File Names

• Train and Eval Data (train 3354, eval 1679 files)

The filenames for the train data follow the format [dataset]_[anomaly type]_[data_index].csv, where:

• [dataset] indicates the data category (train / eval / test).

• [anomaly_type] indicates the type of anomaly (normal / inner / outer / ball).

• [data_index] is a unique identifier representing the index of the data.

• Test Data (Total 1680 files)

The filenames for the test data follow the format [dataset]_[data_index].csv.

• File Structure

Each .csv file contains 1-second length data, with two columns: bearingB_x and bearingB_y, representing the x-axis and y-axis accelerometer data, respectively.

• Added Noise

To create a dataset similar to real-world industrial environments, Weibull Noise and background noise from the ToyADMOS dataset have been added to the vibration signals at various signal-to-noise ratios.

Weibull Noise consists of signals with a signal-to-noise ratio (SNR) between [-5, 5]dB, shape factors ranging from 1.5 to 2, and scale factors ranging from -2 to 2.

Additionally, ToyADMOS Noise has an SNR of [-5, 5]dB, with noise randomly selected from over 380 different types of signals added to both the x-axis and y-axis data.

Data Specification (Train + Evaluation Dataset)