This repository houses a rich and extensive dataset capturing vibration signals emanating from a gearbox operating under a spectrum of variable speed and load conditions. These datasets are meticulously crafted to encompass various fault types, degrees, and operational scenarios, serving as a pivotal resource for researchers and practitioners delving into vibration analysis, machine condition monitoring, and fault diagnosis within the realm of mechanical engineering.
- Repository Name: Multi-mode Fault Diagnosis Datasets of Gearbox Under Variable Working Conditions
- Direct URL: [Mendeley Data] [IEEE Dataport]
- Data Article [Paper]
- Compund Fault Diagnosis
- Fault Diagnosis With Different Fault Severity Degrees
- Fault Diagnosis With Multiple Steady Working Conditions
- Fault Diagnosis With Unknown Working Conditions
- Fault Diagnosis With Variable Working Conditions
- Fault Diagnosis With Transitional Working Conditions
- Data Format: CSV
- Total Datasets: 240 sets of time series data
- Data Collection Sources:
- Vibration signals from the motor drive end and gearbox intermediate shaft
- Key phase signal (speed data) of motor output shaft
- Torque data of gearbox input shaft
- Data Source: Collected by the MCC5 Group Shanghai Co. LTD and the Department of Automation at Tsinghua University
Each dataset was measured with a sampling frequency of 12.8 kHz. The datasets were stored in the standard Excel format, ".csv," in a single column without a time stamp. They were collected at time-varying speeds or time-varying loads for a fixed duration of 60 seconds, with the set speed-time curve and load-time curve depicted as follows. The number of colors in the figure is used to distinguish groups of experiments. Taking the 0-2500-3000 situation marked in blue in Figure 2(a) as an example, it means that the speed is set to 3000 rpm within 10-20 seconds and within 40-50 seconds. At the same time, the rotation speed is set to 2500 rpm within 25-30 seconds.
- If you want to perform visual analysis on a specific file, you can refer to the demo files in the "visual" folder.
- Note: The
.mfile and the CSV file to be plotted must be placed in the same directory. - The demo files use examples with varying speed conditions:
plot_torque_change_speed.mvisualizes the 'torque' channel under varying speed conditions, whileplot_speed_change_speed.mvisualizes the 'speed' channel under varying speed conditions. - First, you need to set the fault type, fault level, and speed/load conditions to retrieve the target CSV file:
% Define the regular expression for searching file names
fault_name = 'gear_pitting';
fault_level = 'L';
mode_name = 'speed';
torque_choice = '20Nm';
speed_choice = '3000rpm';
- For the torque channel signal, during the experiment, the load is applied on the output shaft, while the measurement is taken on the input shaft. Therefore, there is a scaling factor between the measurement and the actual load, approximately 6 times.
% Extract the column named 'torque'
time_series_data = table_data.torque;
time_series_data = time_series_data .* 6;
- For the speed channel signal, since the measured value during the experiment is dimensionless speed, some processing is required to obtain the actual speed variation plot.
% Extract the column named 'speed'
time_series_data = table_data.speed;
time_series_data(time_series_data <= 2) = 0;
time_series_data(time_series_data > 2) = 1;
% Find the positions of the rising edges
rising_edges_index = find(diff(time_series_data) == 1);
% Create a time series
time = linspace(0, 60, numel(time_series_data));
% Extract the time points corresponding to each rising edge
rising_time_point = time(rising_edges_index);
% Calculate the intervals between adjacent rising edges
period = diff(rising_time_point);
frequency = 1 ./ period;
speed = frequency .* 60;
% Compute the moving average of the rising edge time points (with a window of 2)
mean_time_point = movmean(rising_time_point, 2);
% Remove the first value of the moving average
mean_time_point = mean_time_point(2:end);
Each dataset comprises 8 columns, detailing:
- Speed: Key phase signal of the motor output shaft (dimensionless)
- Torque: Torque applied to the gearbox input shaft (Nm)
- Motor Vibration (x, y, z): Axial, horizontal, and vertical vibration acceleration at the motor drive end (g)
- Gearbox Vibration (x, y, z): Axial, horizontal, and vertical vibration acceleration at the gearbox intermediate shaft bearing seat (g)
The experimental setup for this dataset comprises the following components:
-
2.2 kW Three-Phase Asynchronous Motor: The motor serves as the primary power source in the experimental setup.
-
Torque Sensor: A precision torque sensor (model S2001) is employed to measure the torque applied to the gearbox input shaft with a synthetic accuracy of ±0.5%F.S.
-
Two-Stage Parallel Gearbox: The gearbox system used in the setup is of the two-stage parallel type.
