Control Engineering Practice最新文献_第2页

Observer-based sliding mode admittance control for ultrasound robot force-tracking in complex interaction environments 基于观测器的滑模导纳控制在复杂交互环境下的超声机器人力跟踪

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-13 DOI: 10.1016/j.conengprac.2026.106764

Sen Li , Qi Chen , Weihua Li , Mingfeng Wang , Sungsung Pang , Kai Wu

Robotic ultrasound systems can improve imaging consistency and reduce operator workload by maintaining stable probe–tissue contact forces. However, variations in tissue stiffness and surface position often degrade force-tracking performance. This paper presents a disturbance observer–based sliding mode admittance control (DOSMAC) framework to address these challenges. Stiffness and position variations are uniformly modeled as lumped external disturbances, estimated online by a disturbance observer and compensated through sliding mode control. The resulting signal is introduced as a virtual input to the admittance model, enhancing robustness without requiring online stiffness identification or extensive parameter tuning. The stability of the closed-loop system is established through Lyapunov analysis. Simulations and experiments with stiffness ranging from 500 to 3500 N/m verify the effectiveness of the proposed approach. In experiments with simultaneous stiffness and position variations, DOSMAC reduces overshoot by 29.1% and 55.4%, peak force error by 4.36 N and 0.46 N, and root mean square error by 1.79 N and 0.4 N, respectively, compared with conventional admittance control and adaptive variable admittance control. These results demonstrate that the proposed method enables stable and reliable force tracking under complex time-varying conditions, supporting the clinical translation forceof robotic ultrasound.

机器人超声系统可以提高成像的一致性，并通过保持稳定的探针组织接触力来减少操作员的工作量。然而，组织刚度和表面位置的变化通常会降低力跟踪性能。本文提出了一种基于干扰观测器的滑模导纳控制（DOSMAC）框架来解决这些问题。刚度和位置变化统一建模为集总外部扰动，由扰动观测器在线估计，并通过滑模控制进行补偿。结果信号作为虚拟输入引入导纳模型，增强鲁棒性，而不需要在线刚度识别或广泛的参数调整。通过李雅普诺夫分析，建立了闭环系统的稳定性。在500 ~ 3500 N/m的刚度范围内进行了仿真和实验，验证了该方法的有效性。在刚度和位置同时变化的实验中，与传统导纳控制和自适应变导纳控制相比，DOSMAC控制的超调量分别降低29.1%和55.4%，峰值力误差分别降低4.36 N和0.46 N，均方根误差分别降低1.79 N和0.4 N。结果表明，该方法能够在复杂时变条件下实现稳定可靠的力跟踪，为机器人超声的临床平移力提供支持。

{"title":"Observer-based sliding mode admittance control for ultrasound robot force-tracking in complex interaction environments","authors":"Sen Li , Qi Chen , Weihua Li , Mingfeng Wang , Sungsung Pang , Kai Wu","doi":"10.1016/j.conengprac.2026.106764","DOIUrl":"10.1016/j.conengprac.2026.106764","url":null,"abstract":"<div><div>Robotic ultrasound systems can improve imaging consistency and reduce operator workload by maintaining stable probe–tissue contact forces. However, variations in tissue stiffness and surface position often degrade force-tracking performance. This paper presents a disturbance observer–based sliding mode admittance control (DOSMAC) framework to address these challenges. Stiffness and position variations are uniformly modeled as lumped external disturbances, estimated online by a disturbance observer and compensated through sliding mode control. The resulting signal is introduced as a virtual input to the admittance model, enhancing robustness without requiring online stiffness identification or extensive parameter tuning. The stability of the closed-loop system is established through Lyapunov analysis. Simulations and experiments with stiffness ranging from 500 to 3500 N/m verify the effectiveness of the proposed approach. In experiments with simultaneous stiffness and position variations, DOSMAC reduces overshoot by 29.1% and 55.4%, peak force error by 4.36 N and 0.46 N, and root mean square error by 1.79 N and 0.4 N, respectively, compared with conventional admittance control and adaptive variable admittance control. These results demonstrate that the proposed method enables stable and reliable force tracking under complex time-varying conditions, supporting the clinical translation forceof robotic ultrasound.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"169 ","pages":"Article 106764"},"PeriodicalIF":4.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980022","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Full actuation control for rail-less electric bus trains 无轨电动客车列车的全驱动控制

