Zhiyong Hu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Zhiyong Hu has a total of 66 co-authors. Here are the top co-authors by number of shared publications:
- Juncai Song
Anhui University
14 shared publications - Siliang Lu
Anhui University
14 shared publications - Xiaoxian Wang
Anhui University
14 shared publications - Lei Mao
Hefei University of Technology, Hefei, Anhui, China
6 shared publications - Chao Wang
Department of Industrial and Systems Engineering University of Iowa Iowa City IA USA
5 shared publications - Xianhong Wu
College of Electrical Engineering and Automation, Anhui University, Hefei, China
3 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (22) , 34468-34481
A Lightweight Model Based on Time--Frequency Enhancement and Multisensor Fusion for Motor Fault Diagnosis
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (22) , 34468-34481
A Lightweight Model Based on Time--Frequency Enhancement and Multisensor Fusion for Motor Fault Diagnosis
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 37 (11) , 6194-6206
Hybrid Transfer Active Learning for Multistream Processes With Within-Process and Cross-Process Correlation Modeling and Online Updating
Active learning for regression (ALR) is a prevalent tool for learning functional relationships by selectively incorporating the most informative data. However, existing ALR methods suffer from the cold-start problem and focus solely on learning single-stream functions or processes. In this work, a hybrid transfer learning framework is proposed to model within-process and cross-process functional correlations for resolving the cold-start problem in multistream ALR. In this framework, a novel multioutput Gaussian process (MGP) covariance structure is proposed to characterize both within-process and cross-process correlations, and offline learning and online updating are strategically integrated to leverage these correlations for ALR. Specifically, offline learning utilizes cross-process correlation to transfer knowledge from related source processes, providing a robust prior for the target process, while online updating incorporates the offline-learned prior with newly collected data to iteratively update the within-process functional relationship in the target process. Theoretical justifications are provided for the proposed framework, showing that the ALR error decreases monotonically as new data are acquired under the guidance of the proposed learning structure. The accuracy and superiority of the proposed framework are thoroughly verified through comparisons with benchmark methods on various numerical and real-case studies.
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 37 (11) , 6194-6206
Hybrid Transfer Active Learning for Multistream Processes With Within-Process and Cross-Process Correlation Modeling and Online Updating
Active learning for regression (ALR) is a prevalent tool for learning functional relationships by selectively incorporating the most informative data. However, existing ALR methods suffer from the cold-start problem and focus solely on learning single-stream functions or processes. In this work, a hybrid transfer learning framework is proposed to model within-process and cross-process functional correlations for resolving the cold-start problem in multistream ALR. In this framework, a novel multioutput Gaussian process (MGP) covariance structure is proposed to characterize both within-process and cross-process correlations, and offline learning and online updating are strategically integrated to leverage these correlations for ALR. Specifically, offline learning utilizes cross-process correlation to transfer knowledge from related source processes, providing a robust prior for the target process, while online updating incorporates the offline-learned prior with newly collected data to iteratively update the within-process functional relationship in the target process. Theoretical justifications are provided for the proposed framework, showing that the ALR error decreases monotonically as new data are acquired under the guidance of the proposed learning structure. The accuracy and superiority of the proposed framework are thoroughly verified through comparisons with benchmark methods on various numerical and real-case studies.
- 1 August 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (16) , 24550-24564
Position Optimization and Signal Feature Fusion of a Magnetic Sensor Array for Noise Reconstruction and Elimination
- 1 August 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (16) , 24550-24564
Position Optimization and Signal Feature Fusion of a Magnetic Sensor Array for Noise Reconstruction and Elimination
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. PP (1530437X) , 1
Remaining Useful Life Estimation of Bearings Based on Mixed Effects Model with Automatic Group Determination
Bearing is one of the most critical components in rotating machinery, whose accurate remaining useful life (RUL) estimation is essential to reduce economic loss and prevent catastrophic consequences. Due to modeling flexibility, data-driven methods have become the mainstream for RUL estimation. However, existing deep-learning based models focus on capturing the average behavior of bearings and thus cannot provide individualized estimation for a single bearing. Moreover, although existing statistical methods can take personalized characteristics into consideration, they cannot distinguish the heterogeneity among different groups, leading to degenerate estimation performance. To address this research gap, a general modelling framework, denoted as Mixed Effects model with Automatic Group Determination, i.e., AGD-ME, is proposed for the RUL estimation of bearings. Specifically, Gaussian Mixture Model is firstly introduced to automatically determine the number of groups and mixing proportion of each group; then, with the online monitoring data, the fitness of an in-service unit for each group is calculated, based on which, the degradation path is finally predicted to provide individualized RUL estimation for this specific unit. The accuracy and effectiveness of the proposed AGD-ME method is thoroughly analyzed with a run-to-failure bearing datasets by comparing to conventional mixed effects model and other advanced machine learning methods.
