Multi-domain vibration dataset with various bearing types under compound machine fault scenarios: subset 2 (cylindrical roller bearing)
Description
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 cylindrical roller bearing (NTN N204 and NJ204). 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 - ‘N204’: bearing model for H, B, and OR conditions - ‘NJ204’: bearing model for IR condition 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.