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-13 DOI: 10.1016/j.conengprac.2026.106760

Shuo Wang , Baorui Wang , Guoxiang Gu

This paper proposes rail-less trains, composed of electric buses which are coupled together. The dynamic model of the rail-less electric bus train (REBT) involves not only nonlinearities but also unknown and uncertain parameters, which pose significant challenges. To mitigate model nonlinearities and parameter uncertainties in practical REBT system, we develop centralized and distributed adaptive control laws based on high-order full actuation (HOFA) control. For developing the centralized adaptive control law, we propose a novel adaptive control method, integrating gamma-projection operators with sigma-modification in adaptive estimation. This method robustly constrains parameter estimates within the known set while suppressing drift and offering tradeoffs between the magnitudes of control signal and tracking performance. For developing the distributed adaptive control law, we propose a different adaptive control method, employing both autonomous and cooperative control actions and using again the gamma-projection. In addition, adaptive estimation is aided by an off-line least-squares (LS) algorithm that ensures the adaptive estimates to converge to the true system parameters under the persistent excitation condition, leading to asymptotic feedback linearization and global asymptotic stabilization. Disturbance rejection in the framework of

H_{\infty}

-control is studied for the linearized REBT system. It is shown that the two proposed adaptive control laws ensure the

H_{\infty}

-norm from the input disturbance to the output tracking errors for velocity and inter-EB-distance controls to be strictly smaller than any γ > 0, effectively suppressing energy bounded disturbances in the worst-case. The simulation study includes industrial-level simulators and validates the proposed adaptive control methods.

本文提出了一种由电动客车耦合而成的无轨列车。无轨电动客车列车（REBT）的动力学模型不仅涉及非线性，而且涉及未知和不确定参数，这给整车动力学模型的建立带来了很大的挑战。为了缓解实际REBT系统中的模型非线性和参数不确定性，提出了基于高阶全致动（HOFA）控制的集中式和分布式自适应控制律。为了建立集中自适应控制律，我们提出了一种新的自适应控制方法，将伽玛投影算子与自适应估计中的西格玛修正相结合。该方法鲁棒地约束了已知集合内的参数估计，同时抑制了漂移，并在控制信号的大小和跟踪性能之间提供了折衷。为了发展分布式自适应控制律，我们提出了一种不同的自适应控制方法，采用自主和合作控制动作，并再次使用伽马投影。此外，采用离线最小二乘（LS）算法辅助自适应估计，确保自适应估计在持续激励条件下收敛到系统的真实参数，从而实现渐近反馈线性化和全局渐近稳定。研究了线性化REBT系统在H∞控制框架下的抗扰性。结果表明，所提出的两种自适应控制律保证了速度和eb -距离控制从输入扰动到输出跟踪误差的H∞范数严格小于任何γ >； 0，在最坏情况下有效地抑制了能量有界扰动。仿真研究包括工业级模拟器，并验证了所提出的自适应控制方法。

