This bearing datasets has high data quality and obvious fault characteristics, so it is a commonly used bearing fault diagnosis standard dataset. In this datasets, three unbalanced datasets under different loads are constructed to testify the recognition effect of the proposed method. The test bench is composed of 2HP (1.5KW) induction motor, fan end bearing, driver end bearing, torque translator and load motor. By using EDM technology, single point faults with different depths were machined on the inner race, outer race and rolling element of the test bearing. The fault diameters were 7 mils, 14 mils and 21 mils, respectively.
The vibration signals of the bearings were collected under four different loads, so four datasets A, B, C, and D were formed. Each dataset contains five different working states of bearings: normal, rolling element failure (BF), inner race failure (IF), and outer race failure (OF). where BF, IF, and OF contain three states with fault diameters of 7mils, 14mils and 21mils, respectively. Therefore, each dataset contained ten different samples.