MetroAT

About the Dataset

Overview
The MetroAT dataset is a newly introduced open-source benchmark dataset designed for research and development in data-driven maintenance of metro pneumatic systems. It comprises 105 variables and approximately 25 million observations collected continuously over a one-year period, capturing the operational behavior of multiple pneumatic system components.

Dataset Structure

Collection Period: June 2024 – June 2025  
- Sampling Frequency: 1 Hz  
- Data Organization: Files are organized hierarchically by year, month, and day to support efficient storage and retrieval, with the dataset further divided into training and test subsets.
  

Variables

The dataset contains 105 variables, excluding timestamp-related fields (`timestamp`, `day`, `month`, and `year`), categorized as follows:

- 72 continuous sensor variables
- 20 binary sensor variables
- 7 operational state variables
- 4 failure and maintenance variables
- 2 aggregated sensor variables