The flexible pectoral-fin robotic manta ray has become a hot spot in underwater robotics research due to its remarkable manoeuvrability, swimming efficiency, and stealth capabilities. However, its complex dynamics pose challenges to the effectiveness of conventional modeling and control approaches. In this study, we present a novel position control strategy for robotic manta rays by harnessing data-driven methodologies and sample-observed deep reinforcement learning techniques. First, we designed and fabricated a robotic manta ray prototype and introduced a central pattern generator (CPG) model based on a phase oscillator. This model enables smooth transitions between multiple swimming modes by adjusting a single parameter. Next, we developed a data-driven kinematics (DK) model to reduce the simulation-reality gap. We combined the Soft Actor-Critic (SAC) algorithm with the DK model and sample observation (SO) to develop the SODK-SAC control strategy, which improves training efficiency and policy migration capability. Simulation results demonstrate that this control strategy achieves task efficiency improvements of 17.2% and 35.8% over the DK-SAC and PID control strategies, respectively. Experiments verify superior performance from simulation to reality: the proposed approach incurs only a 3.3% increase in path length, significantly outperforming traditional control methods.

ObjectiveEfficient and accurate solutions to the forward kinematic problem are critical technical challenges for achieving real-time control of parallel platforms. Existing algorithms, however, often exhibit poor generalizability, low computational efficiency, and limited accuracy. A forward kinematic solution algorithm for parallel platforms integrating a deep residual network and error compensation was proposed to address these issues.MethodsFirstly, the nonlinear mapping from joint space to pose space was realized using the proposed deep residual network model. Secondly, an iterative process with an error compensation strategy was adopted to further improve the computational accuracy of the deep residual network model. Finally, simulation analysis was carried out on two 3-DOF parallel platforms with different configurations.ResultsThe results show that the efficiency of the proposed algorithm is improved by approximately 91%, 87%, and 58% at minimum compared with the Newton-Raphson method, the back propagation-Newton Raphson hybrid strategy algorithm, and the quasi-Newton method, respectively. The proposed algorithm can efficiently solve the forward kinematic problem of 3-DOF parallel platforms, meet the computational requirements for real-time control of parallel platforms, and be easily applied to other configurations of parallel platforms.

Parallel platforms are widely used in simulators and precision machining due to their high precision and high rigidity. Trajectory planning is crucial to achieve high-performance motion control for platforms, while existing algorithms suffer from complex kinematics, difficult multi-objective optimization and low efficiency. Therefore, this thesis combines Efficient Forward Kinematics (EFK) with Deep Deterministic Policy Gradients (DDPG) to form the EFKD trajectory planning algorithm. Simulation and experimental results show the convergence speed and success rate of the EFKD algorithm are improved by at least 40% compared with the Cartesian Space Linear (CSL) algorithm. The speed and acceleration indicators are continuously smooth, which effectively suppresses the problem of sudden acceleration change.

Parallel platforms are widely used in modern industry due to their superior dynamic and static properties. However, different parallel platforms have different closure structures, and their forward kinematics equations have high nonlinearity, leading to poor generalization, low efficiency, and precision in solving these equations. This study presents a novel RNSA-EC algorithm designed to enhance the computational precision and efficiency of forward kinematics for parallel platforms. The algorithm integrates a Self-Attention optimized Residual Network (RNSA) with an Error Compensation (EC) method. The RNSA model initially achieves high-precision nonlinear mapping from joint coordinates to pose coordinates of parallel platforms. Subsequently, the inverse kinematic model is employed to develop an EC method, which further reduces computational errors in the forward kinematics of the parallel platform by iteratively updating the inputs to the RNSA model. We applied the RNSA-EC algorithm to solve the forward kinematics of two 6-degree-of-freedom (6-DOF) parallel platforms with distinct structures. The results demonstrate that the algorithm is approximately 60%, 50%, and 45% more efficient than the Newton-Raphson (NR), the hybrid Backpropagation Neural Network (BPNN)-NR, and the Iterative Artificial Neural Network (IANN) methods, respectively, on the two parallel platforms. Furthermore, the algorithm can be extended to solve forward kinematics for parallel platforms with different structures and DOFs.

