A novel hybrid method for fault diagnosis of two rolling bearings mounted on the same shaft

Abstract

The mechanical system reliability depends on efficiency incipient fault detection of rolling bearings. Traditionally, fault detection based on signal processing methods from accelerometers, which are mounted on the housing. In contrast, accelerations of the rotating shaft are more sensitive to incipient faults of the bearing. However, the design of some mechanisms limits the measuring accelerations of the shaft near each bearing. This paper proposes a novel hybrid processing shaft accelerations method for faulting diagnosis of several rolling bearings fixed on the same shaft. The method executes mixed processing of linear and angular accelerations by CNN and MLP. The method allows to detect and localize defects of each bearing based on signals from the one measured point. The experimental results show the method allows to detect the faulty bearing with an accuracy of 99,3%. Moreover, the method allows localizing the bearing fault with an accuracy of 97%.

Keywords

• Rolling bearing
• Fault detection
• Angular acceleration
• Wireless acceleration sensor
• CNN
• MLP