Bearing Dataset

The dataset comprises vibration measurements from SKF 6314 deep groove ball bearings and NJ2307 cylindrical roller bearings. Both bearing types were investigated in healthy states and with artificially introduced faults at different load points that reproduce the operating conditions at their respective positions within the SEED axle gearbox. The measurements enable systematic analysis of vibration signatures and support the development and evaluation of bearing fault diagnosis methods. This work was funded by the research projects SEED (Smart Efficient Electric Driveline, FKZ 13FH585KX0) and KI-Werkstatt (FEIH_2486678). The figures below illustrate the design and technical specifications of both bearing types.

The measurements were performed on an APS 58 bearing test rig, which consists of an electric drive unit, a bearing housing, and two pneumatic actuators for applying radial and axial loads. The applied forces were measured using HBM U2B tension/compression force transducers, with radial and axial measurement ranges of ±20 kN and ±10 kN, respectively. The force signals were processed and recorded using MATLAB/Simulink.

The rotational speed was measured with an Autosen A0014 optical sensor at a sampling frequency of 1 kHz. Vibration data were acquired using a PCB 603C01 ICP piezoelectric accelerometer mounted on the bearing housing and a National Instruments (NI) data acquisition system.

For the 6314 deep-groove ball bearings, vibration signals were recorded at sampling frequencies of 51.2 kHz, 25.6 kHz, and 12.8 kHz over a duration of 120 s per measurement. For the NJ2307 cylindrical roller bearing, vibration signals were recorded at 10 kHz over a duration of 60 s per measurement.

Both bearing types were tested on the same test rig and under the same load points. Their corresponding bearing housings were used for deep-groove ball bearings and cylindrical roller bearings during testing respectively. 

The operating conditions were derived from the gearbox input torques of the SEED axle gearbox. Based on a free-body analysis of the drive shaft, two combined load cases were defined: 3257 N radial and 1556 N axial and 4893 N radial and 2337 N axial, corresponding to gearbox input torques of 215 Nm and 323 Nm, respectively. These load cases were applied at rotational speeds of 1000 rpm and 2000 rpm.

The following conditions were investigated:

• Healthy bearing
• Laser-induced defect on the inner raceway
• Laser-induced defect on the outer raceway for ball bearing (Type 6314) and boring based defect on the outer raceway for cylindrical roller bearing (Type NJ2307)
• Corrosion-induced damage (Only bearing type 6314)

The dataset supports the development and evaluation of vibration-based methods for bearing fault detection, fault classification, and anomaly detection. It is suitable for signal processing, feature extraction, conventional machine learning, and deep learning.

The combination of multiple sampling frequencies, operating points, bearing conditions, and bearing types also enables studies on data augmentation, generative models, and the transferability of diagnostic methods between the type 6314 deep groove ball bearing and type NJ2307 cylindrical roller bearing.