Xiangkai Shen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Xiangkai Shen has a total of 60 co-authors. Here are the top co-authors by number of shared publications:
- Jinhai Liu
College of Information Science and Engineering, Northeastern University, Shenyang 110819, China
24 shared publications - Huaguang Zhang
Northeastern University
15 shared publications - Lin Jiang
College of Information Science and Engineering, Northeastern University, Shenyang, China
11 shared publications - He Zhao
College of Information Science and Engineering, Northeastern University, Shenyang, China
7 shared publications - Yan Shi
Beihang University
6 shared publications - Fengyuan Zuo
School of Information Science and Engineering, Northeastern University, Shenyang, China
5 shared publications
Latest Publications
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (18) , 43565-43577
A Physical Information-Guided Masked Contrastive Learning Network for Defect Detection in Intelligent Pipeline Systems With MFL Signals
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (18) , 43565-43577
A Physical Information-Guided Masked Contrastive Learning Network for Defect Detection in Intelligent Pipeline Systems With MFL Signals
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/CAA Journal of Automatica Sinica
- Vol. 13 (5) , 1254-1256
Adaptive Fully-Actuated Control for Lower Limb Exoskeletons with High-Order Disturbance Compensation
Dear Editor, This letter addresses disturbance-resilient, constraint-satisfying control for lower-limb rehabilitation exoskeletons operating under irregular patient-induced perturbations. We combine a fully actuated system (FAS) outer loop with a high-order disturbance observer (HODO) inner loop and a constrained gradient-based self-tuning scheme. The framework was validated in simulation and on a prototype device. In prototype tests, the hip and knee achieved peak tracking errors of 0.069 rad and 0.038 rad, with no violations of physical limits.
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/CAA Journal of Automatica Sinica
- Vol. 13 (5) , 1254-1256
Adaptive Fully-Actuated Control for Lower Limb Exoskeletons with High-Order Disturbance Compensation
Dear Editor, This letter addresses disturbance-resilient, constraint-satisfying control for lower-limb rehabilitation exoskeletons operating under irregular patient-induced perturbations. We combine a fully actuated system (FAS) outer loop with a high-order disturbance observer (HODO) inner loop and a constrained gradient-based self-tuning scheme. The framework was validated in simulation and on a prototype device. In prototype tests, the hip and knee achieved peak tracking errors of 0.069 rad and 0.038 rad, with no violations of physical limits.
- Top Cited Papers
- 19 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23 (15455955) , 9771-9782
Distributed Finite-Time Fuzzy Adaptive Consensus Control for Robot Manipulators With Input Deadzone and Model Uncertainties
The multiple manipulator system (MMS) has strong coupling properties and nonlinearities, which is used to accomplish complex cooperation tasks. In this study, a distributed consensus control algorithm is proposed for uncertain MMS with input deadzone under a directed communication graph. Meanwhile, a fuzzy logic system (FLS) is designed to approximate uncertain dynamics for controller compensation. A fast finite-time convergence backstepping controller is designed to ensure that the state error of the MMS system converges to the zero neighborhood within a finite time. In addition, an adaptive method is used to estimate and compensate for the unknown input dead zone parameters. Finally, the effectiveness of the control method is verified through the simulation model and experimental platform, and its advantages are verified through comparative experiments.
- 19 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23 (15455955) , 9771-9782
Top Cited PapersDistributed Finite-Time Fuzzy Adaptive Consensus Control for Robot Manipulators With Input Deadzone and Model Uncertainties
The multiple manipulator system (MMS) has strong coupling properties and nonlinearities, which is used to accomplish complex cooperation tasks. In this study, a distributed consensus control algorithm is proposed for uncertain MMS with input deadzone under a directed communication graph. Meanwhile, a fuzzy logic system (FLS) is designed to approximate uncertain dynamics for controller compensation. A fast finite-time convergence backstepping controller is designed to ensure that the state error of the MMS system converges to the zero neighborhood within a finite time. In addition, an adaptive method is used to estimate and compensate for the unknown input dead zone parameters. Finally, the effectiveness of the control method is verified through the simulation model and experimental platform, and its advantages are verified through comparative experiments.
