Anomaly Detection Generated via Physics-Based Fault Injection}, author={Lu, Jing and Fang, Yali and Fan, Songhan and Wu, Xuan and Huang, Ziyi}, journal={Scientific Data}, year={2026}}=============================================================================