Wei Kang
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
Wei Kang has a total of 29 co-authors. Here are the top co-authors by number of shared publications:
- Yu Guo
Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming, China
4 shared publications - Jiandong Li
Academy of Air and Missile Defense, Air Force Engineering University, Xi’an 710051, China
3 shared publications - Dawei Gao
School of Marine Science and Technology, Northwestern Polytechnical University, Xian, China
2 shared publications - Ke Yan
Key Laboratory of Education Ministry for Modern Design and Rotor-Bearing System, Xi’an Jiaotong University, Xi’an 710049, China
2 shared publications - Xiaoli Yu
Faculty of Science, Kunming University of Science and Technology, Kunming, Yunnan, China
2 shared publications - Xin Chen
Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming, China
2 shared publications
Latest Publications
- Open Access
- 2 July 2026
- Published by Elsevier in SSRN Electronic Journal
A Multi-Scale Adaptive Enhanced Graph Attention Network for Hydraulic System Fault Diagnosis
Hydraulic systems are critical power hubs in modern industry, yet their fault diagnosis remains inherently challenging. Although Graph Neural Networks (GNNs) excel in diagnosing rotating machinery, their direct application to hydraulic systems is hindered by strong fluid damping and non-stationary signal pulsations. These phenomena inevitably cause static graph distortion and over-smoothing in deep network architectures. To overcome these bottlenecks, we propose a Multi-scale Adaptive Graph Enhanced Graph Attention Network (MSA-EGAT). Specifically, an adaptive graph construction algorithm is formulated to dynamically optimize the nearest neighbor count K and fuzzy weight α, ensuring the topological mapping strictly aligns with fluid physical dynamics. Furthermore, the architecture integrates residual connections, multi-head attention, and multi-scale pooling to effectively suppress feature degradation. Experimental validation on a custom-built hydraulic test rig demonstrates that MSA-EGAT achieves a fault recognition accuracy exceeding 99%, successfully transferring the GNN diagnostic paradigm to fluid power systems. Comprehensive ablation studies confirm its efficacy in mitigating graph distortion and over-smoothing, while visual analysis of the constructed graphs explicitly reveals the topological manifolds of fluid state evolution, significantly improving model interpretability.
- 2 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Multi-Scale Adaptive Enhanced Graph Attention Network for Hydraulic System Fault Diagnosis
Hydraulic systems are critical power hubs in modern industry, yet their fault diagnosis remains inherently challenging. Although Graph Neural Networks (GNNs) excel in diagnosing rotating machinery, their direct application to hydraulic systems is hindered by strong fluid damping and non-stationary signal pulsations. These phenomena inevitably cause static graph distortion and over-smoothing in deep network architectures. To overcome these bottlenecks, we propose a Multi-scale Adaptive Graph Enhanced Graph Attention Network (MSA-EGAT). Specifically, an adaptive graph construction algorithm is formulated to dynamically optimize the nearest neighbor count K and fuzzy weight α, ensuring the topological mapping strictly aligns with fluid physical dynamics. Furthermore, the architecture integrates residual connections, multi-head attention, and multi-scale pooling to effectively suppress feature degradation. Experimental validation on a custom-built hydraulic test rig demonstrates that MSA-EGAT achieves a fault recognition accuracy exceeding 99%, successfully transferring the GNN diagnostic paradigm to fluid power systems. Comprehensive ablation studies confirm its efficacy in mitigating graph distortion and over-smoothing, while visual analysis of the constructed graphs explicitly reveals the topological manifolds of fluid state evolution, significantly improving model interpretability.
