Multi-domain vibration dataset with various bearing types under compound machine fault scenarios: subset 1 (deep groove ball bearing)

Note: Due to the data storage limitation, our dataset is divided into three subsets for each bearing type. This subset includes the data collected from the deep groove ball bearing (MOCHU 6204).

This dataset provides vibration data collected under various fault conditions, including compound faults, and multiple domain environments. The faults include three single-bearing faults, seven single rotating component faults, and 21 compound faults. The domain configurations are categorized into the rotating speed, bearing type, and sampling rate. The data were collected for 160 seconds at an 8 kHz sampling rate and 80 seconds at a 16 kHz sampling rate, resulting in a uniform sample length of 1,280,000 for each raw vibration signal.

The data files are organized in a hierarchical directory structure. The top-level directories are based on the sampling rate (8 kHz and 16 kHz). Each sampling rate directory has subdirectories for six rotating speeds (600, 800, 1,000, 1,200, 1,400, and 1,600 RPM). Each rotating speed subdirectory contains 32 files collected under different fault conditions.

Data samples are stored in binary MATLAB (MAT) file. The naming convention of each file follows the format comprising five properties: {rotating component condition}_{bearing condition}_{sampling rate}_{bearing model}_{rotating speed}.mat.

The fault conditions are denoted by one capital letter, and unbalance and misalignment faults include three severity levels, where a higher number indicates that a more significant fault occurs. Available values of each property in the file name are as follows:
1. Rotating component condition
- ‘H’: healthy (no severity level)
- ‘M’: misalignment (severity level 1-3)
- ‘U’: unbalance (severity level 1-3)
- ‘L’: looseness (no severity level)
2. Bearing condition
- ‘H’: healthy
- ‘B’: ball fault
- ‘IR’: inner race fault
- ‘OR’: outer race fault
3. Sampling rate
- ‘8’: 8 kHz
- ‘16’: 16 kHz
4. Bearing model
- ‘6204’: bearing model for H, B, IR, and OR conditions
5. Rotating speed
- ‘600’, ‘800’, ‘1000’, ‘1200’, ‘1400’, or ‘1600’: numbers represent RPM.

The data are available in two formats: 1-D raw vibration signal and 2-D spectrogram. These time and time-frequency domain data are stored in separate fields of a MAT file. Data fields in each MAT file are as follows:
1. ‘Data’: raw vibration signal (unit: g [9.80665 m/s2]).
2. ‘Spectrogram’: spectrograms (unit: dB scale magnitude).
3. ‘STFTFreq’: frequency instant vector of the spectrogram (unit: Hz).
4. ‘STFTTime’: time instant vector of the spectrogram (unit: seconds).

For each data file, 78 spectrograms with a size of 128 x 128 were generated by the short-time Fourier transform (STFT) method with the Kaiser window function. In doing so, raw vibration data were sliced into 16,384-length segments without overlapping. Each segment was then converted to a spectrogram with a window size of 192, an FFT size of 256, and an overlap size of 65.

注意：由于数据存储限制，本数据集针对每种轴承类型均划分为三个子集。本次公开的子集包含从深沟球轴承（deep groove ball bearing）采集得到的MOCHU 6204相关数据。

本数据集提供多种故障工况下的振动数据，涵盖复合故障以及多域环境下的采集数据。故障类型包括3种单一轴承故障、7种单一旋转部件故障，以及21种复合故障。域配置按旋转转速、轴承类型与采样率三类进行划分。数据采集时长为：8 kHz采样率下采集160秒，16 kHz采样率下采集80秒，最终每段原始振动信号的统一采样长度均为1,280,000。

数据文件采用层级目录结构组织。顶层目录以采样率（8 kHz与16 kHz）划分。每个采样率目录下包含6个转速子目录，转速分别为600、800、1000、1200、1400及1600转每分钟（RPM）。每个转速子目录包含32个不同故障工况下采集的数据文件。

数据样本以二进制MATLAB（MAT）文件格式存储。每个文件的命名规则包含5个属性，格式为：{旋转部件工况}_{轴承工况}_{采样率}_{轴承型号}_{转速}.mat。

故障工况由一个大写字母表示，其中不平衡与不对中故障包含3个严重程度等级，数值越高代表故障越严重。文件名各属性的可选取值如下：
1. 旋转部件工况：
- ‘H’：健康状态（无严重程度等级）
- ‘M’：不对中故障（严重程度等级1-3）
- ‘U’：不平衡故障（严重程度等级1-3）
- ‘L’：松动故障（无严重程度等级）
2. 轴承工况：
- ‘H’：健康状态
- ‘B’：滚动体故障
- ‘IR’：内圈故障
- ‘OR’：外圈故障
3. 采样率：
- ‘8’：8 kHz
- ‘16’：16 kHz
4. 轴承型号：
- ‘6204’：适用于H、B、IR及OR工况的轴承型号
5. 转速：
- ‘600’、‘800’、‘1000’、‘1200’、‘1400’或‘1600’：数值单位为转每分钟（RPM）。

本数据集提供两种数据格式：一维（1-D）原始振动信号与二维（2-D）语谱图。这些时域与时频域数据分别存储于MAT文件的不同字段中。每个MAT文件包含以下字段：
1. ‘Data’：原始振动信号（单位：g [9.80665 m/s²]）。
2. ‘Spectrogram’：语谱图（单位：dB刻度幅值）。
3. ‘STFTFreq’：语谱图的频率瞬时向量（单位：Hz）。
4. ‘STFTTime’：语谱图的时间瞬时向量（单位：秒）。

针对每个数据文件，通过短时傅里叶变换（short-time Fourier transform, STFT）结合凯瑟窗（Kaiser window）函数生成78张尺寸为128×128的语谱图。具体实现流程为：将原始振动数据切割为长度16384的无重叠分段，随后将每个分段转换为语谱图，其窗长为192、快速傅里叶变换（Fast Fourier Transform, FFT）点数为256、重叠长度为65。

搜集汇总

背景与挑战

背景概述

该数据集提供了深沟球轴承在多种故障条件下的振动数据，包括单一和复合故障，覆盖不同的旋转速度和采样率。数据以1-D原始信号和2-D频谱图两种格式存储，适用于机械故障诊断和领域适应研究。