- 13 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23 (15455955) , 9797-9808
A Domain-Incremental Learning Framework Based on Defect Replay and Gaussian Mixup for Industrial Defect Detection
Incremental learning is a critical yet challenging problem in automation engineering, especially across heterogeneous domains. Existing incremental learning methods utilize mixup and mosaic techniques to replay previous knowledge. However, the saliency discrepancy between replayed samples is often ignored, leading to suboptimal accuracy. To address this problem, we propose a novel replay-based incremental learning framework that replays previous defect samples based on mixup and mosaic. To dynamically assign samples to different image augmentation techniques, a saliency score calculation mechanism is proposed by using posterior probability and spatial location. To eliminate the step noise brought by mixup, we propose a Gaussian operator-improved mixup to smoothly merge previous foregrounds with current images. Additionally, a domain-attentive distillation loss is proposed to further alleviate forgetting. Experiments on three scenarios verify the improved accuracy of the proposed method, achieving improvements of 2.4% mAP on weld defect detection, 2.7% mAP on steel defect detection and 4.0% mAP on no-service rail defect detection, respectively.
- 13 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23 (15455955) , 9797-9808
A Domain-Incremental Learning Framework Based on Defect Replay and Gaussian Mixup for Industrial Defect Detection
Incremental learning is a critical yet challenging problem in automation engineering, especially across heterogeneous domains. Existing incremental learning methods utilize mixup and mosaic techniques to replay previous knowledge. However, the saliency discrepancy between replayed samples is often ignored, leading to suboptimal accuracy. To address this problem, we propose a novel replay-based incremental learning framework that replays previous defect samples based on mixup and mosaic. To dynamically assign samples to different image augmentation techniques, a saliency score calculation mechanism is proposed by using posterior probability and spatial location. To eliminate the step noise brought by mixup, we propose a Gaussian operator-improved mixup to smoothly merge previous foregrounds with current images. Additionally, a domain-attentive distillation loss is proposed to further alleviate forgetting. Experiments on three scenarios verify the improved accuracy of the proposed method, achieving improvements of 2.4% mAP on weld defect detection, 2.7% mAP on steel defect detection and 4.0% mAP on no-service rail defect detection, respectively.
- 11 March 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 318
A hierarchical knowledge-embedded fine-grained network for pipeline weak defect detection
- 11 March 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 318
A hierarchical knowledge-embedded fine-grained network for pipeline weak defect detection
- 16 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (7) , 9603-9615
Continuous Learning Method for Incremental Fault Class Identification of Control Valves Based on Weighted Supervised Contrastive Knowledge Distillation
Data-driven methods have achieved high accuracy in diagnosing control valve faults. However, Incremental Learning (IL) methods for step signal fault diagnosis of control valves remain insufficient. First, the distinct characteristics of various step-response signals lead to data imbalance, which makes it difficult for models to extract discriminative features from step-fault data. Second, the arrival of a novel fault class necessitates full retraining and incurs prohibitive computational overhead. Finally, as the number of fault classes increases, the model tends to bias toward newly added fault classes and may suffer from catastrophic forgetting of previously learned ones. These issues collectively degrade the performance of incremental fault diagnosis models. To solve these problems, this paper proposes an incremental fault diagnosis model based on Weighted Supervised Contrastive Knowledge Distillation (WSCKD), which extends the classical Incremental Classifier And Representation Learning (ICARL) framework in three key ways. First, supervised contrastive learning (SCL) is weighted to enhance representation learning for step-response signals. In addition, a new preferred example selection method, Greedy Diversity Sampling (GDS), is proposed to limit computational overhead and is combined with knowledge distillation to mitigate catastrophic forgetting. Furthermore, a Balanced Bagging Classifier (BBC) alleviates the adverse effects of class imbalance. Extensive simulation experiments were conducted on a control valve system. Experimental results demonstrate that the WSCKD method outperforms existing approaches, achieving average incremental accuracies of 94.32% for 5 fault classes.
- 16 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (7) , 9603-9615
Continuous Learning Method for Incremental Fault Class Identification of Control Valves Based on Weighted Supervised Contrastive Knowledge Distillation
Data-driven methods have achieved high accuracy in diagnosing control valve faults. However, Incremental Learning (IL) methods for step signal fault diagnosis of control valves remain insufficient. First, the distinct characteristics of various step-response signals lead to data imbalance, which makes it difficult for models to extract discriminative features from step-fault data. Second, the arrival of a novel fault class necessitates full retraining and incurs prohibitive computational overhead. Finally, as the number of fault classes increases, the model tends to bias toward newly added fault classes and may suffer from catastrophic forgetting of previously learned ones. These issues collectively degrade the performance of incremental fault diagnosis models. To solve these problems, this paper proposes an incremental fault diagnosis model based on Weighted Supervised Contrastive Knowledge Distillation (WSCKD), which extends the classical Incremental Classifier And Representation Learning (ICARL) framework in three key ways. First, supervised contrastive learning (SCL) is weighted to enhance representation learning for step-response signals. In addition, a new preferred example selection method, Greedy Diversity Sampling (GDS), is proposed to limit computational overhead and is combined with knowledge distillation to mitigate catastrophic forgetting. Furthermore, a Balanced Bagging Classifier (BBC) alleviates the adverse effects of class imbalance. Extensive simulation experiments were conducted on a control valve system. Experimental results demonstrate that the WSCKD method outperforms existing approaches, achieving average incremental accuracies of 94.32% for 5 fault classes.