- 23 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-9
Research on High-Order Vese–Osher Model Based on Partial Differential Equation and Its Application in Planetary Gearbox Fault Diagnosis
Planetary gearbox is a crucial variable-speed transmission component in wind turbines, and the solar wheel is prone to local faults under harsh environmental conditions. Firstly, to avoid the influence of time-varying transmission paths, this paper preprocesses the fault vibration signal of solar wheel using window function and further transforms it to the angular domain.. Then, a high-order Vese-Osher model based on partial differential equations is subsequently developed to effectively identify the fault features of solar wheel. By introducing the mathematical expression of second derivative, this approach avoids the over-smoothing problem associated with the first derivative. While the Euler equation is commonly used to solve the total variation model, it cannot handle the G-norm, therefore, the divergence norm is considered to reformulate the model. Additionally, two weighting parameters are introduced between different regularization terms to optimize the solution, enhancing the ability of model to suppress noise in the fault-windowed vibration signal of solar wheel in the angular domain and thereby making the fault spectral lines more distinguishable. The Split Bregman algorithm is adopted to avoid directly solving high-order nonlinear partial differential equations. Furthermore, the method leverages the fast Fourier transform and generalized soft-thresholding equation to decompose the global optimization problem into several subproblems. Since higher-order cumulant exhibit superior noise suppression, the diagonal slices of third-order cumulant for the model output are extracted, and further obtain the 1.5-dimensional demodulation spectrum. This makes the fault feature of solar wheel completely prominent. Finally, The robustness of the above method in fault diagnosis of solar wheel is verified by fault test data.
- 23 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-9
Research on High-Order Vese–Osher Model Based on Partial Differential Equation and Its Application in Planetary Gearbox Fault Diagnosis
Planetary gearbox is a crucial variable-speed transmission component in wind turbines, and the solar wheel is prone to local faults under harsh environmental conditions. Firstly, to avoid the influence of time-varying transmission paths, this paper preprocesses the fault vibration signal of solar wheel using window function and further transforms it to the angular domain.. Then, a high-order Vese-Osher model based on partial differential equations is subsequently developed to effectively identify the fault features of solar wheel. By introducing the mathematical expression of second derivative, this approach avoids the over-smoothing problem associated with the first derivative. While the Euler equation is commonly used to solve the total variation model, it cannot handle the G-norm, therefore, the divergence norm is considered to reformulate the model. Additionally, two weighting parameters are introduced between different regularization terms to optimize the solution, enhancing the ability of model to suppress noise in the fault-windowed vibration signal of solar wheel in the angular domain and thereby making the fault spectral lines more distinguishable. The Split Bregman algorithm is adopted to avoid directly solving high-order nonlinear partial differential equations. Furthermore, the method leverages the fast Fourier transform and generalized soft-thresholding equation to decompose the global optimization problem into several subproblems. Since higher-order cumulant exhibit superior noise suppression, the diagonal slices of third-order cumulant for the model output are extracted, and further obtain the 1.5-dimensional demodulation spectrum. This makes the fault feature of solar wheel completely prominent. Finally, The robustness of the above method in fault diagnosis of solar wheel is verified by fault test data.
- 9 February 2026
- Published by Elsevier in Measurement
- Vol. 270
Multiband spectral amplitude modulation for cyclostationary feature extraction within instantaneous angular speed signals
- 9 February 2026
- Published by Elsevier in Measurement
- Vol. 270
Multiband spectral amplitude modulation for cyclostationary feature extraction within instantaneous angular speed signals
- 13 November 2025
- Published by Elsevier in Control Engineering Practice
- Vol. 167
Generative adversarial augmented multi-scale CNN for machine fault diagnosis
- 13 November 2025
- Published by Elsevier in Control Engineering Practice
- Vol. 167
Generative adversarial augmented multi-scale CNN for machine fault diagnosis
- 22 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-10
Total Variational Model of Structure Tensor Based on Partial Differential Equation and Its Application in Planet-Bearing Fault Diagnosis
When planet-bearing operates under a complex and variable environment, the fault signal is often coupled with signals from other transmission components and is influenced by the transmission path with time-varying characteristics. The angle domain resampling method reduces the phenomenon of spectral line smear caused by speed variation. However, the order spectral line of fault feature remains surrounded by many modulation sidebands and interference components. To address the above situation, For one thing, a novel blind convolution method (Minimum noise amplitude deconvolution) is adopted to preprocess the angular domain signal of planet-bearing inner ring. This method does not directly enhance the fault features but suppresses the noise in the marked area. For another, this paper achieves the goal of fault feature extraction for planet-bearing inner ring, a total variation model based on partial differential equations is constructed. The traditional total variation model takes advantage of the local structural regularity for matrix and treats the computed gradient directly as the regularization penalty term. This method utilizes structure tensor as the regularization term, which could sense neighborhood information. The overall structural features are described based on the local neighborhood information of matrix elements. This achieves a better effect of noise suppression. Finally, based on the above research methods, this paper demonstrates that they are superior to other research methods in fault feature extraction through the fault test data of planet-bearing inner ring.
