Small sample fault diagnosis method for wind turbine gearbox based on optimized generative adversarial networks

Su, Yuanhao; Meng, Lei; Kong, Xiaojia; Xu, Tongle; Lan, Xiaosheng; Li, Yunfeng

doi:10.1016/j.engfailanal.2022.106573

Cited by 39 publications

(19 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although some papers have paid attention to fault diagnosis under the limited fault samples. [19][20][21][22] Note that the existing fault diagnosis methods with limited samples generally need to introduce new technologies such as generative adversarial network, transfer learning into the classification model. These overmuch steps aggravate the difficulty of applying the method to practical fault diagnosis.…”

Section: Introductionmentioning

confidence: 99%

Gearbox fault diagnosis based on frequency-domain Gramian angular difference field and deep convolutional neural network

Zhang

Feng

et al. 2023

Proceedings of the Institution of Mechanical Engineers, Part C:

View full text Add to dashboard Cite

Gearbox fault diagnosis is of great significance to avoid accidents and reduce operation and maintenance costs for machinery such as the quayside container crane. In recent years, mechanical fault diagnosis methods based on deep learning have developed rapidly. How to perform diagnosis simply based on deep learning in the case of limited labeled samples is still a challenging problem. To address this issue, this paper proposes a new fault diagnosis method combining the frequency-domain Gramian angular difference field (FDGADF) and deep convolutional neural network (DCNN), namely, FDGADF-DCNN. To explore the availability of the FDGADF-DCNN method, two experiment datasets are studied. When the proportion of labeled samples is 70%, the recognition accuracy of both datasets exceeds 98%. When only 20% of the labeled samples are available (the average number of training samples in each category is only dozens), FDGADF-DCNN can achieve a recognition accuracy of no less than 95% in two datasets, in which a benchmark dataset is 100% and the dataset collected from a scaled-down test bench is 95%. Compared with other 11 methods, the FDGADF-DCNN method achieves the highest recognition accuracy. These results indicate that the FDGADF-DCNN method is expected to be used for gearbox fault diagnosis under limited labeled samples.

show abstract

Section: Introductionmentioning

confidence: 99%

Gearbox fault diagnosis based on frequency-domain Gramian angular difference field and deep convolutional neural network

Zhang

Feng

et al. 2023

Proceedings of the Institution of Mechanical Engineers, Part C:

View full text Add to dashboard Cite

show abstract

“…Traditional deep learning requires massive amounts of data to improve model generalization, which is constrained to a fixed task [26][27][28]. In contrast to these approaches, meta-learning provides a flexible learning mechanism, which is expected to design a model in which new knowledge can be learned through a few samples.…”

Section: Meta-learningmentioning

confidence: 99%

Prior knowledge-based self-supervised learning for intelligent bearing fault diagnosis with few fault samples

Wu,

Nie,

et al. 2023

Meas. Sci. Technol.

View full text Add to dashboard Cite

Deep learning-based bearing fault diagnosis methods have been developed to learn fault knowledge from massive data. Owing to the deficiency of fault samples and the variability of working conditions, these deep-learning-based methods are limited in industrial applications. To address this problem, this study proposes a prior knowledge-based self-supervised learning (PKSSL) method for bearing fault diagnosis. In the PKSSL method, prior diagnostic knowledge is extracted by meta-learning from a few samples. Prior diagnostic knowledge is then utilized to guide the self-supervised learning (SSL) process to reduce reliance on training data. Furthermore, a graph convolutional network is introduced to fuse the information obtained by meta-learning and SSL, which makes the model fully utilize the learned information and improves the accuracy of the fault diagnosis. The effectiveness of the proposed method was validated using two datasets. The results demonstrate that, compared to other existing approaches, the proposed method exhibits a strong generalization ability to transfer diagnostic knowledge from artificial damage data to real damage data under varied operating conditions.

show abstract

“…In recent years, researchers have widely used various neural network models to solve the problems of unbalanced datasets and small sample size, and to overcome the difficulties in learning new fault types in gearbox fault identification [3][4][5][6]. At the same time, however, they also noticed that the neural network model has a poor interpretation of fault characteristics [7].…”

Section: Introductionmentioning

confidence: 99%

Fault feature extraction for planetary bearing of CRF pump in nuclear power plant based on TFDC-QPSO-optimised MOMEDA

Zhang¹,

Xiong²,

He³

et al. 2022

Meas. Sci. Technol.

View full text Add to dashboard Cite

To address the issue of weak fault features and difficulty in feature extraction for planetary gearbox bearings of the circulation water pump (CRF pump) unit in the conventional island of a nuclear power plant, a deconvolution method, named the Multipoint Optimal Minimum Entropy Deconvolution Adjusted (MOMEDA), is optimised by a mixing operator improved Quantum behaviour Particle Swarm Optimisation (QPSO) algorithm, to extract the periodic fault impulse of planetary gearbox bearings. In the mixing operator, a Differential Evolution (DE) operator is introduced to improve particle swarms' diversity and enhance the algorithm's global optimisation capability. Meanwhile, a proposed adaptive CrossOver (CO) operator is incorporated into the algorithm to increase its convergence speed. Then, combining these two types of operators can construct a parameter optimisation algorithm displaying both global optimisation capability and high algorithm execution efficiency. Consider capturing the periodic impulsive features relating to bearing faults. A fitness function based on the characteristic information of the deconvolution signal in the time and frequency domain is proposed. This function serves as the objective function to facilitate the parameter optimisation on MOMEDA. Thereby, it enhances the optimisation capability of our feature extraction method. Experiments were conducted by adopting the proposed method on the signals collected from the inner- and outer-race faulty planetary bearing in a test bed, in which favourable feature extraction results are obtained. After the comparative analysis, we observed that both significances of features extracted by this method and the execution efficiency of the algorithm are superior compared to other methods. The on-site feature extraction results of the input shaft bearing in the planetary gearbox of a CRF pump unit in a nuclear power plant also demonstrated the engineering practicability of the method proposed in this work.

show abstract

Small sample fault diagnosis method for wind turbine gearbox based on optimized generative adversarial networks

Cited by 39 publications

References 32 publications

Gearbox fault diagnosis based on frequency-domain Gramian angular difference field and deep convolutional neural network

Gearbox fault diagnosis based on frequency-domain Gramian angular difference field and deep convolutional neural network

Prior knowledge-based self-supervised learning for intelligent bearing fault diagnosis with few fault samples

Fault feature extraction for planetary bearing of CRF pump in nuclear power plant based on TFDC-QPSO-optimised MOMEDA

Contact Info

Product

Resources

About