{"title":"Full actuation control for rail-less electric bus trains","authors":"Shuo Wang , Baorui Wang , Guoxiang Gu","doi":"10.1016/j.conengprac.2026.106760","DOIUrl":"10.1016/j.conengprac.2026.106760","url":null,"abstract":"<div><div>This paper proposes rail-less trains, composed of electric buses which are coupled together. The dynamic model of the rail-less electric bus train (REBT) involves not only nonlinearities but also unknown and uncertain parameters, which pose significant challenges. To mitigate model nonlinearities and parameter uncertainties in practical REBT system, we develop centralized and distributed adaptive control laws based on high-order full actuation (HOFA) control. For developing the centralized adaptive control law, we propose a novel adaptive control method, integrating gamma-projection operators with sigma-modification in adaptive estimation. This method robustly constrains parameter estimates within the known set while suppressing drift and offering tradeoffs between the magnitudes of control signal and tracking performance. For developing the distributed adaptive control law, we propose a different adaptive control method, employing both autonomous and cooperative control actions and using again the gamma-projection. In addition, adaptive estimation is aided by an off-line least-squares (LS) algorithm that ensures the adaptive estimates to converge to the true system parameters under the persistent excitation condition, leading to asymptotic feedback linearization and global asymptotic stabilization. Disturbance rejection in the framework of <span><math><msub><mi>H</mi><mi>∞</mi></msub></math></span>-control is studied for the linearized REBT system. It is shown that the two proposed adaptive control laws ensure the <span><math><msub><mi>H</mi><mi>∞</mi></msub></math></span>-norm from the input disturbance to the output tracking errors for velocity and inter-EB-distance controls to be strictly smaller than any <em>γ</em> > 0, effectively suppressing energy bounded disturbances in the worst-case. The simulation study includes industrial-level simulators and validates the proposed adaptive control methods.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"169 ","pages":"Article 106760"},"PeriodicalIF":4.6,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980023","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Longitudinal acceleration shaping and priority-aware control allocation for tilt-trirotor UAVs 倾转三旋翼无人机纵向加速度整形与优先级感知控制分配

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-13 DOI: 10.1016/j.conengprac.2026.106778

Xiaomei Cheng , Xutao Qu , Jiang Zhao , Ningjun Liu , Zhihao Cai , Yingxun Wang

Tilt-trirotor Unmanned Aerial Vehicles (UAVs) face coupled challenges in generating longitudinal acceleration and mapping virtual control inputs to actuator commands. This paper develops an integrated flight control strategy addressing both problems. First, a smooth nonlinear pitch-tilt coordination function is formulated to map desired longitudinal acceleration to attitude commands, ensuring continuous derivatives during acceleration reversals and improving tracking performance. Second, a structure-preserving control allocation framework is established using intermediate variables, expanding the feasible allocation space and enabling a priority-preserving saturation mechanism that safeguards critical channels such as roll while limiting less essential ones like yaw. Flight experiments demonstrate that, compared with baseline approaches, the proposed strategy reduces position error by up to 45.4% and enhances stability under demanding flight conditions.

倾转三旋翼无人机（uav）在产生纵向加速度和将虚拟控制输入映射到执行器命令方面面临着双重挑战。本文提出了一种综合飞行控制策略来解决这两个问题。首先，建立了平滑非线性俯仰-倾斜协调函数，将期望的纵向加速度映射到姿态指令，确保加速度反转时的连续导数，提高跟踪性能。其次，利用中间变量建立了保结构控制分配框架，扩大了可行分配空间，实现了保优先级饱和机制，既保护了滚转等关键通道，又限制了偏航等次要通道。飞行实验表明，与基线方法相比，该策略在苛刻的飞行条件下，将位置误差降低了45.4%，并提高了稳定性。

引用次数: 0

Distributionally robust fault detection trade-off design with prior fault information 基于先验故障信息的分布式鲁棒故障检测权衡设计

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-12 DOI: 10.1016/j.conengprac.2026.106758

Yulin Feng , Hailang Jin , Steven X. Ding , Hao Ye , Chao Shang

The robustness of fault detection algorithms against uncertainty is crucial in the real-world industrial environment. Recently, a new probabilistic design scheme called distributionally robust fault detection (DRFD) has emerged and received immense interest. Despite its robustness against unknown distributions in practice, current DRFD focuses on the overall detectability of all possible faults rather than the detectability of critical faults that are a priori known. Hence, a new DRFD trade-off design scheme is put forward in this work by utilizing prior fault information. The key contribution includes a novel distributional robustness metric of detecting a known fault and a new relaxed distributionally robust chance constraint that ensures robust detectability. Then, a new DRFD design problem of fault detection under unknown probability distributions is proposed, and this offers a flexible balance between the robustness of detecting known critical faults and the overall detectability against all possible faults. To address the resulting semi-infinite chance-constrained problem, we first reformulate it to a finite-dimensional problem characterized by bilinear matrix inequalities. Subsequently, a tailored heuristic solution algorithm is developed, which includes a sequential minimization procedure and an initialization strategy. Finally, case studies on a simulated three-tank system and a real-world battery cell are carried out to showcase the effectiveness of the proposed heuristic algorithm and the advantages of our DRFD method.

