Kewen Zhang
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
Kewen Zhang has a total of 36 co-authors. Here are the top co-authors by number of shared publications:
- Mingyu Du
The College of Mechanical Engineering, Zhejiang University of Technology
5 shared publications - Guanjun Bao
College of Mechanical Engineering, Zhejiang University of Technology, Hangzhou 310023, China
4 shared publications - Shibo Cai
School of Mechanical and Electrical Engineering, China Jiliang University, Hangzhou, China
4 shared publications - Xin Chen
College of Mechanical Engineering, Zhejiang University of Technology, Hangzhou, China
3 shared publications - Bingfei Fan
College of Mechanical Engineering, Zhejiang University of Technology, Hangzhou, China
2 shared publications - Jin Wang
Zhejiang University
2 shared publications
Latest Publications
- 7 May 2026
- Published by Elsevier in Mechatronics
- Vol. 118
Design, modeling, and control of a multi-link horizontal lower-limb rehabilitation robot
- 7 May 2026
- Published by Elsevier in Mechatronics
- Vol. 118
Design, modeling, and control of a multi-link horizontal lower-limb rehabilitation robot
- 24 October 2025
- Published by Springer Nature
- p. 310-322
Development of a Functional Electrical Stimulation Device Combined with Multi-modal Muscle Status Monitoring
- 24 October 2025
- Published by Springer Nature
- p. 310-322
Development of a Functional Electrical Stimulation Device Combined with Multi-modal Muscle Status Monitoring
- 2 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Human-Machine Systems
- Vol. 55 (4) , 650-659
A Method for Improving Online Active Engagement During Lower Limb Rehabilitation Training Based on EEG Signals
Increasing online active engagement has been shown to effectively improve rehabilitation outcomes. This article introduces a method for improving online active engagement during lower limb rehabilitation training based on electroencephalogram (EEG) signals using an artificial neural network (ANN) recognition model as well as a virtual reality-based induced active engagement rehabilitation training system (VATS). The VATS consists of a bicycle-style rehabilitation instrument and a virtual reality game. Experiments were performed to train the ANN model and to validate the effectiveness of the VATS. We compared the accuracy of the ANN with that of k-nearest neighbor and support vector machine models, where active engagement results from eye tracker analysis were used as the gold standard. In addition, we used the surface electromyography engagement index (Er) to compare the VATS with and without active engagement feedback. The results showed that the ANN model had the highest average accuracy of 76.95% and a maximum accuracy of 81.48%. A comparison of the ANN, KNN, and SVM models revealed statistically significant differences in accuracy (p < 0.05). Significant differences were observed in the Er values between the VATS with and without active engagement feedback. The results indicated that the ANN has an apparent advantage in active engagement recognition. Moreover, the VATS was effective in improving active engagement. The proposed method utilized EEG signals to characterize active engagement intention and achieved real-time enhancement of active engagement through VATS. This method could be further used to improve the rehabilitation training effect in clinical applications.
- 2 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Human-Machine Systems
- Vol. 55 (4) , 650-659
A Method for Improving Online Active Engagement During Lower Limb Rehabilitation Training Based on EEG Signals
Increasing online active engagement has been shown to effectively improve rehabilitation outcomes. This article introduces a method for improving online active engagement during lower limb rehabilitation training based on electroencephalogram (EEG) signals using an artificial neural network (ANN) recognition model as well as a virtual reality-based induced active engagement rehabilitation training system (VATS). The VATS consists of a bicycle-style rehabilitation instrument and a virtual reality game. Experiments were performed to train the ANN model and to validate the effectiveness of the VATS. We compared the accuracy of the ANN with that of k-nearest neighbor and support vector machine models, where active engagement results from eye tracker analysis were used as the gold standard. In addition, we used the surface electromyography engagement index (Er) to compare the VATS with and without active engagement feedback. The results showed that the ANN model had the highest average accuracy of 76.95% and a maximum accuracy of 81.48%. A comparison of the ANN, KNN, and SVM models revealed statistically significant differences in accuracy (p < 0.05). Significant differences were observed in the Er values between the VATS with and without active engagement feedback. The results indicated that the ANN has an apparent advantage in active engagement recognition. Moreover, the VATS was effective in improving active engagement. The proposed method utilized EEG signals to characterize active engagement intention and achieved real-time enhancement of active engagement through VATS. This method could be further used to improve the rehabilitation training effect in clinical applications.
