Minghang Zhao 1, Myeongsu Kang 2, Baoping Tang 1, and Michael Pecht 2
1 State Key Laboratory of Mechanical Transmission, Chongqing University, Chongqing, China
2 CALCE, Center for Advanced Life Cycle Engineering, Department of Mechanical Engineering, University of Maryland, College Park, Maryland 20740, USA
Wavelet transform, an effective tool to decompose signals into a series of frequency bands, has been widely used for vibration-based fault diagnosis in machinery. Likewise, the use of deep learning algorithms is becoming popular to automatically learn discriminative features from input data for the sake of improving diagnostic performance. However, the fact that no general consensus has been reached as to which wavelet basis functions are useful for diagnosis motivated this investigation of methods to fuse multiple wavelet transforms into deep learning algorithms. In this paper, two methods—i.e., multiple wavelet coefficients fusion in deep residual networks by concatenation and multiple wavelet coefficients fusion in deep residual networks by maximization—were developed to capture discriminative information from diverse sets of wavelet coefficients for fault diagnosis. The efficacy of the developed methods was verified by applying them to planetary gearbox fault diagnosis.