Zhilei Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zhilei Liu has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Dayong Ning
Naval Architecture and Ocean Engineering College, Dalian Maritime University, Dalian 116026, China
3 shared publications Cheng Zhou
Wuhan 2nd Ship Design & Res Inst, Wuhan 430000, Peoples R China
2 shared publicationsDayong Ning
Naval Architecture and Ocean Engineering College, Dalian Maritime University,Dalian,China
2 shared publicationsFengrui Zhang
Dalian Maritime Univ, Naval Architecture & Ocean Engn Coll, Dalian 116026, Peoples R China
2 shared publicationsGangda Liang
Dalian Maritime Univ, Naval Architecture & Ocean Engn Coll, Dalian 116026, Peoples R China
2 shared publications- Hui Zhang
School of Mechanical Engineering and Automation, Dalian Polytechnic University, Dalian 116034, China
2 shared publications
Latest Publications
- Open Access
RG-RGD: Task-Triggered Small-Target RGB-D Depth Refinement for Robotic Laser Ablation
Robotic 3D manipulation of small targets, such as laser ablation of urban-tree fruit balls, requires locally reliable depth. Mainstream RGB-D depth-refinement methods usually optimize image-wide metrics, which can leave task-relevant regions under-resolved and limit their direct use in precision robotic operation. This paper presents residual-gated RGB-D depth refinement (RG-RGD), a task-triggered local depth-refinement framework that reallocates computation toward task-relevant 3D geometry after a candidate target has been selected. The method contains three coupled designs: a self-play benefit-driven foveation mechanism that focuses network capacity on regions where refinement reduces geometric residuals; residual prediction with Bayesian measurement fusion that anchors predictions to available raw observations; and inertial measurement unit (IMU)-conditioned self-supervised training that improves inter-frame view consistency. Evaluated on the Visual Odometry with Inertial and Depth (VOID) benchmark, RG-RGD obtains competitive standard depth-completion metrics, including a mean absolute error (MAE) of 24.95 mm and an inverse mean absolute error (iMAE) of 10.85. On self-collected London plane fruit-ball sequences, the method reduces region-of-interest geometric error more strongly than full-image error. A qualitative demonstrative laser-ablation use case illustrates how refined local geometry can drive physical branch filtering, cutting-point selection, and gimbal-based execution. The results support task-triggered local depth refinement as a practical perception component for robotic manipulation.
Open AccessRG-RGD: Task-Triggered Small-Target RGB-D Depth Refinement for Robotic Laser Ablation
Robotic 3D manipulation of small targets, such as laser ablation of urban-tree fruit balls, requires locally reliable depth. Mainstream RGB-D depth-refinement methods usually optimize image-wide metrics, which can leave task-relevant regions under-resolved and limit their direct use in precision robotic operation. This paper presents residual-gated RGB-D depth refinement (RG-RGD), a task-triggered local depth-refinement framework that reallocates computation toward task-relevant 3D geometry after a candidate target has been selected. The method contains three coupled designs: a self-play benefit-driven foveation mechanism that focuses network capacity on regions where refinement reduces geometric residuals; residual prediction with Bayesian measurement fusion that anchors predictions to available raw observations; and inertial measurement unit (IMU)-conditioned self-supervised training that improves inter-frame view consistency. Evaluated on the Visual Odometry with Inertial and Depth (VOID) benchmark, RG-RGD obtains competitive standard depth-completion metrics, including a mean absolute error (MAE) of 24.95 mm and an inverse mean absolute error (iMAE) of 10.85. On self-collected London plane fruit-ball sequences, the method reduces region-of-interest geometric error more strongly than full-image error. A qualitative demonstrative laser-ablation use case illustrates how refined local geometry can drive physical branch filtering, cutting-point selection, and gimbal-based execution. The results support task-triggered local depth refinement as a practical perception component for robotic manipulation.
- 1 July 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Oceanic Engineering
- Vol. 51 (3) , 2192-2205
Design and Fabrication of a Compact Waterjet Thruster for Deep-Sea Applications
In the context of the growing demand for compact equipment in deep-sea exploration missions, the development of lightweight propulsion systems is confronted with considerable challenges. In view of the necessity to resist high pressure, the design of deep-sea thrusters generally incorporates a bulky configuration. To address this technical bottleneck, this study proposes a novel compact deep-sea thruster based on electromagnetic soft-coupled actuation. The thruster comprises a body support structure, an electromagnetic drive mechanism, and a soft-bodied water bladder. The mechanism in question generates periodic propulsion forces by altering the volume of the bladder in response to pulsed electrical signals. This study investigates the propulsion force and operational cycles of the thruster under varying structural parameters and control currents. A pressure tank is employed to validate the thruster’s pressure resistance, and two typical operational configurations are designed to test its propulsion effectiveness. The findings indicate that the thruster functions reliably at depths in excess of 11 000 m. It has been demonstrated that the two typical configurations are capable of attaining swimming speeds of 92.7 mm/s (1.55 BL/s) and 68.2 mm/s (1.17 BL/s), respectively, thereby demonstrating exceptional propulsion capabilities.