- 2 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (10) , 16775-16786
Complete Gait Phase Recognition Based on Muscle Synergy Using PSO-CNN-LSTM Algorithm
Accurately recognizing gait phases, by applying proper instrumentation and measurement, is significant in walking rehabilitation training for patients with impaired mobility. In this study, seven phases of complete stand-walk-stand cycle as well as continuous daily walking were recognized based on muscle synergy and PSO-CNN-LSTM model. Firstly, the complete stand-walk-stand walking cycle were divided into starting phase, terminal swing phase, loading response phase, mid-stance phase, terminal stance phase, initial swing phase, and stopping phase, based on the features of surface electromyography (sEMG) collected by portable sEMG acquisition system and measured motion data of lower limb. Secondly, a muscle weight matrix and an activation sequence matrix were calculated by using non-negative matrix factorization (NMF). Finally, a PSO-CNN-LSTM network was designed to recognize the complete stand-walk-stand cycle by applying above mentioned matrices as input. Fourteen subjects volunteered to perform linear walking experiments under procedure to verify the feasibility of the proposed approach by comparing results with other classifiers and features. Experimental results show that the proposed approach was capable of achieving an average recognition accuracy of up to 85.384%. This work will offer promising gait recognition method for rehabilitation robots to achieve natural and flexible human-robot interaction.
- 2 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (10) , 16775-16786
Complete Gait Phase Recognition Based on Muscle Synergy Using PSO-CNN-LSTM Algorithm
Accurately recognizing gait phases, by applying proper instrumentation and measurement, is significant in walking rehabilitation training for patients with impaired mobility. In this study, seven phases of complete stand-walk-stand cycle as well as continuous daily walking were recognized based on muscle synergy and PSO-CNN-LSTM model. Firstly, the complete stand-walk-stand walking cycle were divided into starting phase, terminal swing phase, loading response phase, mid-stance phase, terminal stance phase, initial swing phase, and stopping phase, based on the features of surface electromyography (sEMG) collected by portable sEMG acquisition system and measured motion data of lower limb. Secondly, a muscle weight matrix and an activation sequence matrix were calculated by using non-negative matrix factorization (NMF). Finally, a PSO-CNN-LSTM network was designed to recognize the complete stand-walk-stand cycle by applying above mentioned matrices as input. Fourteen subjects volunteered to perform linear walking experiments under procedure to verify the feasibility of the proposed approach by comparing results with other classifiers and features. Experimental results show that the proposed approach was capable of achieving an average recognition accuracy of up to 85.384%. This work will offer promising gait recognition method for rehabilitation robots to achieve natural and flexible human-robot interaction.
- 7 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (5) , 8515-8527
Predicting Paretic Gait Trajectories Using sEMG With the ISSA-CNN-SVR Model
The continuous prediction of the desired gait trajectory of the paretic leg is a key task for the lower limb rehabilitation robot to assist the coordinated motion of the healthy and paretic legs of stroke patients, which has important research significance. In this paper, we propose a continuous prediction method for the desired gait trajectory of the paretic leg based on the sEMG signal of the healthy leg. In this method, an ISSA-CNN-SVR cascade network prediction model is designed. Based on this model, the mapping relationship between the sEMG signal of the healthy leg and the hip and knee joint angle of the paretic leg is established. The model input is the 8-channel sEMG signal of the healthy leg in the current instantaneous moment, and the output is the desired gait trajectory (hip and knee joint angle trajectories) of the paretic leg in the instantaneous moment after 100 ms. We recruited eight healthy participants for the validation experiments. Results indicate that the proposed method can continuously and smoothly predict the hip and knee joint angle trajectories of the paretic leg for the instantaneous moment after 100 ms based on the sEMG signal of the healthy leg and the ISSA-CNN-SVR model. The average RMSE of the predicted hip and knee joint angle trajectories of the paretic leg are 4.643° and 6.845°, with average R 2 being 0.855 and 0.857, respectively, demonstrating a high level of fitting. Therefore, this study is of great significance to improve the gait coordination of the healthy and paretic legs of robot-assisted stroke patients in the process of rehabilitation training.
- 7 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (5) , 8515-8527
Predicting Paretic Gait Trajectories Using sEMG With the ISSA-CNN-SVR Model
The continuous prediction of the desired gait trajectory of the paretic leg is a key task for the lower limb rehabilitation robot to assist the coordinated motion of the healthy and paretic legs of stroke patients, which has important research significance. In this paper, we propose a continuous prediction method for the desired gait trajectory of the paretic leg based on the sEMG signal of the healthy leg. In this method, an ISSA-CNN-SVR cascade network prediction model is designed. Based on this model, the mapping relationship between the sEMG signal of the healthy leg and the hip and knee joint angle of the paretic leg is established. The model input is the 8-channel sEMG signal of the healthy leg in the current instantaneous moment, and the output is the desired gait trajectory (hip and knee joint angle trajectories) of the paretic leg in the instantaneous moment after 100 ms. We recruited eight healthy participants for the validation experiments. Results indicate that the proposed method can continuously and smoothly predict the hip and knee joint angle trajectories of the paretic leg for the instantaneous moment after 100 ms based on the sEMG signal of the healthy leg and the ISSA-CNN-SVR model. The average RMSE of the predicted hip and knee joint angle trajectories of the paretic leg are 4.643° and 6.845°, with average R 2 being 0.855 and 0.857, respectively, demonstrating a high level of fitting. Therefore, this study is of great significance to improve the gait coordination of the healthy and paretic legs of robot-assisted stroke patients in the process of rehabilitation training.