Currently, the most commonly used method to study the hydrodynamic performance of manta rays is computational fluid dynamics (CFD) simulation. In this research, we investigated the effects of kinematic parameters—specifically wave number, amplitude, and frequency—on the hydrodynamic performance of manta rays during the swimming process by constructing a 2D CFD model. First, we verified the reasonableness of the 2D simulation. Subsequently, a 2D simulation was used to study the hydrodynamic performance of manta ray pectoral fins, and it was concluded that using low-amplitude, high-frequency propulsion with an optimal wave number has better energy utilization. Finally, we conducted orthogonal experiments, which revealed that the thrust reaches a maximum value of 8.55 N at a frequency of 1 Hz, amplitude of 0.3 c, and wave number of 0.4, and the quasi-propulsive efficiency reaches a maximum value of 82.4% at a frequency of 0.8 Hz, amplitude of 0.3 c, and wave number of 0.4. In general, we can regulate the wave number to a range of 0.35 to 0.4, the frequency to between 0.7 and 0.9 Hz, and the amplitude to between 0.3 c and 0.325 c. This configuration yields a thrust exceeding 3.04 N and a quasi-propulsive efficiency surpassing 70.4%.

With the rapid growth of the shipbuilding industry, anchor chain flash welding quality inspection has become a critical concern. Due to the challenge of collecting abnormal state data, most anchor chain samples are normal, making anomaly detection from limited anomalous data a key research problem. Therefore, this paper establishes the LSAC-DPP GAN model based on Least Squares Generative Adversarial Networks (LS GAN) by adding a Determinantal Point Process (DPP)-based extra loss term to the generator's loss function. From a semi-supervised perspective, the LSAC-DPP GAN learns normal and abnormal samples, enabling the model to avoid generating abnormal samples explicitly. The problem of pattern collapse that exists in the model is mitigated by adding a loss term based on DPP to the generator's loss function, allowing the generator to generate samples with better diversity. Finally, we trained the LSAC-DPP GAN with electrode position and current signal data to generate normal displacement and signal data. High recall, accuracy, and F1 scores were obtained on the test set.

The robotic manta has attracted significant interest for its exceptional maneuverability, swimming efficiency, and stealthiness. However, achieving efficient autonomous swimming in complex underwater environments presents a significant challenge. To address this issue, this study integrates Deep Deterministic Policy Gradient (DDPG) with Central Pattern Generators (CPGs) and proposes a CPG-based DDPG control strategy. First, we designed a CPG control strategy that can more precisely mimic the swimming behavior of the manta. Then, we implemented the DDPG algorithm as a high-level controller that adaptively modifies the CPG’s control parameters based on the real-time state information of the robotic manta. This adjustment allows for the regulation of swimming modes to fulfill specific tasks. The proposed strategy underwent initial training and testing in a simulated environment before deployment on a robotic manta prototype for field trials. Both further simulation and experimental results validate the effectiveness and practicality of the proposed control strategy.

High-flow hydraulic servo systems are extensively employed in contemporary industrial applications due to their considerable flow capacity and cost-effectiveness. Nonetheless, hydraulic servo systems frequently encounter unpredictable internal and external disturbances, and high-flow proportional directional valves always have unsatisfactory hydraulic characteristics, compromising the precision and robustness of high-flow hydraulic servo systems. This study proposes a novel control strategy integrating the Soft Actor-Critic (SAC) reinforcement learning algorithm with Adaptive Robust Control (ARC) to enhance system performance. This approach features a two-tiered controller: the upper controller utilizes the SAC algorithm to learn and adapt to the dynamics of the hydraulic servo system, iteratively refining the lower controller’s hyperparameters. Meanwhile, grounded in the ARC strategy, the lower controller executes real-time control of the hydraulic servo system. The simulation and experimental results demonstrate that the proposed control strategy can effectively adjust the control hyperparameters according to the learned system dynamic and tracking errors. Consequently, this approach enhances control precision amidst varying external and internal disturbances. Moreover, this control strategy is anticipated to realize high-flow, high-precision, and high-robust hydraulic servo systems, which can be used in various fields such as marine and offshore engineering.

