RMIS Benchmark Datasets

Introduction

RMIS is a benchmark dataset collection for evaluating representation learning on multi-modal industrial signals. It brings together the datasets used in the RMIS benchmark, covering anomaly detection and fault diagnosis across four modalities: sound, vibration, voltage, and current.

This Hugging Face repository mainly hosts the benchmark datasets themselves. If you are looking for the full benchmark codebase, evaluation pipeline, preprocessing details, or leaderboard, please refer to the RMIS GitHub repository.

RMIS is closely related to FISHER:

• FISHER is the foundation model proposed for industrial signal representation.
• RMIS is the benchmark used to evaluate FISHER and other signal foundation models.
• This Hugging Face repository hosts the dataset side of RMIS, while the GitHub repository hosts the benchmark code and evaluation pipeline.

In the current release, the dataset includes 19 configurations:

• Anomaly detection: dcase20, dcase21, dcase22, dcase23, dcase24, dcase25
• Fault diagnosis: iica, iiee, mafaulda_sound, mafaulda_vib, pu_cur, pu_vib, sdust_bearing, sdust_gear, umged_cur, umged_sound, umged_vib, umged_vol, wtpg

What is included

Each configuration corresponds to one benchmark subset. The exact schema varies by subset, but all configurations provide an audio column together with file-level metadata such as file_name, and task-specific annotations such as status, scene, mt, ori, domain, or related attributes.

The split names follow the benchmark tasks:

• ad: anomaly detection subsets
• fd: fault diagnosis subsets

Please note that this repository mainly serves as a data hosting and distribution endpoint. For RMIS-specific preprocessing, path organization, evaluation protocol, and model integration, please refer to the RMIS GitHub repository.

Recommended usage in the RMIS project

If you want to use these datasets inside the RMIS benchmark workflow, please first download the RMIS GitHub repository, install its dependencies, and then run the provided script from that repository.

git clone https://github.com/jianganbai/RMIS.git
cd RMIS/
pip install -r requirements.txt

Then use the following command to download and extract the Hugging Face data into RMIS-compatible local wav folders:

[HF_ENDPOINT=https://hf-mirror.com] python -m utils.scripts.download_and_extract_hf_data \
    --output_dir OUTPUT_DIR \
    [--subset SUBSET [SUBSET ...]] \
    [--remove_parquet_after_extract] \
    [--force_reextract]

For example:

python -m utils.scripts.download_and_extract_hf_data \
    --output_dir datasets_hf \
    --subset iiee mafaulda_sound \
    --remove_parquet_after_extract

Here --output_dir is required, while HF_ENDPOINT, --subset, --remove_parquet_after_extract, and --force_reextract are optional. If --subset is omitted, the script processes all RMIS subsets.

In the RMIS project, the Hugging Face route is mainly intended to treat this repository as a cloud storage backend and materialize the data into the same local wav-style directory layout used by the other RMIS download paths.

Additional usage

If you are interested in more customized workflows, you may also directly use the metadata and parquet assets attached to this Hugging Face repository to develop your own data loading, conversion, or preprocessing utilities.

However, for most users who want to reproduce RMIS experiments, the recommended path is still:

1. Download the RMIS GitHub repository.
2. Follow the benchmark instructions there.
3. Use the provided Hugging Face download-and-extract script when you prefer the Hugging Face storage route.

For more details about RMIS, including benchmark construction, evaluation, and usage, please refer to the RMIS GitHub repository.

Acknowledgements

RMIS is built from multiple public industrial signal datasets. We thank the original dataset creators for making these resources available.

If you believe that any content in this repository infringes your rights, please contact us and we will address the issue promptly.

Citation

If you find RMIS useful, please cite the following paper.

@article{fan2025fisher,
  title={FISHER: A Foundation Model for Multi-Modal Industrial Signal Comprehensive Representation},
  author={Fan, Pingyi and Jiang, Anbai and Zhang, Shuwei and Lv, Zhiqiang and Han, Bing and Zheng, Xinhu and Liang, Wenrui and Li, Junjie and Zhang, Wei-Qiang and Qian, Yanmin and Chen, Xie and Lu, Cheng and Liu, Jia},
  journal={arXiv preprint arXiv:2507.16696},
  year={2025}
}