appleparan/telemanom

NASA SMAP and MSL Spacecraft Anomaly Detection Dataset

Dataset Description

This dataset contains real spacecraft telemetry data and labeled anomalies from two NASA missions:

• SMAP (Soil Moisture Active Passive satellite)
• MSL (Mars Science Laboratory / Curiosity Rover)

The data was originally used in the 2018 KDD paper Detecting Spacecraft Anomalies Using LSTMs and Nonparametric Dynamic Thresholding and released via the telemanom repository.

Dataset Structure

data/
  train/       # 82 Parquet files (one per channel)
  test/        # 82 Parquet files (one per channel)
  data/
    train/     # 82 .npy files (original format)
    test/      # 82 .npy files (original format)
labeled_anomalies.csv

Each channel is available as a separate config (e.g., A-1, P-1, M-1) that can be selected in the dataset viewer or loaded programmatically.

Data Files

Each Parquet file contains the following columns:

Column	Description
timestep	Integer index of the time step
value	Target telemetry value being monitored
cmd_0 ... cmd_N	One-hot encoded command information (24 or 54 columns depending on channel)

• All telemetry values are anonymized and normalized to [-1, 1] based on min/max in the test set.
• Channel IDs are anonymized; the first letter indicates channel type (P = power, R = radiation, etc.).
• SMAP channels have 24 command columns; MSL channels have 54 command columns.

Anomaly Labels (labeled_anomalies.csv)

Column	Description
chan_id	Anonymized channel ID
spacecraft	SMAP or MSL
anomaly_sequences	Start and end indices of anomalies in the test set
class	Anomaly type: point or contextual
num_values	Number of telemetry values in the test stream

Dataset Statistics

	SMAP	MSL	Total
Total anomaly sequences	69	36	105
Point anomalies (%)	43 (62%)	19 (53%)	62 (59%)
Contextual anomalies (%)	26 (38%)	17 (47%)	43 (41%)
Unique telemetry channels	55	27	82
Telemetry values evaluated	429,735	66,709	496,444

Usage

Using datasets library (recommended)

from datasets import load_dataset

# Load a specific channel (e.g., P-1)
ds = load_dataset("appleparan/telemanom", name="P-1")
print(ds)
# DatasetDict({
#     train: Dataset({features: ['timestep', 'value', 'cmd_0', ...], num_rows: ...})
#     test:  Dataset({features: ['timestep', 'value', 'cmd_0', ...], num_rows: ...})
# })

# Access train/test splits
train_df = ds["train"].to_pandas()
test_df = ds["test"].to_pandas()

Using numpy (original .npy format)

import numpy as np
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="appleparan/telemanom",
    filename="data/data/test/P-1.npy",
    repo_type="dataset",
)
data = np.load(path)
print(data.shape)  # (n_timesteps, n_inputs)

Source

• Original repository: khundman/telemanom
• Paper: arXiv:1802.04431
• Contributors: Kyle Hundman, Valentinos Constantinou, Christopher Laporte, Ian Colwell, Tom Soderstrom (NASA JPL)

Citation

@inproceedings{hundman2018detecting,
  title     = {Detecting Spacecraft Anomalies Using LSTMs and Nonparametric Dynamic Thresholding},
  author    = {Hundman, Kyle and Constantinou, Valentino and Laporte, Christopher and Colwell, Ian and Soderstrom, Tom},
  booktitle = {Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
  year      = {2018}
}

License

This dataset is distributed under the BSD 3-Clause License (Copyright (c) 2018, California Institute of Technology).

Note: The original repository's README states Apache 2.0, but the actual LICENSE.txt file contains a BSD 3-Clause license from Caltech/JPL. This dataset card follows the license file.