Journal
IEEE ACCESS
Volume 8, Issue -, Pages 63027-63042Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/ACCESS.2020.2981289
Keywords
Fault diagnosis; 1-D stacked dilated convolutional neural network (1D-SDCNN); domain adaptation; variable working loads
Categories
Funding
- National Nature Science Foundation of China [61104030, 61672430]
- Fundamental Research Funds for the Central Universities [3102019ZX031]
- Shaanxi Key Research and Development Program [2019ZDLGY14-02-01]
- Research Funds for Interdisciplinary Subject of NPU [19SH030401]
Ask authors/readers for more resources
Data-driven fault diagnosis is critical for the rolling bearing to improve its healthy states and save invaluable cost. Nowadays, various intelligent fault diagnosis methods using machine learning (ML) or deep learning (DL) techniques have achieved much success. The convolutional neural network (CNN) based method, as a representative DL technique, can extract the features of raw data automatically for its excellent sparse connectivity and weight sharing properties. In this paper, a novel data-driven intelligent fault diagnosis method of rolling bearing under variable working loads has been proposed by using 1-D stacked dilated convolutional neural network (1D-SDCNN). First, 1-D vibration signals were used as input data without additional signal processing and diagnostic expertise. Second, the stacked dilated convolution, which can capture larger scale associated information and achieve large receptive fields with a few layers, was used to replace the traditional convolution and pooling structure. Third, the 1D-SDCNN architecture was flexible which is based on the relationship between receptive fields and the length of the input signal. And the number of network layers can be adjusted according to signal length. Further, it can adapt to the changing working loads of the mechanical environment. Finally, the effectiveness of the proposed method was confirmed through the experiment. And the results demonstrated that 1D-SDCNN was able to learn in-deep features under three variable working loads and the average accuracy was 96.8 & x0025;.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available