在现实工业环境中，故障检测算法对不确定性的鲁棒性至关重要。最近，出现了一种新的概率设计方案，称为分布式鲁棒故障检测（DRFD），并引起了人们的极大兴趣。尽管在实践中对未知分布具有鲁棒性，但目前的DRFD侧重于所有可能故障的总体可检测性，而不是先验已知的关键故障的可检测性。因此，本文提出了一种利用先验故障信息的DRFD权衡设计方案。关键贡献包括一种新的检测已知故障的分布鲁棒性度量和一种新的松弛分布鲁棒机会约束，以确保鲁棒可检测性。然后，提出了一种未知概率分布下的故障检测DRFD设计问题，在检测已知关键故障的鲁棒性和对所有可能故障的总体可检测性之间实现了灵活的平衡。为了解决由此产生的半无限机会约束问题，我们首先将其重新表述为一个以双线性矩阵不等式为特征的有限维问题。随后，开发了一种定制的启发式求解算法，该算法包括顺序最小化过程和初始化策略。最后，以模拟的三罐系统和现实世界的电池为例进行了案例研究，以展示所提出的启发式算法的有效性和我们的DRFD方法的优势。

{"title":"Distributionally robust fault detection trade-off design with prior fault information","authors":"Yulin Feng , Hailang Jin , Steven X. Ding , Hao Ye , Chao Shang","doi":"10.1016/j.conengprac.2026.106758","DOIUrl":"10.1016/j.conengprac.2026.106758","url":null,"abstract":"<div><div>The robustness of fault detection algorithms against uncertainty is crucial in the real-world industrial environment. Recently, a new probabilistic design scheme called distributionally robust fault detection (DRFD) has emerged and received immense interest. Despite its robustness against unknown distributions in practice, current DRFD focuses on the overall detectability of all possible faults rather than the detectability of critical faults that are <em>a priori</em> known. Hence, a new DRFD trade-off design scheme is put forward in this work by utilizing prior fault information. The key contribution includes a novel distributional robustness metric of detecting a known fault and a new relaxed distributionally robust chance constraint that ensures robust detectability. Then, a new DRFD design problem of fault detection under unknown probability distributions is proposed, and this offers a flexible balance between the robustness of detecting known critical faults and the overall detectability against all possible faults. To address the resulting semi-infinite chance-constrained problem, we first reformulate it to a finite-dimensional problem characterized by bilinear matrix inequalities. Subsequently, a tailored heuristic solution algorithm is developed, which includes a sequential minimization procedure and an initialization strategy. Finally, case studies on a simulated three-tank system and a real-world battery cell are carried out to showcase the effectiveness of the proposed heuristic algorithm and the advantages of our DRFD method.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"169 ","pages":"Article 106758"},"PeriodicalIF":4.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979888","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Modeling and cascade-based robust dynamic control of a hyper-constrained parallel mechanism for high-precision robotic tasks 高精度机器人任务超约束并联机构建模及基于级联的鲁棒动态控制

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-12 DOI: 10.1016/j.conengprac.2025.106756

Marco Fernandes dos Santos Xaud , Pål Johan From , Antonio Candea Leite

This paper presents the mechanical design, constructive aspects, kinematic modeling and control design methodology for a hyper-constrained parallel mechanism (HCPM), composed of six closed kinematic chains and twelve spherical joints, which is developed to carry out high-precision robotic tasks in industrial and agricultural environments. Here, we employ a systematic modeling methodology which considers the kinematic constraints of the mechanism from its structure equations, rather than explicitly using the constraint equations. This allows us to describe the kinematic constraints in the operation velocity space instead of the joint configuration space. From this analytical approach, we can compute the velocity of the non-actuated joints as a function of the velocity of the actuated ones. The control design uses an inverse kinematics algorithm based on the pseudo-inverse Jacobian matrix. In order to deal with many potential singular configurations which may occur during the task execution, we consider a recently proposed approach, called the Filtered Inverse method, which dynamically estimates the inverse of the Jacobian matrix instead of computing its true inverse instantaneously. The dynamic control of the HCPM is achieved using a simplified, Lagrangian-derived model to provide the computed-torque feedforward term for nonlinear compensation and decoupling, enabling motion simulation for high-speed trajectories while dismissing the need for a full explicit model, which is complex and difficult to obtain. Despite the simplifications, a robust cascade control strategy is proposed—featuring a second-order sliding mode (STA) compensator—to handle modeling uncertainties effectively. 3D computer modeling, numerical simulations, and laboratory experiments with two prototypes of the hyper-constrained parallel mechanism were conducted to validate the proposed approach and demonstrate its feasibility for high-precision robotic tasks.