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Reliability
- Vol. 75 (00189529) , 133-144
Joint Domain Adaptation via Cluster Centers and Thermal Imaging for Detecting Leakages in Different Pneumatic Components
Pneumatic components are essential for precise control and automation in mechanical manufacturing. Leakage faults in pneumatic components can seriously undermine the reliability of manufacturing processes. When high-pressure gas leaks and expands, heat transfer between the gas and the component wall creates a localized low-temperature area. This thermal anomaly can be detected using thermal imaging techniques. However, thermal images captured from different pneumatic components exhibit distinct distribution patterns, which can significantly degrade the accuracy of existing detection methods. Conventional domain adaptation methods typically use the sample mean to represent feature distributions, neglecting intraclass dispersion and the interdependence among feature learning, classifier learning, and pseudolabel learning. To address these limitations, we propose a joint unsupervised domain adaptation method via cluster centers and thermal imaging for leakage detection in different pneumatic components (JCC-LDC). Specifically, JCC-LDC accounts for the fact that samples of a single class may disperse into multiple clusters, and it integrates feature learning, classifier learning, and pseudolabel learning into a unified framework based on cluster centers. Experimental results demonstrate that JCC-LDC increases the average detection accuracy from 71.5% to 82.3%, effectively enhancing the generalization capability of thermal-imaging-based leakage detection in pneumatic components.
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Reliability
- Vol. 75 (00189529) , 133-144
Joint Domain Adaptation via Cluster Centers and Thermal Imaging for Detecting Leakages in Different Pneumatic Components
Pneumatic components are essential for precise control and automation in mechanical manufacturing. Leakage faults in pneumatic components can seriously undermine the reliability of manufacturing processes. When high-pressure gas leaks and expands, heat transfer between the gas and the component wall creates a localized low-temperature area. This thermal anomaly can be detected using thermal imaging techniques. However, thermal images captured from different pneumatic components exhibit distinct distribution patterns, which can significantly degrade the accuracy of existing detection methods. Conventional domain adaptation methods typically use the sample mean to represent feature distributions, neglecting intraclass dispersion and the interdependence among feature learning, classifier learning, and pseudolabel learning. To address these limitations, we propose a joint unsupervised domain adaptation method via cluster centers and thermal imaging for leakage detection in different pneumatic components (JCC-LDC). Specifically, JCC-LDC accounts for the fact that samples of a single class may disperse into multiple clusters, and it integrates feature learning, classifier learning, and pseudolabel learning into a unified framework based on cluster centers. Experimental results demonstrate that JCC-LDC increases the average detection accuracy from 71.5% to 82.3%, effectively enhancing the generalization capability of thermal-imaging-based leakage detection in pneumatic components.
- 27 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 37 (5) , 2321-2334
Leader-Based Multiexpert Neural Network for High-Level Visual Tasks
Remarkable progress has been achieved in the detection and segmentation of the baseline; however, for high-level visual tasks in complex scenes (e.g., dense, occlusion, scale diversity, high background noise, etc.), existing frameworks often fail to provide satisfactory performance. To further improve the object recognition ability, this article introduces a leader-based multiexpert mechanism into the detection and segmentation tasks. In this work, we first design a leader-based attention learning layer to fully integrate multilevel features from the backbone network, which can effectively obtain global semantics and assign instructions to detection experts. Then, we propose multiple feature pyramids with dual fusion paths to replace the traditional single pipeline using semantic and spatial allocators. With this strategy, we can further establish deep supervision for multiple experts during training and sufficiently utilize the multiexpert detection results from leaders’ assignments during reasoning, thereby comprehensively improving the performance of the model in complex scenarios. In the experiment, we established ablation studies and performance comparisons on COCO 2017 detection and segmentation tasks. Finally, we demonstrated the model’s performance in three complex application scenarios (remote sensing, autonomous driving, and industrial fields), and the results showed our advantages.
