A novel transfer learning method for bearing fault diagnosis under different working conditions

Zou, Yisheng; Liu, Yongzhi; Deng, Jialin; Jiang, Yuliang; Zhang, Weihua

doi:10.1016/j.measurement.2020.108767

Cited by 52 publications

(18 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As a result, domain information is obfuscated and class information is retained. DANN easily implements synchronous training of the generator and discriminator by a GRL and has been borrowed in many subsequent studies [39][40][41]. In view of the insufficient robustness of the existing domain adaptation methods to deal with interference, Yu et al [42] transmitted the label prediction information to the domain discriminator to carry out condition adversarial training.…”

Section: Related Workmentioning

confidence: 99%

Step-by-step gradual domain adaptation for rotating machinery fault diagnosis

Sun¹,

Zeng²,

Wang³

et al. 2022

Meas. Sci. Technol.

View full text Add to dashboard Cite

The deep learning-based fault diagnosis methods have shown remarkable superiority in the field of PHM. However, the performance of deep neural network relies heavily on a substantial of labeled training dataset, which is rare in actual industrial scenario. Moreover, the extracted fault features under different working conditions follow different joint distribution. As a result, the deep model trained in one condition cannot be extended to others. Aiming at the existing problem, this paper proposes a Step-by-step Gradual Domain Adaptive Neural Network (SGDAN) to conduct cross-domain fault diagnosis which can realize precise alignment between source domain and target domain. Firstly, the maximum mean discrepancy is used to perform primary domain adaptation. Furthermore, two classifiers are set up to limit the discrepancy of target domain data in the classification decision. Finally, the exact alignment of class-level features is achieved by category prototype alignment. The extensive experimental results have shown the superiority and stability of the proposed method when compared with other conventional approaches.

show abstract

Section: Related Workmentioning

confidence: 99%

Step-by-step gradual domain adaptation for rotating machinery fault diagnosis

Sun¹,

Zeng²,

Wang³

et al. 2022

Meas. Sci. Technol.

View full text Add to dashboard Cite

show abstract

“…In addition, the DL model has problems of tedious parameter configuration and weak generalization ability, which greatly reduces the diagnostic performance of the model. The emergence of transfer learning (TL) [20] theory broke the deadlock that hindered the development of DL models. Incorporating TL into deep models leverages the similarities among data, models, and knowledge to enable leapfrogging between similar and even different domains.…”

Section: Introductionmentioning

confidence: 99%

A novel wind turbine fault diagnosis based on deep transfer learning of improved residual network and multi-target data

Zhang¹,

Liu²,

Gu³

et al. 2022

Meas. Sci. Technol.

View full text Add to dashboard Cite

In industrial production, the efficiency of fault diagnosis is seriously affected by such problems as lack of data, complex data types and low popularity of machine learning models. Therefore, to solve the above problems, we developed a novel wind turbine fault diagnosis model with high universality and strong generalization ability. Firstly, the one-dimensional original vibration data under different working conditions are divided into large-scale data, mid-scale data, small-scale data and class imbalance data. At the same time, all kinds of one-dimensional data are reshaped into two-dimensional form. Then, the residual network structure was redesigned to implement deep transfer learning in a pre-training-fine-tuning manner based on the sensitivity of vibration data features to the size of the convolutional kernel. In addition, a super parameter optimizer is added in this study to optimize some super parameter settings. In addition, the optimization of some hyperparameter settings was achieved by adding a hyperparameter search mechanism in this study. Finally, we used the developed deep transfer learning model for fault diagnosis of different types of target data of wind turbine. Experiments show that the model proposed in this study can show superior diagnostic performance in the face of different data sizes, unbalanced data types and various working conditions, and has excellent universality and reliability.

show abstract

“…Signal-based fault diagnosis methods can be divided into non-machine learning methods and machine learning methods. The machine learning-based methods, such as transfer learning [1,2], can effectively realize the classification and recognition of multiple faults, but require a large amount of sample data and labeling data. Among non-machine learning methods, statistical indicator-based methods that benefit from easy implementation are commonly used in bearing fault detection, such as kurtosis [3] and entropy [4,5], but they cannot accurately identify fault types and usually require some reference data.…”

Section: Introductionmentioning

confidence: 99%