本文介绍了一种由6个封闭运动链和12个球面关节组成的超约束并联机构（HCPM）的机械设计、构造、运动学建模和控制设计方法，该机构是为在工农业环境中执行高精度机器人任务而开发的。在这里，我们采用了一种系统的建模方法，该方法从其结构方程中考虑机构的运动约束，而不是明确地使用约束方程。这使得我们可以在运动速度空间而不是关节位形空间中描述运动约束。根据这种分析方法，我们可以计算出非驱动关节的速度作为驱动关节速度的函数。控制设计采用基于伪逆雅可比矩阵的逆运动学算法。为了处理在任务执行过程中可能出现的许多潜在的奇异构型，我们考虑了最近提出的一种方法，称为滤波逆方法，它动态估计雅可比矩阵的逆，而不是立即计算它的真逆。HCPM的动态控制采用简化的拉格朗日模型来实现，该模型为非线性补偿和解耦提供了计算扭矩前馈项，从而实现了高速轨迹的运动仿真，同时省去了复杂且难以获得的完整显式模型的需要。尽管进行了简化，但提出了一种鲁棒级联控制策略-具有二阶滑模（STA）补偿器-以有效地处理建模不确定性。对两种超约束并联机构进行了三维计算机建模、数值模拟和实验室实验，验证了所提出的方法，并证明了其在高精度机器人任务中的可行性。

{"title":"Modeling and cascade-based robust dynamic control of a hyper-constrained parallel mechanism for high-precision robotic tasks","authors":"Marco Fernandes dos Santos Xaud , Pål Johan From , Antonio Candea Leite","doi":"10.1016/j.conengprac.2025.106756","DOIUrl":"10.1016/j.conengprac.2025.106756","url":null,"abstract":"<div><div>This paper presents the mechanical design, constructive aspects, kinematic modeling and control design methodology for a hyper-constrained parallel mechanism (HCPM), composed of six closed kinematic chains and twelve spherical joints, which is developed to carry out high-precision robotic tasks in industrial and agricultural environments. Here, we employ a systematic modeling methodology which considers the kinematic constraints of the mechanism from its structure equations, rather than explicitly using the constraint equations. This allows us to describe the kinematic constraints in the operation velocity space instead of the joint configuration space. From this analytical approach, we can compute the velocity of the non-actuated joints as a function of the velocity of the actuated ones. The control design uses an inverse kinematics algorithm based on the pseudo-inverse Jacobian matrix. In order to deal with many potential singular configurations which may occur during the task execution, we consider a recently proposed approach, called the Filtered Inverse method, which dynamically estimates the inverse of the Jacobian matrix instead of computing its true inverse instantaneously. The dynamic control of the HCPM is achieved using a simplified, Lagrangian-derived model to provide the computed-torque feedforward term for nonlinear compensation and decoupling, enabling motion simulation for high-speed trajectories while dismissing the need for a full explicit model, which is complex and difficult to obtain. Despite the simplifications, a robust cascade control strategy is proposed—featuring a second-order sliding mode (STA) compensator—to handle modeling uncertainties effectively. 3D computer modeling, numerical simulations, and laboratory experiments with two prototypes of the hyper-constrained parallel mechanism were conducted to validate the proposed approach and demonstrate its feasibility for high-precision robotic tasks.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"169 ","pages":"Article 106756"},"PeriodicalIF":4.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979889","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A continual causal learning architecture with multiscale graph attention for robust mold level fluctuation prediction in smart continuous casting systems 基于多尺度图关注的连续因果学习体系结构用于智能连铸系统中模位波动的鲁棒预测