-
Magnetic Powder Brake: Acting as a torque generator, the magnetic powder brake applies a torque load to the gearbox.
-
Measurement and Control System: An advanced measurement and control system is utilized to monitor and regulate the experimental parameters.
-
The model of the data acquisition system: CMS-ONE-DAQ16, with a maximum sampling frequency of 128 kHz and an analog-to-digital conversion accuracy of 24 bits.
This dataset aims to simulate and meticulously document various fault conditions associated with the 36-tooth gear on the intermediate shaft and its adjacent support bearings under diverse operating modes. The magnetic powder brake is pivotal in applying torque loads to the gearbox, enabling the simulation of real-world operational conditions.
The actual torque endured by the gearbox input shaft is precisely measured by the torque sensor, ensuring accurate data collection. Additionally, a speed sensor is employed to measure the key phase signal of the motor output shaft, enabling the calculation of motor output shaft speed from the key phase signal.
The test rig is equipped with two three-axis vibration acceleration sensors (model TES001V) capable of measuring motor and gearbox intermediate shaft triaxial vibrations along the x-, y-, and z-axes at a sampling frequency of 12.8 kHz.
Datasets were collected and meticulously processed under 12 distinct working conditions. To minimize experimental and measurement errors induced by temperature variations, laboratory temperature differences were rigorously controlled within a range of 2$^o$C.
Researchers utilizing this dataset are kindly requested to cite the following article:
@article{CHEN2024110453,
title = {Multi-mode Fault Diagnosis Datasets of Gearbox Under Variable Working Conditions},
journal = {Data in Brief},
pages = {110453},
year = {2024},
doi = {https://doi.org/10.1016/j.dib.2024.110453},
author = {Shijin Chen and Zeyi Liu and Xiao He and Dongliang Zou and Donghua Zhou},
}
The following research works have utilized the MCC5-THU Gearbox Benchmark Datasets for experiments and analysis. Researchers can refer to the following articles for further insights:
| Title | Journal / Conference | Year |
|---|---|---|
| A dynamic anchor-based online semi-supervised learning approach for fault diagnosis under variable operating conditions | Neurocomputing | 2025 |
| MHSNet: A Multi-Scale Hidden State Interaction Network for Fault Diagnosis of Rotating Machinery | Tsinghua Sci. Technol | 2025 |
| Knowledge-informed FIR-based cross-category filtering framework for interpretable machinery fault diagnosis under small samples | Reliability Engineering & System Safety | 2025 |
| Vibration-based gear wear area monitoring for quantitative assessment of wear severity under variable speed conditions | Mechanical Systems and Signal Processing | 2025 |
| Imbalanced real-time fault diagnosis based on minority-prioritized online semi-supervised random vector functional link network | IEEE Transactions on Instrumentation and Measurement | 2024 |
| A graph representation learning-based method for fault diagnosis of rotating machinery under time-varying speed conditions | Nonlinear Dynamics | 2025 |
| Compound fault diagnosis method of rotating machinery using multi-view multi-label feature selection based on label compression and local label correlation | Advanced Engineering Informatics | 2025 |
| PLL-WCAN: Pseudo-label progressive learning guided wavelet class-aware adaptive network for gearbox cross-domain fault diagnosis | Mechanical Systems and Signal Processing | 2025 |
| An interpretable wavelet Kolmogorov--Arnold convolutional LSTM for spatial-temporal feature extraction and intelligent fault diagnosis | Journal of Dynamics, Monitoring and Diagnostics | 2025 |
| IBL-AE: An Interpretable Base Learning Autoencoder for Intelligent Fault Diagnosis of Rotating Machinery | Knowledge-Based Systems | 2025 |
| Accurate and efficient instantaneous angular speed estimation method for rolling bearing under time-varying speed | IEEE Transactions on Instrumentation and Measurement | 2024 |
| Multi-Channel and Multi-Scale Weight Adaptive Neural Network for Intelligent Rotating Speed Extraction | Expert Systems with Applications | 2025 |
| Local direction-synchrosqueezing transform: An energy-concentrated post-processing method and its application in mechanical fault feature extraction | Mechanical Systems and Signal Processing | 2025 |
| Multi-scale distance similarity entropy: A novel complexity measurement for gearbox fault diagnosis | Engineering Applications of Artificial Intelligence | 2025 |
| A mechanism-informed contrastive learning framework for zero-shot gear anomaly detection | Measurement | 2025 |
| An advanced framework for efficient Variational Mode Extraction in gearbox vibration response analysis | Measurement | 2026 |
| Residual feature-based semi-supervised anomaly detection for Gearbox under varying rotational speeds | Measurement | 2025 |
