Jet Engine Simulated Time Series for Prognostic Health Management

Four datasets contain multivariate time series from simulated jet engines, with each row representing a snapshot from a single operational cycle. The data includes 26 columns per row, covering unit number, time in cycles, three operational settings, and 21 sensor measurements. Engine faults develop over time, and the objective is to predict the Remaining Useful Life (RUL) before system failure.

Use Cases

• Train a model to predict the Remaining Useful Life (RUL) using time series of sensor measurements and operational settings.
• Analyze fault progression patterns, such as HPC Degradation or Fan Degradation, across different operational conditions.
• Develop anomaly detection algorithms by monitoring deviations in the 21 sensor measurement channels from normal operational baselines.
• Compare prognostic model performance across datasets with varying fault modes and operational conditions.

Strengths

• Includes four distinct datasets (FD001-FD004) with up to 260 training trajectories, providing multiple scenarios for model validation.
• Each data row contains 26 explicitly defined variables, including three operational settings and 21 sensor measurements.
• Data is designed for a specific prognostic task with provided true RUL values for the test set, enabling supervised learning.

Limitations

• Data is simulated and contaminated with sensor noise, which may not fully capture the complexity and noise profiles of real-world engine sensors.
• The datasets have different sizes and fault modes (e.g., FD001 has one fault mode, FD003 has two), which can complicate direct comparisons if not accounted for.
• Lack of real-world validation; performance on this simulated data may not translate directly to physical jet engines.

Provenance

Source

National Aeronautics and Space Administration (NASA).

Collection Method

Generated via run-to-failure simulation of jet engines under different operational conditions and fault modes, as described in the referenced PHM08 conference paper.

Time Range

Simulation data; no specific real-world temporal coverage.

Freshness

Dataset metadata was last updated in 2025, though the underlying simulation and reference publication are from 2008.

Geography

Not applicable; data is from simulated engine models.