- 1 July 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Oceanic Engineering
- Vol. 51 (3) , 2192-2205
Design and Fabrication of a Compact Waterjet Thruster for Deep-Sea Applications
In the context of the growing demand for compact equipment in deep-sea exploration missions, the development of lightweight propulsion systems is confronted with considerable challenges. In view of the necessity to resist high pressure, the design of deep-sea thrusters generally incorporates a bulky configuration. To address this technical bottleneck, this study proposes a novel compact deep-sea thruster based on electromagnetic soft-coupled actuation. The thruster comprises a body support structure, an electromagnetic drive mechanism, and a soft-bodied water bladder. The mechanism in question generates periodic propulsion forces by altering the volume of the bladder in response to pulsed electrical signals. This study investigates the propulsion force and operational cycles of the thruster under varying structural parameters and control currents. A pressure tank is employed to validate the thruster’s pressure resistance, and two typical operational configurations are designed to test its propulsion effectiveness. The findings indicate that the thruster functions reliably at depths in excess of 11 000 m. It has been demonstrated that the two typical configurations are capable of attaining swimming speeds of 92.7 mm/s (1.55 BL/s) and 68.2 mm/s (1.17 BL/s), respectively, thereby demonstrating exceptional propulsion capabilities.
- 18 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 601-605
A Novel Improved Generalized Normal Distribution Optimization Algorithm for Path Planning of AUV in a 3D Environment
Path planning technology is the key to realizing the precise operation of underwater robots. However, due to the complexity of the underwater environment and the limitations of multiple planning conditions, it is difficult for underwater robots to generate reliable and reasonable paths. In response to this challenge, this paper proposes an improved Generalized Normal Distribution Optimization (GDNO) algorithm. Firstly, the circle chaotic mapping strategy is introduced to ensure the uniform distribution of the initialized population. Secondly, the mathematical optimizer acceleration function is utilized to achieve a balance between the exploration and exploitation strategies. Finally, an improved search strategy based on basic addition and subtraction operations and optimal value tracking is put forward. The experimental results show that the improved GDNO exhibits excellent performance in path planning, providing new perspectives and methods for the research on path planning technology and swarm intelligence algorithms.
- 18 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 601-605
A Novel Improved Generalized Normal Distribution Optimization Algorithm for Path Planning of AUV in a 3D Environment
Path planning technology is the key to realizing the precise operation of underwater robots. However, due to the complexity of the underwater environment and the limitations of multiple planning conditions, it is difficult for underwater robots to generate reliable and reasonable paths. In response to this challenge, this paper proposes an improved Generalized Normal Distribution Optimization (GDNO) algorithm. Firstly, the circle chaotic mapping strategy is introduced to ensure the uniform distribution of the initialized population. Secondly, the mathematical optimizer acceleration function is utilized to achieve a balance between the exploration and exploitation strategies. Finally, an improved search strategy based on basic addition and subtraction operations and optimal value tracking is put forward. The experimental results show that the improved GDNO exhibits excellent performance in path planning, providing new perspectives and methods for the research on path planning technology and swarm intelligence algorithms.
- 20 December 2024
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 36 (1) , 016231
Research on fault component extraction and fault type identification of rotating machinery based on MDSM and a novel convolutional neural network
Given the complexity and difficulty in extracting and recognizing multi-axis mechanical fault components, a method for fault extraction and identification based on the multi-Axis displacement superposition method (MDSM) and a novel convolutional neural network (NCNN) is proposed. In the proposed MDSM method, first, correlation analysis is used to determine the operational status of the mechanical system and to identify the location of faults in the multi-axis rotating mechanical system. Secondly, a simplified initial point selection process is introduced to segment the collected fault component. Subsequently, a signal superposition method with position offset correction is employed to perform position correction and superposition operations on the segmented signals, enhancing the accuracy of the fault signal. Finally, the front end of the superimposed signals is extracted as the fault component, completing the separation and extraction of the fault components. For the extracted fault signals, an NCNN is designed for fault-type identification. NCNN improves computational efficiency and effectively completes fault feature identification through a lightweight network architecture and a nonlinear learning rate scheduling strategy. The results of the experiment show that the proposed method can accurately determine the fault occurrence location, extract the fault components, and achieve high-accuracy fault type identification.