- 22 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-10
Total Variational Model of Structure Tensor Based on Partial Differential Equation and Its Application in Planet-Bearing Fault Diagnosis
When planet-bearing operates under a complex and variable environment, the fault signal is often coupled with signals from other transmission components and is influenced by the transmission path with time-varying characteristics. The angle domain resampling method reduces the phenomenon of spectral line smear caused by speed variation. However, the order spectral line of fault feature remains surrounded by many modulation sidebands and interference components. To address the above situation, For one thing, a novel blind convolution method (Minimum noise amplitude deconvolution) is adopted to preprocess the angular domain signal of planet-bearing inner ring. This method does not directly enhance the fault features but suppresses the noise in the marked area. For another, this paper achieves the goal of fault feature extraction for planet-bearing inner ring, a total variation model based on partial differential equations is constructed. The traditional total variation model takes advantage of the local structural regularity for matrix and treats the computed gradient directly as the regularization penalty term. This method utilizes structure tensor as the regularization term, which could sense neighborhood information. The overall structural features are described based on the local neighborhood information of matrix elements. This achieves a better effect of noise suppression. Finally, based on the above research methods, this paper demonstrates that they are superior to other research methods in fault feature extraction through the fault test data of planet-bearing inner ring.
- Open Access
Rolling bearing fault diagnosis method based on generative adversarial enhanced multi-scale convolutional neural network model
In order to solve the problem that current convolutional neural networks can not capture the correlation features between the time domain signals of rolling bearings effectively, and the model accuracy is limited by the number and quality of samples, a rolling bearing fault diagnosis method based on generative adversarial enhanced multi-scale convolutional neural network model is proposed. Firstly, Gram angular field coding technique is used to encode the time domain signal of the rolling bearing and generate the feature map to retain the complete information of the vibration signal. Then, the re-sulting data is divided into a training set, a validation set, and a test set. Among them, the training set is input into the gradient penalty Wasserstein distance generation adversarial network to complete the training, and a new sample with similar features to the training sample is obtained, and then the original training set is expanded. Next, multi-scale convolution is used to extract the fault features of the extended training set, and the feature graph is normalized by example to overcome the influence of the difference in feature distribution. Finally, the attention mechanism is applied to the adaptive weighting of normalized features and the extraction of deep features, and the fault diagnosis is completed by the softmax classifier. Compared with ResNet method, the experimental results show that the proposed method has better generalization performance and anti-noise performance.
Open AccessRolling bearing fault diagnosis method based on generative adversarial enhanced multi-scale convolutional neural network model
In order to solve the problem that current convolutional neural networks can not capture the correlation features between the time domain signals of rolling bearings effectively, and the model accuracy is limited by the number and quality of samples, a rolling bearing fault diagnosis method based on generative adversarial enhanced multi-scale convolutional neural network model is proposed. Firstly, Gram angular field coding technique is used to encode the time domain signal of the rolling bearing and generate the feature map to retain the complete information of the vibration signal. Then, the re-sulting data is divided into a training set, a validation set, and a test set. Among them, the training set is input into the gradient penalty Wasserstein distance generation adversarial network to complete the training, and a new sample with similar features to the training sample is obtained, and then the original training set is expanded. Next, multi-scale convolution is used to extract the fault features of the extended training set, and the feature graph is normalized by example to overcome the influence of the difference in feature distribution. Finally, the attention mechanism is applied to the adaptive weighting of normalized features and the extraction of deep features, and the fault diagnosis is completed by the softmax classifier. Compared with ResNet method, the experimental results show that the proposed method has better generalization performance and anti-noise performance.