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. PP (1530437X) , 1
Remaining Useful Life Estimation of Bearings Based on Mixed Effects Model with Automatic Group Determination
Bearing is one of the most critical components in rotating machinery, whose accurate remaining useful life (RUL) estimation is essential to reduce economic loss and prevent catastrophic consequences. Due to modeling flexibility, data-driven methods have become the mainstream for RUL estimation. However, existing deep-learning based models focus on capturing the average behavior of bearings and thus cannot provide individualized estimation for a single bearing. Moreover, although existing statistical methods can take personalized characteristics into consideration, they cannot distinguish the heterogeneity among different groups, leading to degenerate estimation performance. To address this research gap, a general modelling framework, denoted as Mixed Effects model with Automatic Group Determination, i.e., AGD-ME, is proposed for the RUL estimation of bearings. Specifically, Gaussian Mixture Model is firstly introduced to automatically determine the number of groups and mixing proportion of each group; then, with the online monitoring data, the fitness of an in-service unit for each group is calculated, based on which, the degradation path is finally predicted to provide individualized RUL estimation for this specific unit. The accuracy and effectiveness of the proposed AGD-ME method is thoroughly analyzed with a run-to-failure bearing datasets by comparing to conventional mixed effects model and other advanced machine learning methods.
- Open Access
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-12
A Bayesian Hierarchical Framework for Non-linear and Iterative Transfer Learning of Gaussian Process
Classical multi-output GP (MGP) models often suffer from limitations such as the assumption of linear inter-process correlations, sensitivity to data imbalance, and lack of support for online updates. This paper proposes a novel Bayesian hierarchical framework to address these challenges. Specifically, we introduce a three-layer architecture that models shared hyper-parameters as random variables and updates them asymmetrically through a hierarchical structure. The proposed framework treats source data as historical information to form informative priors and uses target data for posterior update, which facilitates a flexible and process-specific knowledge transfer. We theoretically establish that this treatment induces nonlinear inference for the target process and allows efficient, iterative updates without re-optimizing the entire model. Comprehensive evaluations on synthetic data and real-world case studies in battery diagnostics and nano-sensor design demonstrate that our method significantly outperforms existing single-output and hierarchical MGP baselines in both accuracy and computational efficiency. The results suggest that our approach offers a principled and scalable solution for transfer learning under data-scarce and heterogeneous conditions.
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-12
Open AccessA Bayesian Hierarchical Framework for Non-linear and Iterative Transfer Learning of Gaussian Process
Classical multi-output GP (MGP) models often suffer from limitations such as the assumption of linear inter-process correlations, sensitivity to data imbalance, and lack of support for online updates. This paper proposes a novel Bayesian hierarchical framework to address these challenges. Specifically, we introduce a three-layer architecture that models shared hyper-parameters as random variables and updates them asymmetrically through a hierarchical structure. The proposed framework treats source data as historical information to form informative priors and uses target data for posterior update, which facilitates a flexible and process-specific knowledge transfer. We theoretically establish that this treatment induces nonlinear inference for the target process and allows efficient, iterative updates without re-optimizing the entire model. Comprehensive evaluations on synthetic data and real-world case studies in battery diagnostics and nano-sensor design demonstrate that our method significantly outperforms existing single-output and hierarchical MGP baselines in both accuracy and computational efficiency. The results suggest that our approach offers a principled and scalable solution for transfer learning under data-scarce and heterogeneous conditions.
- 22 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ASME Transactions on Mechatronics
- Vol. 31 (3) , 3665-3676
Air-Gap Asymmetry Fault Diagnosis of PMSLM Based on External Magnetic Field Data and Few-Shot Learning Framework
Air-gap asymmetry faults (AAFs) in permanent magnet synchronous linear motors (PMSLMs) increase thrust fluctuation, reduce positioning accuracy, and impair reliable operation. Existing methods exhibit insufficient feature extraction capability and heavy reliance on data. To address these drawbacks, this article presents an efficient approach for detecting AAFs in PMSLMs. First, external stray magnetic field signals corresponding to different AAFs are extracted through a noninvasive signal acquisition scheme that utilizes tunnel magnetoresistance sensors. Second, an optimized recurrence plot signal processing method is proposed to enable visual representation and feature enhancement. Next, a few-shot learning framework named efficient multiscale faster prototype network is proposed to realize precise diagnosis of AAF types, achieving a higher classification accuracy of 97.75% and a shorter inference time of 11.15 ms compared with other models. Notably, in contrast to traditional deep learning-based diagnostic methods, the proposed approach yields substantial improvements in two key aspects: namely the required training dataset size and the time incurred in its construction. Furthermore, the method has been rigorously validated via both simulations and physical motor prototype experiments.
