smilegeng/nasa_cmapss_rul

NASA C-MAPSS RUL converted to TsFile

This dataset is a TsFile conversion of penikmatrumput/nasa-cmapss-rul. The source dataset is a modified NASA C-MAPSS turbofan engine degradation simulation dataset prepared for Remaining Useful Life (RUL) prediction.

The original dataset card describes each row as one operating cycle for one engine unit. Engines start with different initial wear and manufacturing variation, operate normally, and then degrade toward failure. The modeling task is to predict RUL from operating settings and sensor measurements.

Source License and Citation

• Source dataset license: cc0-1.0
• Source dataset: https://huggingface.co/datasets/penikmatrumput/nasa-cmapss-rul
• Original NASA C-MAPSS citation:

@inproceedings{saxena2008damage,
  title={Damage Propagation Modeling for Aircraft Engine Run-to-Failure Simulation},
  author={Saxena, A. and Goebel, K. and Simon, D. and Eklund, N.},
  booktitle={Proceedings of the 1st International Conference on Prognostics and Health Management (PHM08)},
  year={2008},
  address={Denver, CO}
}

Source Subsets

Subset	Train Units	Test Units	Operating Conditions	Fault Modes
FD001	100	100	1	High Pressure Compressor
FD002	260	259	6	High Pressure Compressor
FD003	100	100	1	HPC + Fan Degradation
FD004	248	249	6	HPC + Fan Degradation

Converted Files

The TsFile was built from the source C-MAPSS time-series CSV files:

• train_FD001.csv to train_FD004.csv
• test_FD001.csv to test_FD004.csv
• RUL_FD001.csv to RUL_FD004.csv

The source Hugging Face dataset loader exposes merged train and test splits, which do not preserve the FD001-FD004 subset identity. This conversion therefore reads the repository CSV files directly and restores the subset from the filename.

TsFile Layout

• TsFile: penikmatrumput_nasa-cmapss-rul.tsfile
• Table: nasa_cmapss_rul
• Time column: Time, copied from time_in_cycles
• Time precision: ms
• Tags: subset, split, unit_id
• Fields: unit_number, time_in_cycles, op_setting_1 to op_setting_3, sensor_1 to sensor_21, and RUL
• Row count: 265,256

unit_id is generated as <subset>_<split>_unit_<unit_number> so repeated unit numbers across subsets and splits remain separate TsFile devices.

RUL Handling

The training CSV files contain per-cycle RUL values. The test CSV files do not contain per-cycle RUL; their final RUL labels are provided separately in RUL_FD001.csv to RUL_FD004.csv. For test rows, per-cycle RUL is expanded as:

RUL = final_RUL + max(time_in_cycles for unit) - time_in_cycles

Row Ordering

Before writing TsFile, rows are sorted by:

subset, split, unit_number, Time

This keeps each TsFile device sequence contiguous and monotonically increasing in time. Because split is sorted lexically, test rows appear before train rows within each FD subset. To compare values against the source CSVs, filter by subset, split, and unit_number, then match by time_in_cycles.

Files Not Included

The source repository also contains documentation and synthetic business/order/production CSV files. Those files are not included in this TsFile because they are not engine sensor time-series measurements at the same grain as the C-MAPSS train/test files.