Redundant degree-of-freedom (DOF) manipulators offer increased flexibility and are better suited for obstacle avoidance, yet precise control of these systems remains a significant challenge. This paper addresses the issues of slow training convergence and suboptimal stability that plague current deep reinforcement learning (DRL)-based control strategies for redundant DOF manipulators. We propose a novel DRL-based intelligent control strategy, FK-DRL, which integrates the manipulator’s forward kinematics (FK) model into the control framework. Initially, we conceptualize the control task as a Markov decision process (MDP) and construct the FK model for the manipulator. Subsequently, we expound on the integration principles and training procedures for amalgamating the FK model with existing DRL algorithms. Our experimental analysis, applied to 7-DOF and 4-DOF manipulators in simulated and real-world environments, evaluates the FK-DRL strategy’s performance. The results indicate that compared to classical DRL algorithms, the FK-DDPG, FK-TD3, and FK-SAC algorithms improved the success rates of intelligent control tasks for the 7-DOF manipulator by 21%, 87%, and 64%, respectively, and the training convergence speeds increased by 21%, 18%, and 68%, respectively. These outcomes validate the proposed algorithm’s effectiveness and advantages in redundant manipulator control using DRL and FK models.

Flash butt welding, a mainstream welding method employed in producing anchor chains, is a critical manufacturing process affecting the quality of anchor chains. Ultrasonic and load testing are used to evaluate the welding quality of anchor chains, but the cost of checking and replacing unqualified chain links is high. A deep learning-based quality evaluation method for flash butt welding is proposed to reduce the cost of detecting and replacing substandard chain links. First, displacement and current sensors collect electrode position and current signals during welding. Second, since the number of qualified anchor links is much larger than that of unqualified ones, a new data synthesis method is proposed: nearest-neighbor splicing sampling, which achieves the enhancement of minority samples by segmenting and combining existing data samples according to the features of anchor chain welding. Then, a piecewise linear interpolation method is used to handle the varying data length problem, thus satisfying the input requirements of the convolutional neural network (CNN). Finally, a CNN model is established, and dropout is used to reduce the over-fitting phenomenon. The experimental results show that the accuracy of the under-sampling method, over-sampling method, and nearest-neighbor splicing sampling method are 93.8%, 95.9%, and 96.3%, respectively, and the sensitivity, specificity, and accuracy of the CNN model are 95.7%, 93%, and 94.3%, respectively, which are better than those of the support vector machine (SVM).

Proportional–integral–derivative (PID) control is the most common control technique used in hydraulic servo control systems. However, the nonlinearity and uncertainty of the hydraulic system make it challenging for PID control to achieve high-precision control. This paper proposes a novel control strategy that combines the soft actor-critic (SAC) reinforcement learning algorithm with the PID method to address this issue. The proposed control strategy consists of an upper-level controller based on the SAC algorithm and a lower-level controller based on the PID control method. The upper-level controller continuously tunes the control parameters of the lower-level controller based on the tracking error and system status. The lower-level controller performs real-time control for the hydraulic servo system with a control frequency 10 times higher than the upper controllers. Simulation experiments demonstrate that the proposed SAC-PID control strategy can effectively address disturbances and achieve high precision control for hydraulic servo control systems in uncertain working conditions compared with PID and fuzzy PID control methods. Therefore, the proposed control strategy offers a promising approach to improving the tracking performance of hydraulic servo systems.