- 27 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 37 (5) , 2321-2334
Leader-Based Multiexpert Neural Network for High-Level Visual Tasks
Remarkable progress has been achieved in the detection and segmentation of the baseline; however, for high-level visual tasks in complex scenes (e.g., dense, occlusion, scale diversity, high background noise, etc.), existing frameworks often fail to provide satisfactory performance. To further improve the object recognition ability, this article introduces a leader-based multiexpert mechanism into the detection and segmentation tasks. In this work, we first design a leader-based attention learning layer to fully integrate multilevel features from the backbone network, which can effectively obtain global semantics and assign instructions to detection experts. Then, we propose multiple feature pyramids with dual fusion paths to replace the traditional single pipeline using semantic and spatial allocators. With this strategy, we can further establish deep supervision for multiple experts during training and sufficiently utilize the multiexpert detection results from leaders’ assignments during reasoning, thereby comprehensively improving the performance of the model in complex scenarios. In the experiment, we established ablation studies and performance comparisons on COCO 2017 detection and segmentation tasks. Finally, we demonstrated the model’s performance in three complex application scenarios (remote sensing, autonomous driving, and industrial fields), and the results showed our advantages.
- Top Cited Papers
- 31 July 2025
- Published by Elsevier in Mechanical Systems and Signal Processing
- Vol. 237
A multi-expert diffusion model for surface defect detection of valve cores in special control valve equipment systems
- 31 July 2025
- Published by Elsevier in Mechanical Systems and Signal Processing
- Vol. 237
Top Cited PapersA multi-expert diffusion model for surface defect detection of valve cores in special control valve equipment systems
- Top Cited Papers
- 26 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-13
VLCIM: A Vision-Language Cyclic Interaction Model for Industrial Defect Detection
Accurate defect detection is an important element in ensuring product quality and safe equipment operation. However, due to the lack of deep cross-modal interactions during vision feature extraction, existing methods often suffer from attention bias, which ultimately limits detection accuracy. To address this issue, this paper proposes a Vision-Language Cyclic Interaction Model (VLCIM), which progressively optimizes vision feature extraction by integrating domain prior knowledge and generic large model, effectively bridging the dual-domain barrier between “generic-specific” and “vision-language”. Specifically, progressive cyclic interaction learning is proposed for the first time, which integrates a recursive guidance module (RGM) and cross-modal interaction (CMI) strategy to realize bidirectional dynamic fusion and collaborative optimization of vision and language features. Furthermore, the proposed dual-view synergistic detection mechanism enhances discriminative decision responses, significantly improving the model’s boundary perception ability and decision-making accuracy in complex scenarios. VLCIM achieves high-precision defect detection by establishing a cyclic interaction mechanism between domain-specific language features and vision representations. Experimental results on three industrial datasets demonstrate that VLCIM achieves improvements of 5.9%, 5.6%, and 4.1% in mIoU over the state-of-the-art (SOTA) methods, indicating its validity and generalization in different scenarios.
- 26 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-13
Top Cited PapersVLCIM: A Vision-Language Cyclic Interaction Model for Industrial Defect Detection
Accurate defect detection is an important element in ensuring product quality and safe equipment operation. However, due to the lack of deep cross-modal interactions during vision feature extraction, existing methods often suffer from attention bias, which ultimately limits detection accuracy. To address this issue, this paper proposes a Vision-Language Cyclic Interaction Model (VLCIM), which progressively optimizes vision feature extraction by integrating domain prior knowledge and generic large model, effectively bridging the dual-domain barrier between “generic-specific” and “vision-language”. Specifically, progressive cyclic interaction learning is proposed for the first time, which integrates a recursive guidance module (RGM) and cross-modal interaction (CMI) strategy to realize bidirectional dynamic fusion and collaborative optimization of vision and language features. Furthermore, the proposed dual-view synergistic detection mechanism enhances discriminative decision responses, significantly improving the model’s boundary perception ability and decision-making accuracy in complex scenarios. VLCIM achieves high-precision defect detection by establishing a cyclic interaction mechanism between domain-specific language features and vision representations. Experimental results on three industrial datasets demonstrate that VLCIM achieves improvements of 5.9%, 5.6%, and 4.1% in mIoU over the state-of-the-art (SOTA) methods, indicating its validity and generalization in different scenarios.