- 20 December 2024
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 36 (1) , 016231
Research on fault component extraction and fault type identification of rotating machinery based on MDSM and a novel convolutional neural network
Given the complexity and difficulty in extracting and recognizing multi-axis mechanical fault components, a method for fault extraction and identification based on the multi-Axis displacement superposition method (MDSM) and a novel convolutional neural network (NCNN) is proposed. In the proposed MDSM method, first, correlation analysis is used to determine the operational status of the mechanical system and to identify the location of faults in the multi-axis rotating mechanical system. Secondly, a simplified initial point selection process is introduced to segment the collected fault component. Subsequently, a signal superposition method with position offset correction is employed to perform position correction and superposition operations on the segmented signals, enhancing the accuracy of the fault signal. Finally, the front end of the superimposed signals is extracted as the fault component, completing the separation and extraction of the fault components. For the extracted fault signals, an NCNN is designed for fault-type identification. NCNN improves computational efficiency and effectively completes fault feature identification through a lightweight network architecture and a nonlinear learning rate scheduling strategy. The results of the experiment show that the proposed method can accurately determine the fault occurrence location, extract the fault components, and achieve high-accuracy fault type identification.
- 28 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (5) , 5476-5487
Improving Deep Q Network Based on Marketing Psychology for AUV Path Planning in Unknown Marine Environments
Autonomous underwater vehicles (AUVs) are commonly utilized in fields like marine environmental testing because of their adaptable and independent characteristics. However, the complexity and variability of the underwater environment present challenges for AUVs. AUVs need to safely and stably navigate through complex unknown underwater environments. The existing path planning technologies for AUVs in unknown marine environments face challenges, such as slow convergence speed and low success rate. In this article, a deep Q-learning (DQN) approach grounded in marketing psychology (MDQN) is suggested to tackle the challenges mentioned above. First, to reasonably allocate the exploration and exploitation abilities in DQN, a mathematical optimizer acceleration function is developed. Additionally, to improve the environmental adaptability of AUVs and speed up the convergence of deep networks, a reward mechanism that takes into account ocean currents and the shortest path principle is suggested. Finally, the reward feedback and experience stratification mechanism are designed based on marketing psychology principles, aiming to promote a virtuous cycle between the training process of deep networks and the learning process of the agent. Various simulation experiments comprehensively demonstrate that MDQN has excellent path planning capabilities in unknown marine environments, effectively addressing the slow convergence speed and low success rate issues of DQN.
- 28 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (5) , 5476-5487
Improving Deep Q Network Based on Marketing Psychology for AUV Path Planning in Unknown Marine Environments
Autonomous underwater vehicles (AUVs) are commonly utilized in fields like marine environmental testing because of their adaptable and independent characteristics. However, the complexity and variability of the underwater environment present challenges for AUVs. AUVs need to safely and stably navigate through complex unknown underwater environments. The existing path planning technologies for AUVs in unknown marine environments face challenges, such as slow convergence speed and low success rate. In this article, a deep Q-learning (DQN) approach grounded in marketing psychology (MDQN) is suggested to tackle the challenges mentioned above. First, to reasonably allocate the exploration and exploitation abilities in DQN, a mathematical optimizer acceleration function is developed. Additionally, to improve the environmental adaptability of AUVs and speed up the convergence of deep networks, a reward mechanism that takes into account ocean currents and the shortest path principle is suggested. Finally, the reward feedback and experience stratification mechanism are designed based on marketing psychology principles, aiming to promote a virtuous cycle between the training process of deep networks and the learning process of the agent. Various simulation experiments comprehensively demonstrate that MDQN has excellent path planning capabilities in unknown marine environments, effectively addressing the slow convergence speed and low success rate issues of DQN.