Modern industries face increasing demands to improve the precision and workability of equipment, leading to stringent requirements for hydraulic servo systems in terms of flow, accuracy, and output power. High‐flow servo valves and multi‐valve synchronous control methods have seen more applications in the industry to meet these demands. However, high‐flow servo valves are expensive and have unsatisfactory performance, and synchronous control methods require servo valves to have the same hydraulic characteristics, resulting in higher cost and lower accuracy of high‐flow hydraulic servo systems. Also, very little has been conducted on controlling single actuators using multiple valves. In this study, we design a novel dual‐valve hydraulic servo system structure with a high‐flow proportional directional valve that is connected in parallel with a low‐flow servo valve and further develop the mathematical model of this dual‐valve system. Next, a harmonic control scheme was proposed, which included a flow allocation layer and trajectory tracking layer. Thereby, we achieved optimal control of two valves, and robust performance was guaranteed. Finally, we investigate the relationships between the flows of two valves, the deadband of the proportional directional valve, and trajectory tracking errors. Experimental results show that the proposed control scheme can effectively adjust the output flow of each valve dynamically according to the tracking trajectory and the hydraulic characteristics of the two valves, and the tracking performance of the servo system is significantly improved. This method is promised to enable the realization of an economical, high‐flow, high‐precision hydraulic servo system that can be generally used in various fields, such as marine engineering and construction.

船舶锚泊系统设计与性能仿真实验是一项设计性、综合性、实践性很强的实验,将"虚拟现实+互联网"技术融入实验教学项目,坚持"学生中心"的原则,面向船舶与海洋工程行业,突出锚泊舾装设计关键技术,开发了虚拟仿真实验项目,丰富了船舶及海工机械装备设计课程内涵.学生通过参加具有开发性、发散性的设计过程,提高了对大型复杂船舶辅助机械系统的设计能力、创新开发能力及独立解决问题能力.该虚拟仿真实验在培养具有高效安全可靠设计理念的现代船舶及海工机械装备复合型创新人才中发挥了积极作用.

传统的锚链质量检测方法主要是出厂前的破断试验和拉力试验,这使得企业无法在焊接过程中及时发现有缺陷的链环,从而导致需要耗费巨大的成本去替换不合格链环.针对这一问题,提出一种基于卷积神经网络的锚链闪光焊接质量检测方法.对于锚链闪光焊接的合格样本远多于不合格样本这一现象,提出一种新的不平衡数据预处理方法:最近邻拼接采样;通过分段线性插值的方法使不同样本的数据长度一致;搭建一个卷积神经网络,用于学习锚链闪光焊接中电极位置曲线和电流曲线的特征;采用增量学习的方式训练新样本.实验结果表明,该方法能高效地检测锚链闪光焊接的质量,准确率可达96.8％,且增量学习后的模型对旧样本同样能准确识别.

With advancements in photoelectric technology and computer image processing technology, the visual measurement method based on point clouds is gradually being applied to the 3D measurement of large workpieces. Point cloud registration is a key step in 3D measurement, and its registration accuracy directly affects the accuracy of 3D measurements. In this study, we designed a novel MPCR-Net for multiple partial point cloud registration networks. First, an ideal point cloud was extracted from the CAD model of the workpiece and used as the global template. Next, a deep neural network was used to search for the corresponding point groups between each partial point cloud and the global template point cloud. Then, the rigid body transformation matrix was learned according to these correspondence point groups to realize the registration of each partial point cloud. Finally, the iterative closest point algorithm was used to optimize the registration results to obtain the final point cloud model of the workpiece. We conducted point cloud registration experiments on untrained models and actual workpieces, and by comparing them with existing point cloud registration methods, we verified that the MPCR-Net could improve the accuracy and robustness of the 3D point cloud registration.