- 22 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ASME Transactions on Mechatronics
- Vol. 31 (3) , 3665-3676
Air-Gap Asymmetry Fault Diagnosis of PMSLM Based on External Magnetic Field Data and Few-Shot Learning Framework
Air-gap asymmetry faults (AAFs) in permanent magnet synchronous linear motors (PMSLMs) increase thrust fluctuation, reduce positioning accuracy, and impair reliable operation. Existing methods exhibit insufficient feature extraction capability and heavy reliance on data. To address these drawbacks, this article presents an efficient approach for detecting AAFs in PMSLMs. First, external stray magnetic field signals corresponding to different AAFs are extracted through a noninvasive signal acquisition scheme that utilizes tunnel magnetoresistance sensors. Second, an optimized recurrence plot signal processing method is proposed to enable visual representation and feature enhancement. Next, a few-shot learning framework named efficient multiscale faster prototype network is proposed to realize precise diagnosis of AAF types, achieving a higher classification accuracy of 97.75% and a shorter inference time of 11.15 ms compared with other models. Notably, in contrast to traditional deep learning-based diagnostic methods, the proposed approach yields substantial improvements in two key aspects: namely the required training dataset size and the time incurred in its construction. Furthermore, the method has been rigorously validated via both simulations and physical motor prototype experiments.
- 20 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-11
Physics-Embedded Neural Network for the Acquisition of High-Resolution 2-D Magnetic Fields
Magnetic field signals have been widely applied for the monitoring and diagnosis of different systems. However, acquisition of high-resolution (HR) magnetic fields usually requires a huge amount of time, posing great challenge for real-time application. To address this issue, a novel strategy is proposed to acquire 2-D HR magnetic field data with advanced super-resolution (SR) imaging techniques. Specifically, the magnetic field data are first assembled into images with different resolutions; then, an image SR neural network is developed to recover HR images from low-resolution (LR) images to realize efficient acquisition of HR data. Moreover, to facilitate image SR, a physics-embedded neural network is developed, where the network-reconstructed SR image is employed to generate the physics-guided SR image to ensure the fidelity to physics laws. In the SR reconstruction network SR residual network (SRResNet), physical information is incorporated to form a physics-embedded SRResNet (PESRResNet). In this way, the proposed network can not only take advantage of advanced deep-learning techniques to realize SR imaging but also improve the generality of conventional networks. Various simulation and real-case studies have been introduced to validate the performance of our method. The experimental results demonstrate that the proposed framework has a good application prospect for the measurement of magnetic fields and provides new ideas for the future development of novel instrumentation for magnetic fields detection.
- 20 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-11
Physics-Embedded Neural Network for the Acquisition of High-Resolution 2-D Magnetic Fields
Magnetic field signals have been widely applied for the monitoring and diagnosis of different systems. However, acquisition of high-resolution (HR) magnetic fields usually requires a huge amount of time, posing great challenge for real-time application. To address this issue, a novel strategy is proposed to acquire 2-D HR magnetic field data with advanced super-resolution (SR) imaging techniques. Specifically, the magnetic field data are first assembled into images with different resolutions; then, an image SR neural network is developed to recover HR images from low-resolution (LR) images to realize efficient acquisition of HR data. Moreover, to facilitate image SR, a physics-embedded neural network is developed, where the network-reconstructed SR image is employed to generate the physics-guided SR image to ensure the fidelity to physics laws. In the SR reconstruction network SR residual network (SRResNet), physical information is incorporated to form a physics-embedded SRResNet (PESRResNet). In this way, the proposed network can not only take advantage of advanced deep-learning techniques to realize SR imaging but also improve the generality of conventional networks. Various simulation and real-case studies have been introduced to validate the performance of our method. The experimental results demonstrate that the proposed framework has a good application prospect for the measurement of magnetic fields and provides new ideas for the future development of novel instrumentation for magnetic fields detection.