| Author Name | Affiliation | Postcode | | Weizhong Zhang | School of Mechatronics Engineering, Nanjing Forestry University, Nanjing 210037, China | 210037 | | Xiaoan Yan* | School of Mechatronics Engineering, Nanjing Forestry University, Nanjing 210037, China | 210037 | | Maoyou Ye | School of Mechatronics Engineering, Nanjing Forestry University, Nanjing 210037, China | 210037 | | Xing Hua | School of Mechatronics Engineering, Nanjing Forestry University, Nanjing 210037, China | 210037 | | Dong Jiang | School of Mechatronics Engineering, Nanjing Forestry University, Nanjing 210037, China | 210037 |
|
| Abstract: |
| Owing to the harsh conditions, wind turbine bearings are prone to faults, and the resulting fault information is easily submerged by strong noise disturbance, making conventional diagnosis challenging. Therefore, this study presents an innovative bearing fault diagnosis approach predicated on parameter-optimized symplectic geometry mode decomposition (POSGMD) and improved convolutional neural network (ICNN). Firstly, assisted by the relative entropy-based adaptive selection of embedding dimension, a POSGMD is presented to adaptively decompose the collected bearing vibration signals into various symplectic geometry components (SGC), which can solve the problem of manual selection of the embedding dimension in the raw symplectic geometry mode decomposition (SGMD). Meanwhile, the signal reconstruction on the decomposed SGC is conducted based on kurtosis-weighted principle to obtain the reconstructed signals. Subsequently, the continuous wavelet transform (CWT) of the reconstructed signals is calculated to generate the corresponding time-frequency images as sample set. Finally, an ICNN is introduced for model training and automatic recognition of bearing faults. Two case studies are used to validate the presented method""s efficacy. Comparing the presented method to traditional fault diagnosis methods, experimental results show that it can achieve greater identification accuracy and superior anti-noise resilience. This work provides a practical and effective solution for fault diagnosis in wind turbine bearings, contributing to the timely detection of faults and the reliable operation of wind turbines or other rotational machinery in industrial applications. |
| Key words: symplectic geometry mode decomposition convolutional neural network deep learning rolling bearing fault diagnosis anti-noise robustness |
| DOI：10.11916/j.issn.1005-9113.2024102 |
| Clc Number:TM315 |
| Fund: |
|
| Descriptions in Chinese: |
| Owing to the harsh conditions, wind turbine bearings are prone to faults, and the resulting fault information is easily submerged by strong noise disturbance, making conventional diagnosis challenging. Therefore, this study presents an innovative bearing fault diagnosis approach predicated on parameter-optimized symplectic geometry mode decomposition (POSGMD) and improved convolutional neural network (ICNN). Firstly, assisted by the relative entropy-based adaptive selection of embedding dimension, a POSGMD is presented to adaptively decompose the collected bearing vibration signals into various symplectic geometry components (SGC), which can solve the problem of manual selection of the embedding dimension in the raw symplectic geometry mode decomposition (SGMD). Meanwhile, the signal reconstruction on the decomposed SGC is conducted based on kurtosis-weighted principle to obtain the reconstructed signals. Subsequently, the continuous wavelet transform (CWT) of the reconstructed signals is calculated to generate the corresponding time-frequency images as sample set. Finally, an ICNN is introduced for model training and automatic recognition of bearing faults. Two case studies are used to validate the presented method""s efficacy. Comparing the presented method to traditional fault diagnosis methods, experimental results show that it can achieve greater identification accuracy and superior anti-noise resilience. This work provides a practical and effective solution for fault diagnosis in wind turbine bearings, contributing to the timely detection of faults and the reliable operation of wind turbines or other rotational machinery in industrial applications. |