新型冠状病毒肺炎的蔓延影响了高校的正常课堂教学,尤其是对于国际学生(留学生),因此文章基于微信群平台,采用"基于工程问题"的教学方法,以"复杂零件精度设计"为目标,开展了"互换性与测量技术基础"课程线上教学实践,突破了时间、空间及新型冠状病毒肺炎的限制,保障了此期间的教学质量和效果.

机器固定约束作业车间规定加工某道工序的机器仅有一台,不符合车间实际生产情况.针对其局限性,考虑某工序有多台机器可供选择的可变机器约束,建立综合机器使用成本和延期惩罚费用两方面因素的单目标优化模型,提出改进反转变异法、双交叉以及指数衰减法的遗传退火算法,求解可变机器约束作业车间调度问题(variable machine constraints job-shop scheduling problems,VMCJSSP).仿真发现,与传统遗传算法相比,该算法使生产成本节约45％,最小加工等待时间缩短37％;最后,基于该算法对VMCJSSP、机器固定约束问题进行调度仿真.结果表明,相对于机器固定约束,该算法模拟的生产成本降低58％、最小加工等待时间缩短11％,具有求解大计算量车间调度问题的高效性.

Taking the 300 kg-class single blade of the controllable pitch propeller as the research object, based on the principle of three-point gravity measurement, a secondary weighing method is developed, which obtained the relative coordinate value of the gravity center by making a difference between the two weighing data. According to the concept that each axis is independent, an error model of spatial attitude is established. The attitude deviation is obtained by using sensor scanning, and the actual coordinate value of the blade gravity center is obtained by combining with the space attitude transformation matrix. The experimental results indicated that the secondary weighing method with the space attitude transformation matrix in this paper was useful and highly accurate. The method has certain guiding significance for the gravity center measurement for workpieces with large and complex curved surface.

Recurrence plots are widely used to represent and characterize recurrence behaviors of complex systems. Although the recurrence plot captures a great deal of useful information, very little has been done to leverage convolutional neural networks (CNNs) regarding the learning of recurrence dynamics. In this paper, we develop a new CNN approach to investigate recurrence patterns in multisensor signals for real-time anomaly detection. The proposed methodology is evaluated and validated in both simulation studies and a real-world case study for quality control of a flash welding process, which is commonly used to manufacture anchor chains in the shipbuilding industry. First, we develop a new sensing system to collect electrical-current and electrode-position profiles in the flash welding process. Second, recurrence plots are derived with multisensor signals collected from the manufacturing process of each workpiece in the anchor chain. Third, CNNs are developed to learn workpiece-to-workpiece variations in the recurrence plots. Experimental results show that the proposed CNN models of recurrence plots yield superior performance for anomaly detection of welding quality variations, improving the automation level of flash welding processes.

This article presents a novel thruster fault diagnosis approach for an autonomous underwater vehicle. In the novel approach, a time-frequency entropy enhancement is used to extract feature, and then a boundary constraint–assisted relative gray relational grade is applied to identify thruster fault. The time-frequency entropy enhancement is developed from the smoothed pseudo Wigner–Ville distribution combined with Shannon entropy. First, the energy distributions of autonomous underwater vehicle dynamic signals are given in the time-frequency plane. And then the energy concentration in the energy distribution is enhanced based on a serial signal processing, including wavelet decomposition, modified Bayes’ classification algorithm, and two dimensional convolution operation, successively. After that the Shannon entropy of the energy distribution is calculated. The boundary constraint–assisted relative gray relational grade comes from the gray relational analysis. A mapping function between the relative gray relational grade and the fault severity is established. And then the boundary constraints of relative gray relational grades at each standard fault level are determined. Moreover, the mapping function is modified based on the boundary constraints. Experiments are performed on an experimental prototype autonomous underwater vehicle in a pool. The experimental results demonstrate the effectiveness of the developed approaches in terms of improving the sensitivity of the fault feature to the fault severity, compared with the smoothed pseudo Wigner–Ville distribution combined with Shannon entropy, and increasing the identification accuracy, compared with the gray relational analysis.