Siyu Zhu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Siyu Zhu has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Yuening Zhu
Beijing Jiaotong University
3 shared publications - Lu Wei
Beihang School, Beihang University, Beijing 100191, China
2 shared publications - Wei Shangguan
School of Automation and Intelligence, Beijing Jiaotong University, Beijing, China
2 shared publications - Zheng Qian
Beihang University
2 shared publications - Baigen Cai
Beijing Jiaotong University
1 shared publication - Cong Peng
School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing, PR China
1 shared publication
Latest Publications
- 13 November 2026
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 37 (45) , 456108
Reliability-regulated evidential fusion for multimodal diagnosis of wind turbine mechanical systems under nonstationary operation and noise perturbations
Fault diagnosis of wind-turbine mechanical components under variable operating conditions is complicated by the condition-dependent diagnostic contributions of vibration and acoustic-emission (AE) measurements and by noise-induced corruption of multimodal evidence. This study proposes a reliability-regulated multimodal evidential framework that incorporates latent distributional familiarity into evidence formation and fusion. Modality-specific encoders retain the distinct signal characteristics of vibration and AE measurements, while Gaussian–Bernoulli restricted Boltzmann machines estimate their distributional familiarity from latent free energy. The resulting reliability factors regulate Dirichlet evidential opinions and discount class-specific support before Dempster–Shafer aggregation, yielding fused diagnostic belief and fused evidential uncertainty. The framework was evaluated on a wind-driven planetary gearbox under nonstationary speed excitation and a pitch-bearing benchmark under time-varying multi-axial loading. The fused model achieved a diagnostic accuracy of 98.30 % in the planetary-gearbox task. The pitch-bearing waveform-input model achieved accuracy of 99.650 ± 0.349 % and Macro- F 1 of 99.648 ± 0.353 % under five-fold temporal-block evaluation. The original class-wise comparisons showed higher performance than five representative multimodal methods. Under controlled additive white Gaussian noise perturbations, the proposed method retained stronger and more uniform fault discrimination across the evaluated signal-to-noise ratios, while the fused evidential uncertainty shifted overall toward higher values as the perturbation increased. These results demonstrate that reliability discounting and evidential fusion provide an effective approach to integrating heterogeneous vibration and AE measurements for wind-turbine mechanical fault diagnosis under variable and noise-perturbed operating conditions.
- 13 November 2026
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 37 (45) , 456108
Reliability-regulated evidential fusion for multimodal diagnosis of wind turbine mechanical systems under nonstationary operation and noise perturbations
Fault diagnosis of wind-turbine mechanical components under variable operating conditions is complicated by the condition-dependent diagnostic contributions of vibration and acoustic-emission (AE) measurements and by noise-induced corruption of multimodal evidence. This study proposes a reliability-regulated multimodal evidential framework that incorporates latent distributional familiarity into evidence formation and fusion. Modality-specific encoders retain the distinct signal characteristics of vibration and AE measurements, while Gaussian–Bernoulli restricted Boltzmann machines estimate their distributional familiarity from latent free energy. The resulting reliability factors regulate Dirichlet evidential opinions and discount class-specific support before Dempster–Shafer aggregation, yielding fused diagnostic belief and fused evidential uncertainty. The framework was evaluated on a wind-driven planetary gearbox under nonstationary speed excitation and a pitch-bearing benchmark under time-varying multi-axial loading. The fused model achieved a diagnostic accuracy of 98.30 % in the planetary-gearbox task. The pitch-bearing waveform-input model achieved accuracy of 99.650 ± 0.349 % and Macro- F 1 of 99.648 ± 0.353 % under five-fold temporal-block evaluation. The original class-wise comparisons showed higher performance than five representative multimodal methods. Under controlled additive white Gaussian noise perturbations, the proposed method retained stronger and more uniform fault discrimination across the evaluated signal-to-noise ratios, while the fused evidential uncertainty shifted overall toward higher values as the perturbation increased. These results demonstrate that reliability discounting and evidential fusion provide an effective approach to integrating heterogeneous vibration and AE measurements for wind-turbine mechanical fault diagnosis under variable and noise-perturbed operating conditions.
- Open Access
- 30 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 52-59
An Adaptive Multi-Scale Fusion and Temporal Attention-Driven Method for Gearbox Fault Diagnosis
- 30 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 52-59
Open AccessAn Adaptive Multi-Scale Fusion and Temporal Attention-Driven Method for Gearbox Fault Diagnosis