- 6 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 71 (8) , 9351-9363
Adaptive Composite Observer-Based Global Finite Time Control With Prescribed Performance for Robots
As humans focus control only on task–space variables to achieve dexterous manipulation, robots could strongly profit from advanced task-space control, which is still blocked by kinematic and dynamic uncertainties now. This article proposes an adaptive composite observer (ACO) with finite-time sliding-mode manifold to compensate uncertain kinematics and dynamics synthetically, then develops a novel nonsingular terminal sliding mode controller based on an adaptive neural network (NN) to stabilize the task–space tracking errors directly with prescribed performance. The global finite-time stability of the entire observer–controller system is achieved by Lyapunov method, and the observer and controller errors will converge to zeros in finite time with preassigned transient and steady-state performance. Simulation and experimental studies are also presented to verify the effectiveness of the designed control methods.
- 6 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 71 (8) , 9351-9363
Adaptive Composite Observer-Based Global Finite Time Control With Prescribed Performance for Robots
As humans focus control only on task–space variables to achieve dexterous manipulation, robots could strongly profit from advanced task-space control, which is still blocked by kinematic and dynamic uncertainties now. This article proposes an adaptive composite observer (ACO) with finite-time sliding-mode manifold to compensate uncertain kinematics and dynamics synthetically, then develops a novel nonsingular terminal sliding mode controller based on an adaptive neural network (NN) to stabilize the task–space tracking errors directly with prescribed performance. The global finite-time stability of the entire observer–controller system is achieved by Lyapunov method, and the observer and controller errors will converge to zeros in finite time with preassigned transient and steady-state performance. Simulation and experimental studies are also presented to verify the effectiveness of the designed control methods.
- 7 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 8 (9) , 5307-5314
Bat Planner: Aggressive Flying Ball Player
In this letter, an aggressive quadrotor ball playing system called Bat is proposed, whose goal is to intercept a flying ball and volley it towards a designated target. Aggressive means Bat operates the quadrotor aggressively to intercept balls that are far away and hit them to distant positions in ways that are beyond the reach of existing methods. The trajectory prediction of the ball is achieved by integrating forward the current position and velocity estimates using an extended kalman filter, and implementing cubic interpolation at the time resolution to calculate the continuous gradient for optimization. Facing the challenge of finding feasible hitting actions under extreme circumstances, we propose a two-stage planning approach, including transition point design and hitting primitive generation, with a simplified expression of uncoupled hitting actions. To obtain the best hitting motion, a trajectory optimization method is proposed, which can jointly optimize the hitting terminal states and time cost, considering dynamic feasibility and anticollision constraints. To avoid pathological hitting, a defensive rule constraint and its constraint transcription method are proposed. The largest difference from the existing methods is that Bat Planner can independently decide how to execute more aggressive keyvolleying maneuvers. A large number of simulation and real-world experiments are conducted, which prove the flying ball player can hit arriving balls from different directions and distances to arbitrary targets. To the best of our knowledge, Bat is currently the closest a quadrotor ball player has approached to human ball players' volleying ability.
- 7 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 8 (9) , 5307-5314
Bat Planner: Aggressive Flying Ball Player
In this letter, an aggressive quadrotor ball playing system called Bat is proposed, whose goal is to intercept a flying ball and volley it towards a designated target. Aggressive means Bat operates the quadrotor aggressively to intercept balls that are far away and hit them to distant positions in ways that are beyond the reach of existing methods. The trajectory prediction of the ball is achieved by integrating forward the current position and velocity estimates using an extended kalman filter, and implementing cubic interpolation at the time resolution to calculate the continuous gradient for optimization. Facing the challenge of finding feasible hitting actions under extreme circumstances, we propose a two-stage planning approach, including transition point design and hitting primitive generation, with a simplified expression of uncoupled hitting actions. To obtain the best hitting motion, a trajectory optimization method is proposed, which can jointly optimize the hitting terminal states and time cost, considering dynamic feasibility and anticollision constraints. To avoid pathological hitting, a defensive rule constraint and its constraint transcription method are proposed. The largest difference from the existing methods is that Bat Planner can independently decide how to execute more aggressive keyvolleying maneuvers. A large number of simulation and real-world experiments are conducted, which prove the flying ball player can hit arriving balls from different directions and distances to arbitrary targets. To the best of our knowledge, Bat is currently the closest a quadrotor ball player has approached to human ball players' volleying ability.
- 1 August 2019
- Published by Elsevier in Molecular and Cellular Endocrinology
- Vol. 494, 110494
Microtubule actin crosslinking factor 1 (MACF1) knockdown inhibits RANKL-induced osteoclastogenesis via Akt/GSK3β/NFATc1 signalling pathway
- 1 August 2019
- Published by Elsevier in Molecular and Cellular Endocrinology
- Vol. 494, 110494
Microtubule actin crosslinking factor 1 (MACF1) knockdown inhibits RANKL-induced osteoclastogenesis via Akt/GSK3β/NFATc1 signalling pathway