Shanze Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shanze Wang has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
- He Li
School of Automation, Shenyang Aerospace University, Shenyang, China
3 shared publications - Guanglei Meng
College of Automation, Shenyang Aerospace University, Shenyang 110136, China
1 shared publication Quan-Yong Fan
School of Automation, Northwestern Polytechnical University, Xian, China
1 shared publication- YangQuan Chen
University of California, Merced
1 shared publication - Yanmei Liu
Shenyang Aerospace University, No.37 Daoyi South Avenue, Daoyi District, Shenyang, 110136, Liaoning Province, China
1 shared publication
Latest Publications
- 22 March 2026
- Published by Elsevier in Neural Networks
- Vol. 201, 108880
Fractional-order gradient descent learning for Elman neural networks
- 22 March 2026
- Published by Elsevier in Neural Networks
- Vol. 201, 108880
Fractional-order gradient descent learning for Elman neural networks
- 4 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (2) , 3354-3367
A Dual-Channel Deep Learning Model for Fault Diagnosis Fusing Time and Fractional Domains
To address the limited feature representation capability of existing time-frequency methods in analyzing non-stationary vibration signals and the inability of fixed transforms to capture dynamic characteristics of bearing faults, this paper proposes a dual-channel deep learning model integrating time-domain and fractional-domain features. The time-domain channel employs multi-scale convolutions with varying kernel sizes to simultaneously capture both fine-grained transient impulses and coarse-grained temporal patterns. The fractional-domain channel systematically processes nine-order fractional Fourier transforms, where temporal convolutional networks extract complex temporal dependencies through dilated causal convolutions, and channel attention mechanisms adaptively weight different fractional orders to enhance discriminative features while suppressing redundancy. This dual-channel architecture achieves synergistic modeling of both transient temporal information and non-stationary fractional-domain energy distributions, overcoming the limitations of traditional single-domain and fixed time-frequency resolution methods. Experimental validation on CWRU and SUFD datasets demonstrates superior diagnostic performance compared to existing advanced methods, with comprehensive ablation studies, multi-condition scenarios, and noise robustness tests confirming the method’s effectiveness and robustness for intelligent maintenance of rotating machinery.
- 4 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (2) , 3354-3367
A Dual-Channel Deep Learning Model for Fault Diagnosis Fusing Time and Fractional Domains
To address the limited feature representation capability of existing time-frequency methods in analyzing non-stationary vibration signals and the inability of fixed transforms to capture dynamic characteristics of bearing faults, this paper proposes a dual-channel deep learning model integrating time-domain and fractional-domain features. The time-domain channel employs multi-scale convolutions with varying kernel sizes to simultaneously capture both fine-grained transient impulses and coarse-grained temporal patterns. The fractional-domain channel systematically processes nine-order fractional Fourier transforms, where temporal convolutional networks extract complex temporal dependencies through dilated causal convolutions, and channel attention mechanisms adaptively weight different fractional orders to enhance discriminative features while suppressing redundancy. This dual-channel architecture achieves synergistic modeling of both transient temporal information and non-stationary fractional-domain energy distributions, overcoming the limitations of traditional single-domain and fixed time-frequency resolution methods. Experimental validation on CWRU and SUFD datasets demonstrates superior diagnostic performance compared to existing advanced methods, with comprehensive ablation studies, multi-condition scenarios, and noise robustness tests confirming the method’s effectiveness and robustness for intelligent maintenance of rotating machinery.
- 25 November 2025
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 36 (11) , 116208
A novel intelligent fault diagnosis method for planetary gearboxes with high accuracy under limited sample conditions
Fault diagnosis of planetary gearboxes is hindered by scarce fault samples and difficult feature extraction from complex non-Gaussian vibration signals. This paper proposes an intelligent diagnosis method combining empirical refined composite multiscale dispersion entropy (ERCMDE) and human evolutionary optimization algorithm-support vector machine (HEOA-SVM). ERCMDE employs empirical quantile symbolization to overcome Gaussian distribution assumptions in existing entropy methods, enabling robust feature extraction from non-Gaussian signals. HEOA-SVM utilizes a four-role collaborative search mechanism to achieve global parameter optimization while avoiding local optima. Experimental results on two planetary gearbox datasets demonstrate that with only two training samples per category, the proposed method achieves recognition accuracies of 98.78% and 96.57% for five gear health states, outperforming existing entropy-based and machine learning methods. This provides an effective solution for equipment health monitoring under sample-scarce conditions.
- 25 November 2025
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 36 (11) , 116208
A novel intelligent fault diagnosis method for planetary gearboxes with high accuracy under limited sample conditions
Fault diagnosis of planetary gearboxes is hindered by scarce fault samples and difficult feature extraction from complex non-Gaussian vibration signals. This paper proposes an intelligent diagnosis method combining empirical refined composite multiscale dispersion entropy (ERCMDE) and human evolutionary optimization algorithm-support vector machine (HEOA-SVM). ERCMDE employs empirical quantile symbolization to overcome Gaussian distribution assumptions in existing entropy methods, enabling robust feature extraction from non-Gaussian signals. HEOA-SVM utilizes a four-role collaborative search mechanism to achieve global parameter optimization while avoiding local optima. Experimental results on two planetary gearbox datasets demonstrate that with only two training samples per category, the proposed method achieves recognition accuracies of 98.78% and 96.57% for five gear health states, outperforming existing entropy-based and machine learning methods. This provides an effective solution for equipment health monitoring under sample-scarce conditions.