Bearing Fault Diagnosis Integrating Improved Adaptive and Sparest Time-Frequency Analysis with 2D-CNN

Time-frequency analysis is necessary in machine monitoring and fault diagnosis. Adaptive and sparest time-frequency analysis (ASTFA) is recently developed method, which is enlightened by empirical mode decomposition (EMD) and compressive sensing theory. However, it is a hard challenge to set proper initial values in ASTFA to obtain accurate intrinsic mode functions(IMFs) .Thus, an improved ASTFA is introduced by using artificial chemical reaction optimization algorithm, named as ACROA-ASTFA.Research shows that two-dimensional convolution neural networks (2D-CNN) have extreme image processing capabilities. In this paper, a fault diagnosis model integrating the ACROA-ASTFA with 2D-CNN is present. Firstly, a vibration signal is decomposed into a number of intrinsic mode functions (IMFs) by the ACROA-ASTFA. Secondly, the time-frequency diagram is calculated to demonstrate the information related to fault types. Thirdly, and 2D-CNN is exploited to capture useful information and fulfill intelligent fault diagnosis. Finally, experimental application is made to verify its advantage. The results show that the proposed model can achieve fault diagnosis for rolling bearings as a total accuracy of 100% for multiple classification.