UPATRAS Rotating Machinery Vibration Dataset for Incipient Fault Diagnosis under Varying Rotating Speed
Description
This dataset contains vibration measurements acquired from a rotating machinery test rig developed at the University of Patras, Greece, at the Stochastic Mechanical Systems and Automation (SMSA) Laboratory. The test rig consists of two foot-mounted electric motors coupled via a claw clutch and instrumented with a single uniaxial accelerometer mounted on the drive motor. The dataset has been designed for research on vibration-based condition monitoring, fault detection, fault diagnosis, signal processing, feature extraction, machine learning, deep learning, and related data-driven methodologies for rotating machinery operating under varying speed conditions. In contrast to many rotating machinery datasets that focus on a limited number of operating conditions, the present dataset provides dense coverage of a wide rotating-speed range. The dataset comprises eight machinery states, namely one healthy state and seven incipient fault scenarios associated with three fault families: limited unbalance, mechanical looseness, and coupler wear. The limited unbalance scenarios are implemented by replacing the main coupler mounting bolt with a heavier bolt, leading to two fault levels, denoted as Unbalance 3g and Unbalance 5g. The mechanical looseness scenarios are implemented through torque reduction of the drive-motor mounting bolts A and B, leading to four fault cases: Bolt A 50%, Bolt A 100%, Bolt B 50%, and Bolt B 100%. The coupler wear scenario corresponds to incipient wear at the base of a single spider tooth of the claw clutch. The healthy state is characterized by mounting torques [A,B] = [5,5] N m, while the looseness scenarios are defined through the corresponding reduced torque values. The measurements are acquired under 75 different rotating speeds ranging from 35.0 Hz to 49.8 Hz with a step of 0.2 Hz. Four measurement sequences are provided for the healthy state and five for each faulty state, leading to a total of 2925 individual vibration signals. Each signal contains 3500 samples, corresponding to 3.42 s with sampling frequency 1024 Hz and frequency bandwidth [0 - 512] Hz. The dataset is provided entirely in CSV format. If you use this dataset in your work, please cite the following publication: https://doi.org/10.1016/j.ymssp.2025.113204 Further details on the data are available in the README.pdf file included in this dataset.
Files
Institutions
- University of PatrasWest Greece, Pátrai