A novel wind turbine fault diagnosis method based on compressed sensing and DTL-CNN

Significance 

With the dramatic growth in the consumption of wind energy, significant efforts have been devoted to improving wind energy generation and output. The efficiency and power generation capacity of wind firms are directly dependent on the normal operation of wind turbines. Presently, a significant portion of wind turbine-related mechanical faults (about 30%) are caused by rolling bearings, the most widely used mechanical component in this field. Therefore, developing effective strategies for rapid and accurate detection of rolling bearing faults is highly desirable.

Typically, rolling bearing fault signals are periodic shock signals that exhibit nonlinear and non-stationary characteristics. Thus, they cannot be accurately analyzed using conventional signal processing methods based on stationary state signal processing. To overcome this problem, a number of new signal analysis theories, such as wavelet transform and wavelet analysis, have been developed. Although these methods show good performance in processing non-stationary nonlinear signals, they still have some drawbacks.

Previous studies have revealed that compressed sensing (CS) technology offers numerous advantages in feature extraction and signal processing and is a promising method for extracting more information from rolling bearing vibration signals. Moreover, the relevant fault characteristics extracted from the original vibration signal can be imported into shallow machine-learning models like Decision Tree for further analysis. However, these machine learning models often fail to provide the anticipated results, especially when dealing with large and complex data sets. Consequently, their optimized counterparts also have some limitations despite showing improved mechanical fault detection. Deep learning, especially convolution neural network (CNN), has recently been identified as a promising approach, owing to its efficiency in solving different signal types.

To overcome the above challenges, Mr. Yan Zhang, Professor Wenyi Liu, Mr. Xin Wang and Mr. Heng Gu from Jiangsu Normal University developed a fault diagnosis method based on compressed sensing and a combination of deep transfer learning and CNN (DTL-CNN) to identify various faults in gears and rolling bearings of wind turbines. In their approach, compressed sensing technology was used to process the vibration signals collected during wind turbine bearing operation. The processed data was then imported into the newly designed DTL-CNN for fault diagnosis. Their work is currently published in the journal, Renewable Energy.

The authors showed that the new CNN was successfully used to implement the DTL, thereby enhancing the model’s capability in processing non-stationary nonlinear signals of the rolling bearing faults. As a result, the proposed diagnosis model could achieve rapid and more accurate fault diagnosis and identification with only a few steps and a small batch of gear and rolling bearing data. The compressed sensing algorithm exhibited high efficiency in sparsing the vibration data of the equipment and subsequently eliminating the interference and redundant parts of the data. Consequently, it helped with the reconstruction of the signals to achieve complete and clean vibration signals. This allowed the maximization of the real signal operating state of the wind turbine.

In summary, the development of a novel vibration processing and fault diagnosis method for rapid and accurate wind turbine bearing-to-gear and bearing-to-bearing transfer fault diagnosis was reported. The performance of this model was validated by comparing the results with those obtained via numerical experiments from the perspectives of data distribution, data density, information amount and time-domain curve. A good agreement was reported. In a statement to Advances in Engineering, Professor Wenyi Liu, the lead and corresponding author stated that the proposed method would not only contribute to rapid and accurate gear and rolling bearing fault detection in wind turbines but also improve the overall performance of such turbines.

About the author

Yan Zhang received the B. S. degree in automation from North China Institute of Science and Technology, in 2019. He is currently pursuing the M.S. degree in mechanical engineering at Jiangsu Normal University in Xuzhou, Jiangsu Province, China.

His research interest includes deep learning for rotating machinery intelligent fault diagnosis and health monitoring, signal processing and transfer learning.

About the author

Dr. Wenyi Liu received the Ph.D. degree in Mechatronic Engineering in 2010 from Chongqing University in China. He joined the School of Mechatronic Engineering, Jiangsu Normal University as an Associate professor in 2011. He is a visiting researcher in Case Western Reserve University between Jan 2013 and Jan 2014.

He is the Associate dean of School of Mechatronic Engineering, Jiangsu Normal University from Jan 2019.

He is a fellow of the Institution of Electronics and Telecommunication Engineers, a council of Chinese Institute on Measurement technology in Mechanical Engineering (CIMTME), a member of IEEE. He is also a Member of CSVE, Senior Member of CMES.

He is also Young and middle-aged science and technology leaders of the fifth 333 project in Jiangsu Province, outstanding young backbone teachers of “Qinglan Project” in Jiangsu province, 10th director of national college Mechanical Engineering Testing Technology Research Association, 8th director of fault diagnosis Professional Committee of China vibration engineering society, 7th director of dynamic testing professional committee of China vibration engineering society.

E-mail: [email protected], [email protected]

Reference

Zhang, Y., Liu, W., Wang, X., & Gu, H. (2022). A novel wind turbine fault diagnosis method based on compressed sensing and DTL-CNN. Renewable Energy, 194, 249–258.

Go To Renewable Energy

Check Also

A fundamental approach to understanding turbulent premixed flames - Advances in Engineering

A fundamental approach to understanding turbulent premixed flames