Zeyu Zhao
Zeyu Zhao has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Jing Yang
Navy Submarine College
1 shared publication - Pu Li
School of Energy and Power Engineering, Xi’an Jiaotong University, Xi’an 710049, China
1 shared publication - Qi Yuan
Fudan University, Shanghai, China and Ant Group, Hangzhou, China
1 shared publication - Siguo Wen
School of Energy and Power Engineering, Xi’an Jiaotong University , Xi’an 710049, China
1 shared publication
Latest Publications
- Open Access
- 23 September 2026
- Published by Elsevier in SSRN Electronic Journal
Hyperspherical Multiscale Centroid Alignment and Repulsion Network for Fault Diagnosis of Curvic Coupling Tie Rod Rotor
The curvic coupling tie rod rotor (CCTR) is a critical component in gas turbines, yet its discontinuous assembly induces nonlinear interface characteristics and unique faults that are hard to distinguish. To achieve effective cross domain diagnosis, this paper proposes a Hyperspherical Multiscale Centroid Alignment and Repulsion Network (HMCAS-Net). Initially, this method employs a multiscale compression and block-wise residual design to filter out domain specific noise. Subsequently, the method nonlinearly maps the extracted features onto a unit hypersphere. Within the task space of this unit hypersphere, the model dynamically constructs self-supervised unbiased centroids based on the data manifold. This mechanism pulls intra class features closer across domains while pushing inter class distributions apart both within and across domains, significantly enhancing intra class compactness and inter class separability. Finally, to elucidate the underlying failure mechanisms and corroborate the superiority of HMCAS-Net, this study establishes a CCTR fault simulation test rig. The rig simulates fundamental rotor faults and CCTR specific faults. In comparative experiments against advanced methods across 60 generalized target domains, the proposed method achieves an average accuracy of 92.98%, which outperforms other advanced approaches. This performance comprehensively demonstrates the effectiveness and robustness of this framework for domain generalization fault diagnosis in CCTRs.
- 23 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessHyperspherical Multiscale Centroid Alignment and Repulsion Network for Fault Diagnosis of Curvic Coupling Tie Rod Rotor
The curvic coupling tie rod rotor (CCTR) is a critical component in gas turbines, yet its discontinuous assembly induces nonlinear interface characteristics and unique faults that are hard to distinguish. To achieve effective cross domain diagnosis, this paper proposes a Hyperspherical Multiscale Centroid Alignment and Repulsion Network (HMCAS-Net). Initially, this method employs a multiscale compression and block-wise residual design to filter out domain specific noise. Subsequently, the method nonlinearly maps the extracted features onto a unit hypersphere. Within the task space of this unit hypersphere, the model dynamically constructs self-supervised unbiased centroids based on the data manifold. This mechanism pulls intra class features closer across domains while pushing inter class distributions apart both within and across domains, significantly enhancing intra class compactness and inter class separability. Finally, to elucidate the underlying failure mechanisms and corroborate the superiority of HMCAS-Net, this study establishes a CCTR fault simulation test rig. The rig simulates fundamental rotor faults and CCTR specific faults. In comparative experiments against advanced methods across 60 generalized target domains, the proposed method achieves an average accuracy of 92.98%, which outperforms other advanced approaches. This performance comprehensively demonstrates the effectiveness and robustness of this framework for domain generalization fault diagnosis in CCTRs.