Fault diagnosis of Planetary Gearbox under Time-varying Speed Conditions Based on Convolutional Neural Network

EasyChair Preprint 977

Abstract

Fault diagnosis of planetary gearboxes under time-varying running conditions is a highly challenging topic due to the frequency complexity and time variability of vibration signals. Conventional statistics are unsuitable to describe such nonstationary signals. Time-frequency analysis can extract the frequency components of nonstationary signals and their time variability, but expertise knowledge is required. In order to address the issue of fault diagnosis under time-varying conditions, an intelligent fault diagnosis method is proposed, by exploiting the capability of convolutional neural networks in image processing. Firstly, the time-frequency representations of signals are constructed, and are treated as images. Such images are compressed, and their RGB weighted averages are used for further image processing. Secondly, a convolutional neural network (CNN) is established for intelligent fault pattern identification. Convolutional calculation is exploited to adaptively extract the features of time-frequency images, and a multi-layer perceptron network is trained to diagnose planetary gearbox faults under time-varying speeds. The proposed method is validated experimentally.

Keyphrases: convolutional neural work, fault diagnosis, planetary gearbox, time-varying speed

@booklet{EasyChair:977,
  author    = {Chuan Zhao and Zhipeng Feng},
  title     = {Fault diagnosis of Planetary Gearbox under Time-varying Speed Conditions Based on Convolutional Neural Network},
  howpublished = {EasyChair Preprint 977},
  year      = {EasyChair, 2019}}