Marine Water Pump Bearing Fault Feature Enhancement and Diagnosis Base on BM-MTF

Objectives Marine water pump bearings operate in a complex environment, the fault features of the acquired are easily submerged by noise, resulting in low fault diagnosis accuracy. This paper proposed a Butterworth Mean Filtering Markov Transition Field (BM-MTF) technique combined with the ResNet18 network to solve the problem. Methods First, the BM filter is employed to improve the fault impulse waveform of the signal, suppress the interference of noise and enhance the fault characteristics; then, the two-dimensional image is drawn through MTF to effectively visualize and enhance the signal characteristics. Then, the MTF images after BM signal filtered are input into the ResNet18 network for fault diagnosis. Finally, the method is verified by the public bearing fault dataset of Western Reserve University, the laboratory bearing fault dataset, the marine water pump bearing fault dataset, and compared with other methods. Results The results on three bearing fault data sets show that the accuracy of the proposed method is 100%. The comparative experiment show the proposed method can effectively extract fault features and has higher recognition accuracy. Conclusions This paper can provide a new method for fault diagnosis of marine water pump bearings.