- Top Cited Papers
- 25 January 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 247
An improved algorithm optimization algorithm based on RungeKutta and golden sine strategy
- 25 January 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 247
Top Cited PapersAn improved algorithm optimization algorithm based on RungeKutta and golden sine strategy
- Open Access
A Multi-Strategy Improved Arithmetic Optimization Algorithm
To improve the performance of the arithmetic optimization algorithm (AOA) and solve problems in the AOA, a novel improved AOA using a multi-strategy approach is proposed. Firstly, circle chaotic mapping is used to increase the diversity of the population. Secondly, a math optimizer accelerated (MOA) function optimized by means of a composite cycloid is proposed to improve the convergence speed of the algorithm. Meanwhile, the symmetry of the composite cycloid is used to balance the global search ability in the early and late iterations. Thirdly, an optimal mutation strategy combining the sparrow elite mutation approach and Cauchy disturbances is used to increase the ability of individuals to jump out of the local optimal. The Rastrigin function is selected as the reference test function to analyze the effectiveness of the improved strategy. Twenty benchmark test functions, algorithm time complexity, the Wilcoxon rank-sum test, and the CEC2019 test set are selected to test the overall performance of the improved algorithm, and the results are then compared with those of other algorithms. The test results show that the improved algorithm has obvious advantages in terms of both its global search ability and convergence speed. Finally, the improved algorithm is applied to an engineering example to further verify its practicability.
Open AccessA Multi-Strategy Improved Arithmetic Optimization Algorithm
To improve the performance of the arithmetic optimization algorithm (AOA) and solve problems in the AOA, a novel improved AOA using a multi-strategy approach is proposed. Firstly, circle chaotic mapping is used to increase the diversity of the population. Secondly, a math optimizer accelerated (MOA) function optimized by means of a composite cycloid is proposed to improve the convergence speed of the algorithm. Meanwhile, the symmetry of the composite cycloid is used to balance the global search ability in the early and late iterations. Thirdly, an optimal mutation strategy combining the sparrow elite mutation approach and Cauchy disturbances is used to increase the ability of individuals to jump out of the local optimal. The Rastrigin function is selected as the reference test function to analyze the effectiveness of the improved strategy. Twenty benchmark test functions, algorithm time complexity, the Wilcoxon rank-sum test, and the CEC2019 test set are selected to test the overall performance of the improved algorithm, and the results are then compared with those of other algorithms. The test results show that the improved algorithm has obvious advantages in terms of both its global search ability and convergence speed. Finally, the improved algorithm is applied to an engineering example to further verify its practicability.
- Open Access
- 25 April 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12 (9) , 4358
Design of a Parallel Quadruped Robot Based on a Novel Intelligent Control System
In order to make a robot track a desired trajectory with high precision and steady gait, a novel intelligent controller was designed based on a new mechanical structure and optimized foot trajectory. Kinematics models in terms of the D-H method were established to analyze the relationship between the angle of the driving joint and the foot position. Inspired by a dog’s diagonal trot on a flat terrain, foot trajectory planning in the swing and support phases without impact were fulfilled based on the compound cycloid improved by the Bézier curve. Both the optimized cascade proportional–integral–derivative (PID) control system and improved fuzzy adaptive PID control system were applied to realize the stable operation of a quadruped robot, and their parameters were optimized by the sparrow search algorithm. The convergence speed and accuracy of the sparrow search algorithm were verified by comparing with the moth flame optimization algorithm and particle swarm optimization algorithm. Finally, a co-simulation with MATLAB and ADAMS was utilized to compare the effects of the two control systems. The results of both displacement and velocity exhibit that the movement of a quadruped bionic robot with fuzzy adaptive PID control systems optimized by the sparrow search algorithm possessed better accuracy and stability than cascade PID control systems. The motion process of the quadruped robot in the co-simulation process also demonstrates the effectiveness of the designed mechanical structure and control system.
- 25 April 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12 (9) , 4358
Open AccessDesign of a Parallel Quadruped Robot Based on a Novel Intelligent Control System
In order to make a robot track a desired trajectory with high precision and steady gait, a novel intelligent controller was designed based on a new mechanical structure and optimized foot trajectory. Kinematics models in terms of the D-H method were established to analyze the relationship between the angle of the driving joint and the foot position. Inspired by a dog’s diagonal trot on a flat terrain, foot trajectory planning in the swing and support phases without impact were fulfilled based on the compound cycloid improved by the Bézier curve. Both the optimized cascade proportional–integral–derivative (PID) control system and improved fuzzy adaptive PID control system were applied to realize the stable operation of a quadruped robot, and their parameters were optimized by the sparrow search algorithm. The convergence speed and accuracy of the sparrow search algorithm were verified by comparing with the moth flame optimization algorithm and particle swarm optimization algorithm. Finally, a co-simulation with MATLAB and ADAMS was utilized to compare the effects of the two control systems. The results of both displacement and velocity exhibit that the movement of a quadruped bionic robot with fuzzy adaptive PID control systems optimized by the sparrow search algorithm possessed better accuracy and stability than cascade PID control systems. The motion process of the quadruped robot in the co-simulation process also demonstrates the effectiveness of the designed mechanical structure and control system.