A Machine Learning Dataset of Artificial Inner Ring Damages on Cylindrical Roller Bearings Measured Under Varying Cross-Influences

This dataset provides a high-resolution, well-annotated collection of vibration measurements from cylindrical roller bearings, both healthy and with artificially induced inner ring damage. It is designed to support machine learning research addressing domain shift by enabling robust evaluation of model generalization across realistic variations in rotational speed, applied load, and mounting position.

Unlike existing bearing datasets, this resource follows a structured experimental design with controlled covariates known to cause domain shifts. It includes 1,151 multi-axis recordings (20 kHz, 60 s) across multiple bearing instances, damage states, and operating conditions.

Optimized for Leave-One-Group-Out Cross-Validation (LOGOCV), the dataset facilitates rigorous assessment of model robustness to unseen conditions. It also includes:

Detailed metadata on testbed setup, damage geometry, and environmental parameters

Transparent labeling of assembly deviations for anomaly detection research

MATLAB scripts for streamlined data loading and segmentation

This dataset is particularly suited for work in robust ML, domain generalization, fault diagnosis, and industrial condition monitoring.

A detailed description of the data can be found at Data Descriptor.

This research was performed in the context of project VProSaar (“Verteilte Produktion für die saarländische Automotivindustrie: Nachhaltig, Vernetzt, Resilient ”) carried out at the Centre for Mechatronics and Automation Technology gGmbH and funded by the Ministry of Economic Affairs, Innovation, Digital and Energy (MWIDE) and the European Fonds for Regional Development (EFRE).