| --- |
| license: apache-2.0 |
| tags: |
| - industrial-maintenance |
| - question-answering |
| - benchmark |
| language: |
| - en |
| --- |
| |
| # Dataset Card for ibm-research/AssetOpsBench |
|
|
| AssetOpsBench is a curated benchmark dataset for evaluating agentic systems and large language models (LLMs) in **industrial asset operations**, particularly focusing on tasks like condition monitoring, failure analysis, and sensor-based diagnostics. |
|
|
| The dataset provides structured question-answering and multi-agent scenario prompts based on real-world operational contexts, such as HVAC systems (e.g., chillers, air handling units), and is part of the AssetOpsBench framework developed at IBM Research. |
|
|
| --- |
|
|
| ## Dataset Summary |
|
|
| This dataset contains two major categories: |
|
|
| 1. **Scenario QA (`scenarios`)**: |
| Human-authored evaluation prompts designed to test the performance of multi-agent orchestration strategies. |
|
|
| 2. **FailureSensorIQ (`failuresensoriq_...`)**: |
| A family of question-answering datasets focused on sensor-failure relationships, supporting multiple variants for robustness and evaluation. |
| |
| --- |
| |
| ## Supported Configurations |
| |
| | Config Name | Type | Description | |
| |-----------------------------------|------------------|----------------------------------------------------------------| |
| | `scenarios` | Scenario QA | Human-authored prompt scenarios for agent orchestration | |
| | `failuresensoriq_all` | QA Dataset | Full dataset with diverse QA covering failure-sensor mappings | |
| | `failuresensoriq_10_options` | QA Dataset | 10-option MCQ format | |
| | `failuresensoriq_perturbed_complex` | QA Dataset | Complex perturbed choices for adversarial robustness | |
| | `failuresensoriq_perturbed_simple` | QA Dataset | Simpler perturbed options | |
| | `failuresensoriq_multi_answers` | QA Dataset | Questions with multiple correct options | |
| | `failuresensoriq_sample_50` | QA Subset | ⭐ Expert-authored subset of 50 questions for benchmarking | |
| |
| --- |
| |
| ## Languages |
| |
| - English |
| |
| --- |
| |
| ## Use Cases |
| |
| - Benchmarking LLMs (GPT, LLaMA, Claude, etc.) |
| - Evaluating multi-agent reasoning systems |
| - Industrial diagnostic model prototyping |
| - Agent planning and verification |
| - Robustness testing via perturbed QA |
| |
| --- |
| |
| ## Dataset Structure |
| |
| Each dataset split is provided in `.jsonl` format where each line is a QA pair or prompt dictionary. Common fields include: |
| |
| - `question`: The user query |
| - `options`: List of possible answers (if applicable) |
| - `answer`: Correct answer(s) |
| - `metadata`: (Optional) Includes asset ID, failure mode, or sensor references |
| |
| --- |
| |
| ## Citation |
| |
| If you use this dataset, please cite: |
| |
| @misc{patel2025assetopsbenchbenchmarkingaiagents, |
| title={AssetOpsBench: Benchmarking AI Agents for Task Automation in Industrial Asset Operations and Maintenance}, |
| author={Dhaval Patel and Shuxin Lin and James Rayfield and Nianjun Zhou and Roman Vaculin and Natalia Martinez and Fearghal O'donncha and Jayant Kalagnanam}, |
| year={2025}, |
| eprint={2506.03828}, |
| archivePrefix={arXiv}, |
| primaryClass={cs.AI}, |
| url={https://arxiv.org/abs/2506.03828}, |
| } |
| |
| |
| --- |
| |
| ## License |
| |
| Apache 2.0 |
| |
| --- |
| |
| ## Source and Acknowledgements |
| |
| Developed by [IBM Research](https://research.ibm.com/) as part of the [AssetOpsBench](https://github.com/ibm-research/AssetOpsBench) project. |
| |
| Special thanks to domain experts who authored the high-quality benchmark subset (`failuresensoriq_sample_50`). |
| |
| --- |
| |
| ## Tags |
| |
| - domain: industrial-maintenance |
| - type: expert-curated, QA, benchmark |
| - task: question-answering, multi-agent evaluation |
| - modality: text, tabular, sensor |
| |
