A Prognostics Approach for Gearbox based on Spectrogram and Deep Learning

##plugins.themes.bootstrap3.article.main##

##plugins.themes.bootstrap3.article.sidebar##

Published Jul 14, 2017
Dengwei Song Hua Su Jian Ma Likun Chao Yu Ding Chen Lu

Abstract

This paper presents a method and solution for Asia Pacific Conference of the Prognostics and Health Management 2017 Data Challenge. Considering the complex of gearbox signal, a novel approach utilizing spectrogram and deep learning is proposed. Spectrogram can transform the raw signal into time-frequency space that is an effective method to process vibration signal of gearbox. Deep learning is a useful method that has the ability of automatically feature learning. The spectrogram is obtained by using short-time Fourier transform (STFT). Then, the way of the stacked auto-encoders (SAE), is trained to auto-extract feature and realize the prediction. For diagnosis, the cepstrum algorithm is performed to simplify the spectrum and find the most severely-degraded component. This solution earned the 3rd highest score of the results from all teams in the competition, which demonstrate the effective of proposed method.

Abstract 47 | PDF Downloads 59

##plugins.themes.bootstrap3.article.details##

Keywords

PHM

References
[1] Lampert, Thomas A., and S. E. M. O'Keefe. "On the detection of tracks in spectrogram images." Pattern Recognition 46.5(2013):1396-1408
[2] Ma, J., (2015). Study on Fault Diagnosis of Electromechanical Products Based on Cognitive Computing: A Deep Learning and Transfer Learning method. Doctoral dissertation. Beihang Uniersicy, Beijing, China.
[3] Liu, H., Li, L., & Ma, J. (2016). Rolling bearing fault diagnosis based on stft-deep learning and sound signals. , 2016(2), 12.
[4] Randall, R. B. (2016). A history of cepstrum analysis and its application to mechanical problems. Mechanical Systems & Signal Processing.
[5] Liang, B., Iwnicki, S. D., & Zhao, Y. (2013). Application of power spectrum, cepstrum, higher order spectrum and neural network analyses for induction motor fault diagnosis. Mechanical Systems & Signal Processing, 39(1–2), 342-360.
Section
Data Challenge Papers