| Single-trend component extraction for fault diagnosis of rotating machinery under time-varying speed conditions | Measurement | 2025 |
| Enhancing SDP-CNN for Gear Fault Detection Under Variable Working Conditions via Multi-Order Tracking Filtering | Journal of Dynamics, Monitoring and Diagnostics | 2025 |
| Information-theoretic continuously indexed domain adaptation network with wavelet-scale-wise convolution for fault diagnosis under continuously varying working conditions | ISA transactions | 2025 |
| Composite fault diagnosis of gearbox based on deep graph residual convolutional network | Engineering Research Express | 2024 |
| Synchrosqueezing-based demodulation transform for analyzing complicated signal containing strong noise | IEEE Transactions on Instrumentation and Measurement | 2025 |
| A Novel Compound Fault Diagnosis Method for Rotating Machinery Based on Dynamic Adaptive MWPE and Dual-Graph Regularization Strategy | IEEE Sensors Journal | 2025 |
| Vibration characteristic analysis of a two-stage spur gear transmission system with tooth crack and profile shifted | Journal of Sound and Vibration | 2025 |
| Sparse Attention Coding-Timestep Shrinking Spiking Neural Network for Gearbox Fault Diagnosis | IEEE Sensors Journal | 2025 |
| Double synchronous and non-synchronous symbol fuzzy similarities: A feature extraction approach for gear fault diagnosis | Measurement | 2025 |
| HDNAT: A Transformer-Based Fault Diagnosis Model For Rotating Machinery Across Varying Damage Degrees | IEEE Transactions on Instrumentation and Measurement | 2025 |
| A multi-sensor fault diagnosis model with adaptive spatial-temporal dual-scale re-modeling | IEEE Sensors Journal | 2024 |
| Continuously Indexed Domain Generalization for Fault Diagnosis under Continuously Varying Working Conditions | IEEE Transactions on Instrumentation and Measurement | 2025 |
| A fault diagnosis method for rotating machinery based on spatiotemporal feature fusion | Journal of Mechanical Science and Technology | 2025 |
| An efficient fault diagnosis method for rotating machinery based on PageRank-guided cluster sampling and Pearson-Shannon feature selection | Journal of Mechanical Science and Technology | 2025 |
| Maximum reweighted local kurtosis deconvolutional filtering and application for time-varying speed rotating machinery fault diagnosis | Transactions of the Institute of Measurement and Control | 2025 |
| A lightweight gearbox fault diagnosis method based on multi-sensor information multi-dimensional fusion under small sample conditions | Measurement Science and Technology | 2025 |
| Gearbox fault diagnosis method based on adaptive multi-sensor data level fusion and fine-grained domain adaptation | Advances in Mechanical Engineering | 2025 |
| Analysis of shaft alignment on vibration and motor power consumption in cooling water pumps | Mechanical Science Reports | 2025 |
Furthermore, we sincerely appreciate the NASA Ames Research Center and its team for their development and contribution to the open-source Python prognostics tool, ProgPy. We are especially grateful for their analysis and interpretation of the MCC5-THU gearbox fault diagnosis dataset's organizational structure in their Technical Report. This effort not only enhances the dataset’s applicability in the field of Condition Monitoring and Diagnostics (CM&D) but also provides valuable insights for further research. We look forward to the continued development of ProgPy in the field of Prognostics and Health Management (PHM), driving advancements in intelligent maintenance and fault diagnosis technologies.
- If you have used this dataset in your research, please feel free to share your published work with us. We are happy to include it in this list to support the research community. We sincerely appreciate the support of the research community and encourage researchers to share their publications using this dataset. Thank you for your contributions!
This project is licensed under the MIT License - see the LICENSE file for details.
We extend our sincere gratitude to the THUFDD Group, led by Prof. Xiao He and Prof. Donghua Zhou, for their invaluable support and contributions to the development of this scheme.
We express our gratitude to the MCC5 Group Shanghai Co. LTD for their invaluable support.
As stated in our paper, it is important to account for a multiplication factor due to the load applied to the output shaft, with measurements being taken from the input shaft. This factor is approximately 6. Additionally, please be aware that some unavoidable noise may occur during the start-stop phase of the equipment, although this is typically negligible and can generally be disregarded.
Furthermore, in Version 1 of the dataset, an issue was identified with the torque measurements, specifically that the sign of the torque was reversed. This issue has been rectified in Version 2 of the dataset. Kindly ensure that the [Latest Version] is being used for accurate results. We will continue to provide updates as needed.