Journal
IEEE ACCESS
Volume 7, Issue -, Pages 149736-149749Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/ACCESS.2019.2947194
Keywords
Fault diagnosis; Gallium nitride; Training; Generative adversarial networks; Feature extraction; Generators; Fault diagnosis; rotating machinery; generative adversarial network; small sample size conditions
Categories
Funding
- National Natural Science Foundation of China [61973011, 61803013, 61903015]
- Fundamental Research Funds for the Central Universities [ZG140S1993]
- National key Laboratory of Science and Technology on Reliability and Environmental Engineering [6142004180501, WDZC2019601A304]
- China Postdoctoral Science Foundation [2019M650438]
Ask authors/readers for more resources
Rotating machinery plays a key role in mechanical equipment, and the fault diagnosis of rotating machinery is a popular research topic. To overcome the dependency on expert knowledge regarding conventional time-frequency analysis diagnosis methods, machine learning (ML) and artificial intelligence (AI)-based methods are commonly studied. Although these methods can achieve high-accuracy diagnosis results, they are based on a large number of training samples. A generative adversarial network (GAN) is an algorithm with the capability of generating realistic samples that are similar to the real samples, and it can be applied to solve fault diagnosis problems with insufficient training data, which is called the small sample size condition in this study. However, a single-GAN model cannot achieve a good diagnostic result. To achieve adaptive feature extraction and high diagnosis accuracy, this study proposes an intelligent fault diagnosis method for rotating machinery based on GANs under small sample size conditions. The effectiveness and performance of the proposed method are validated using rolling bearing and gearbox datasets. In these datasets, only 10 and 20 of the samples are selected as the training data. Samples associated with different health conditions and various working conditions are included in the datasets. Compared with those of other diagnosis methods, the high-accuracy and low-volatility diagnosis results indicate that the proposed method can stably distinguish fault modes under different working conditions in an adaptive way, even though few training samples are available.
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