- Open Access
Modeling and Fault Size Estimation for Non-Penetrating Damage in the Outer Raceway of Tapered Roller Bearing
The fault quantification of a tapered roller bearing (TRB) can provide a reliable guide for predictive maintenance. Currently, damage size estimation based on vibration signals has been proposed and developed. However, most approaches are focused on the theory of ball bearings. Unlike the point contact of a ball bearing, the contact between the tapered roller and the raceway is a line contact. So, the current practices based on the micro-motion theory of ball bearings are limited when estimating the TRB’s fault. To accurately estimate the TRB’s damage size, a dynamic model of non-penetrating damaged TRB was established to research the vibration response mechanism and explain the influence. The model takes the deflection factor of a tapered roller into consideration and uses the elastohydrodynamic lubrication model to simulate the influence of the lubrication factor. Then, a revised formula for estimating the TRBs’ fault size is proposed by uncovering the relationship between the impact time and the damage location. Simulation analysis and experimental analysis prove the correctness of the dynamic model and the effectiveness of the size estimation formula.
Open AccessModeling and Fault Size Estimation for Non-Penetrating Damage in the Outer Raceway of Tapered Roller Bearing
The fault quantification of a tapered roller bearing (TRB) can provide a reliable guide for predictive maintenance. Currently, damage size estimation based on vibration signals has been proposed and developed. However, most approaches are focused on the theory of ball bearings. Unlike the point contact of a ball bearing, the contact between the tapered roller and the raceway is a line contact. So, the current practices based on the micro-motion theory of ball bearings are limited when estimating the TRB’s fault. To accurately estimate the TRB’s damage size, a dynamic model of non-penetrating damaged TRB was established to research the vibration response mechanism and explain the influence. The model takes the deflection factor of a tapered roller into consideration and uses the elastohydrodynamic lubrication model to simulate the influence of the lubrication factor. Then, a revised formula for estimating the TRBs’ fault size is proposed by uncovering the relationship between the impact time and the damage location. Simulation analysis and experimental analysis prove the correctness of the dynamic model and the effectiveness of the size estimation formula.
- Open Access
- 22 June 2022
- Published by MDPI AG in Lubricants
- Vol. 10 (7) , 132
Research on the Effect of Spindle Speed on the Softening and Hardening Characteristics of the Axial Operating Stiffness of Machine Tool Spindle
Spindle stiffness is one of the most critical indicators for evaluating and measuring the service performance of spindles. The traditional static stiffness indexes only involve static analysis and rarely focus on the study of spindle-carrying capacity under operating conditions. In this paper, the explicit solution approach is used to develop a mechanical model of the spindle’s axial operating stiffness. This model was then used to explore the influence of rotational speed on the softening and hardening features of the spindle axial operating stiffness, and experimental verification was carried out. According to studies, the speed of a fixed-position preload spindle can lead its operating stiffness to exhibit a “stiffness-hardening” feature. However, when the axial displacement of the spindle is small, the operating stiffness curve of the spindle displays a noticeable “fluctuation” phenomenon for low-speed spindles. Furthermore, the speed-induced preload has a significant impact on the test results when testing spindle axial operating stiffness.
- 22 June 2022
- Published by MDPI AG in Lubricants
- Vol. 10 (7) , 132
Open AccessResearch on the Effect of Spindle Speed on the Softening and Hardening Characteristics of the Axial Operating Stiffness of Machine Tool Spindle
Spindle stiffness is one of the most critical indicators for evaluating and measuring the service performance of spindles. The traditional static stiffness indexes only involve static analysis and rarely focus on the study of spindle-carrying capacity under operating conditions. In this paper, the explicit solution approach is used to develop a mechanical model of the spindle’s axial operating stiffness. This model was then used to explore the influence of rotational speed on the softening and hardening features of the spindle axial operating stiffness, and experimental verification was carried out. According to studies, the speed of a fixed-position preload spindle can lead its operating stiffness to exhibit a “stiffness-hardening” feature. However, when the axial displacement of the spindle is small, the operating stiffness curve of the spindle displays a noticeable “fluctuation” phenomenon for low-speed spindles. Furthermore, the speed-induced preload has a significant impact on the test results when testing spindle axial operating stiffness.
- 1 April 2022
- Published by Elsevier in ISA Transactions
- Vol. 123, 381-397
Research on extracting weak repetitive transients of fault rolling element bearing
- 1 April 2022
- Published by Elsevier in ISA Transactions
- Vol. 123, 381-397
Research on extracting weak repetitive transients of fault rolling element bearing