Intelligent bearing fault diagnosis method combining mixed input and hybrid CNN-MLP model

Abstract

Rolling bearings are one of the most widely used bearings in industrial machines. The technical condition of the rolling bearings has a significant impact on the total condition of rotating machinery. Typically, the rolling bearing condition monitoring is based on signal processing from accelerometers mounted on the machine housing. AI-based signal processing methods allow achieving high results in the diagnosis of rolling bearings. Moreover, hybrid NN-based signal processing methods provide the best diagnostic results. However, the diagnostic value of each signal from the housing-mounted accelerometers is highly depend on the location of the corresponding accelerometer. On the other hand, mounting a sensor on the rotating shaft allows expanding and increasing the signal diagnostic value. This paper proposes a novel hybrid CNN-MLP model-based diagnostic method which combines mixed input to perform rolling bearing diagnostics. The method successfully detects and localizes bearing defects using acceleration data from a wireless acceleration sensor which is mounted on a rotating shaft of the machine. The detection and localization efficiency of the hybrid model with various size datasets (more than 27 k samples and about 1900 samples) for training was estimated. Moreover, the detection and localization efficiency of the trained hybrid model with a dataset of another shaft rotating speed for testing was estimated. The experimental results show that applying the hybrid model allows detecting and localizing the bearing faults with up to 99.6% accuracy.