Fault Diagnosis Method for Pumping Units with Small Samples Based on Improved ConvNeXt and Transfer Learning

Ze Li

doi:10.54097/w1ftcm96

Authors

Ze Li

DOI:

https://doi.org/10.54097/w1ftcm96

Keywords:

Fault Diagnosis, Pumping Unit, Dynamometer Card, Few-shot Learning, ConvNeXt, Transfer Learning, Channel Attention

Abstract

The scarcity of real fault samples in pumping units degrades the generalization of deep learning models. An improved ConvNeXt diagnosis method integrating physical feature perception and hierarchical transfer learning is proposed. First, according to the “low texture, high geometric structure” of dynamometer cards, ConvNeXt V2 is lightweight adapted and embedded with a channel attention module (CAM). By enhancing sensitivity to feature channels representing displacement and load variations, an adapted model with 22.1M parameters is constructed. Second, a two-stage “mechanism-driven pre-training and measured hierarchical fine-tuning” strategy is designed: simulation dynamometer cards generated by a dynamic model inject physical priors; during fine-tuning, shallow general features are frozen, deep semantic layers are differentially fine-tuned, and the classification head is retrained, achieving efficient cross-domain knowledge transfer. Experiments on a measured dataset containing eight working conditions show that the proposed method achieves an accuracy of 96.12% and a macro-average F1 score of 95.84%, with convergence speed approximately 30% faster than mainstream models. Ablation experiments confirm the synergistic effect of the physical perception structure and hierarchical transfer; for the extremely small-sample “sand sticking” fault, the F1 score significantly increases from 88.5% to 93.8%. Visualization analysis reveals the response mechanism of the CAM module to physical signals such as load surges, providing a high-confidence solution for industrial intelligent maintenance under small samples.

Downloads

Download data is not yet available.

References

[1] J. H. Lin, Y. Zhou, T. Ding, et al., “Fault classification evaluation method of pumping wells based on improved EfficientNet,” Science Technology and Engineering, 2023. [Online]. Availablehttps://kns.cnki. net/kcms/ detail/ detail. aspx?dbcode=CJFD&dbname=CJFDLAST2023&filename=KJYJ202311029.

[2] Z. Liu, H. Mao, C.-Y. Wu, C. Feichtenhofer, and S. Xie, “A ConvNet for the 2020s,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2022, pp. 11976–11986. doi: 10.1109/CVPR52688.2022.01594.

[3] S. W. Pan and S. J. Yang, “An overview of transfer learning in deep learning,” IEEE Trans. Cybern., 2021. doi: 10.1109/ TCYB. 2021.3106516.

[4] R. S. Wang and Y. F. Fu, “Fault diagnosis of rotating machinery using a new deep learning method with data augmentation,” Measurement, 2018. doi: 10.1016/j. measurement. 2017.11.020.

[5] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, pp. 770–778. doi: 10.1109/ CVPR. 2016.90.

[6] M. Tan and Q. V. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” in Proc. Int. Conf. Mach. Learn. (ICML), 2019, pp. 6105–6114. [Online]. Available: https: // proceedings.mlr.press/v97/tan19a.html.

[7] S. J. Pan and Q. Yang, “A comprehensive review on transfer learning,” IEEE Trans. Knowl. Data Eng., vol. 22, no. 10, pp. 1345–1359, Oct. 2010. doi: 10.1109/TKDE.2009.191.

[8] W. Sun, S. Shao, R. Zhao, R. Yan, X. Zhang, and R. Chen, “Deep transfer learning for fault diagnosis with a small number of samples,” Neurocomputing, 2020. doi: 10.1016/j. neucom. 2019. 09.088.

[9] J. Sun, C. Ding, and S. Gong, “Domain adaptation for visual recognition,” IEEE Trans. Pattern Anal. Mach. Intell., 2016. doi: 10.1109/TPAMI.2016.2570081.

[10] J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2018, pp. 7132–7141. doi: 10.1109/CVPR.2018.00745.

[11] J. Woo, J. Park, J.-Y. Lee, and I. S. Kweon, “CBAM: Convolutional block attention module,” in Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 3–19. doi: 10.1007/978-3-030-01237-3_1.

[12] X. Zhai, A. Kolesnikov, L. Beyer, et al., “ConvNeXt V2: Co-designing and scaling ConvNets with masked autoencoders,” arXiv preprint arXiv:2301.00808, 2023.

[13] J. Lv, W. Li, S. Wu, and M. Zhang, “Deep domain adaptation: A survey of the state of the art,” Neural Networks, 2019. doi: 10.1016/j.neunet.2019.04.004.

[14] T. Y. Chen, D. M. Zhang, and X. W. Sun, “A review of few-shot fault diagnosis methods for rotating machinery based on deep learning,” Journal of Vibration and Shock, 2023. [Online]. Available: https://www.sciopen.com/article/10.19693/j.issn.1673-3185.04175.

[15] X. Ding, H. Mao, H. Zhang, Q. Xie, and X. Zhang, “Improving transferability of convolutional neural networks via structural re-parameterization,” in Adv. Neural Inf. Process. Syst. (NeurIPS), 2021. [Online].Available: https:// proceedings. neurips.cc/paper/2021/hash/c8d5040e34f664a78086064f51e7ca30-Abstract.html.

[16] Y. F. Li, P. Y. Song, C. M. Zheng, et al., “Production prediction of ultra-high water-cut oil wells based on convolutional neural network and transfer learning,” Petroleum Geology and Recovery Efficiency, 2020. [Online]. Available: https:// kns. cnki.net/kcms/detail/detail.aspx?dbcode=CJFD&dbname=CJFDLAST2020&filename=YQDC202005018.

[17] P. Ren, Z. Xu, Z. Luo, and W. Liu, “Deep learning based multi-fault diagnosis of rolling bearings using 1D convolutional neural network,” Measurement, 2019. doi: 10.1016/j. measurement. 2018.10.046.

[18] J. Snell, K. Swersky, and R. S. Zemel, “Meta-learning for few-shot image classification,” in Int. Conf. Learn. Representations (ICLR).

[19] J. Zhang, J. Hu, Y. Wang, and L. Yang, “A survey on attention mechanisms in deep learning,” Pattern Recognit., 2020. doi: 10.1016/j.patcog.2020.107402.

[20] H. Li, X. Liu, M. Cao, and X. Li, “Transfer learning with deep convolutional neural networks for fault diagnosis of rolling bearings,” Appl. Soft Comput., 2019. doi: 10.1016/j. asoc. 2019. 105470.

[21] X. G. Li, Z. G. Wang, and L. Q. Sun, “Review of transfer learning methods for rolling bearing faults under variable working conditions and small samples,” Machinery & Electronics, 2024. [Online]. Available: http://www. stae.com. cn/ jsygc/article/issue/2024_0_10.