[1]LI Feilong,HE Weihui,LIU Lifang,et al.Real time fault diagnosis method of rolling bearing based on CWT and LightweightNet[J].CAAI Transactions on Intelligent Systems,2023,18(3):496-505.[doi:10.11992/tis.202204020]
Copy
CAAI Transactions on Intelligent Systems[ISSN 1673-4785/CN 23-1538/TP] Volume:
18
Number of periods:
2023 3
Page number:
496-505
Column:
学术论文—机器感知与模式识别
Public date:
2023-07-05
- Title:
-
Real time fault diagnosis method of rolling bearing based on CWT and LightweightNet
- Author(s):
-
LI Feilong1; HE Weihui2; LIU Lifang1; QI Xiaogang3
-
1. School of Computer Science and Technology, Xidian University, Xi’an 710071, China;
2. Xi’an Satellite Control Center, Xi’an 710049, China;
3. School of Mathematics and Statistics, Xidian University, Xi’an 710071, China
-
- Keywords:
-
rolling bearing; fault diagnosis; continuous wavelet transform; time-frequency feature; lightweight neural network; group convolution; channel shuffle; inverted residual structure
- CLC:
-
TP183
- DOI:
-
10.11992/tis.202204020
- Abstract:
-
In order to solve the problem of large computation and high cost when common deep learning algorithm is applied to bearing fault diagnosis and classification, a real-time rolling bearing fault diagnosis method combining continuous wavelet transform and lightweight neural network is proposed in this paper. Firstly, the Morlet mother wavelet function is used to carry out continuous wavelet transform on the bearing vibration acceleration data, extracting the time-frequency domain features and converting the one-dimensional signals into two-dimensional images. Then, LightweightNet, a lightweight convolutional neural network, is designed for time-frequency image fault classification by combining lightweight neural network design elements such as group convolution, channel shuffle and inverted residual structure. LightweightNet not only ensures sufficient feature extraction ability, but also has lightweight characteristics. The bearing failure experiment data sets from Case Western Reserve University show that, compared with other methods using classic lightweight neural network, this method has less parameters, the highest degree of accuracy and faster diagnosis speed, the real-time fault diagnosis of rolling bearing can be achieved basically, far less than other similar methods in memory consumption and model storage space.