JNU Bearing Dataset

The JNU Bearing Dataset, developed by Jiangnan University in China, is widely used in the field of fault diagnosis for rotating machinery. It contains high-resolution vibration signals collected from single-row spherical roller bearings, specifically types N205 and NU205. These signals were recorded under three different rotational speeds: 600, 800, and 1000 revolutions per minute (rpm), which allows the dataset to be used in scenarios involving varying operating conditions.

The data were sampled at a high frequency of 50 kHz, and each signal segment spans 20 seconds, offering sufficient temporal resolution to capture subtle fault characteristics. The dataset includes four primary health conditions: healthy bearings, inner-ring faults, outer-ring faults, and ball (rolling-element) faults. Each of these fault types is represented at each of the three operating speeds, resulting in a total of 12 distinct class combinations.

To prepare the data for training machine learning models, the continuous vibration signals are often segmented into smaller overlapping samples, typically containing 1,024 or 2,048 data points. Depending on the study, the number of samples per class may vary, but a common configuration involves approximately 600 to 976 samples per class, divided into training, validation, and test sets, often using a 60/20/20 split.

Despite the relatively small size of the dataset compared to larger industrial benchmarks, its high quality and structured fault conditions make it a valuable resource for evaluating the performance of deep learning models, such as convolutional neural networks, transformers, and domain adaptation methods. Researchers often augment the data through overlapping windowing or advanced generative models to enhance the robustness of their fault classification models. The variation in rotational speeds also makes the dataset well-suited for tasks involving domain generalization and transfer learning.