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-11 DOI: 10.1016/j.conengprac.2026.106759

Yun-Jie Pan , Yan-Ning Sun , Wei Qin , Zeng-Gui Gao , Li-Lan Liu

In continuous casting operations, mold level fluctuation (MLF) triggers slag entrapment within molten steel, resulting in surface imperfections of cast slabs, which underscores the imperative for anticipatory MLF prediction to guarantee both quality reliability and system safety. The nonlinear coupling interactions among multiple process parameters and dynamic time-varying operating conditions pose significant challenges to maintaining the sustained robust performance of current data-driven prediction models. Therefore, this study proposes a novel continual causal learning architecture integrating multiscale transfer entropy graph attention network with elastic weight consolidation. The framework achieves temporal-adaptive prediction through multiscale transfer entropy quantification of nonlinear parameter interactions, encoded as dynamic causal graphs to guide attention-based feature aggregation. An elastic weight consolidation mechanism preserves critical information learned from historical conditions while assimilating new operational knowledge, overcoming the catastrophic forgetting dilemma in traditional models. Sensor-specific embedding modules further enhance cross-condition generalization by extracting invariant features from different operating conditions. Experimental validation using real-world continuous casting data demonstrates superior performance, with RMSE reductions of 3.43% – 34.9% compared to mainstream methods across three operating conditions, while achieving an average R² of 0.745. The proposed architecture provides a deployable solution for online MLF monitoring, enabling adaptive quality control in smart continuous casting systems through its interpretable causal reasoning and multi-condition adaptability.

在连铸作业中，结晶器液位波动（MLF）会引发钢水中夹渣，导致铸坯表面缺陷，因此提前预测结晶器液位波动是保证质量可靠性和系统安全的必要条件。多过程参数和动态时变工况之间的非线性耦合相互作用对当前数据驱动预测模型的持续鲁棒性提出了重大挑战。因此，本研究提出了一种将多尺度转移熵图注意网络与弹性权巩固相结合的连续因果学习架构。该框架通过非线性参数相互作用的多尺度传递熵量化来实现时间自适应预测，并将其编码为动态因果图来指导基于注意力的特征聚合。弹性权重巩固机制在吸收新的操作知识的同时，保留了从历史条件中学习到的关键信息，克服了传统模型中的灾难性遗忘困境。针对传感器的嵌入模块通过从不同工况中提取不变性特征，进一步增强了跨工况泛化能力。使用实际连铸数据进行的实验验证表明，与主流方法相比，该方法在三种工况下的RMSE降低了3.43% - 34.9%，平均R2为0.745。所提出的体系结构为在线MLF监测提供了可部署的解决方案，通过其可解释的因果推理和多条件适应性，实现智能连铸系统的自适应质量控制。

{"title":"A continual causal learning architecture with multiscale graph attention for robust mold level fluctuation prediction in smart continuous casting systems","authors":"Yun-Jie Pan , Yan-Ning Sun , Wei Qin , Zeng-Gui Gao , Li-Lan Liu","doi":"10.1016/j.conengprac.2026.106759","DOIUrl":"10.1016/j.conengprac.2026.106759","url":null,"abstract":"<div><div>In continuous casting operations, mold level fluctuation (MLF) triggers slag entrapment within molten steel, resulting in surface imperfections of cast slabs, which underscores the imperative for anticipatory MLF prediction to guarantee both quality reliability and system safety. The nonlinear coupling interactions among multiple process parameters and dynamic time-varying operating conditions pose significant challenges to maintaining the sustained robust performance of current data-driven prediction models. Therefore, this study proposes a novel continual causal learning architecture integrating multiscale transfer entropy graph attention network with elastic weight consolidation. The framework achieves temporal-adaptive prediction through multiscale transfer entropy quantification of nonlinear parameter interactions, encoded as dynamic causal graphs to guide attention-based feature aggregation. An elastic weight consolidation mechanism preserves critical information learned from historical conditions while assimilating new operational knowledge, overcoming the catastrophic forgetting dilemma in traditional models. Sensor-specific embedding modules further enhance cross-condition generalization by extracting invariant features from different operating conditions. Experimental validation using real-world continuous casting data demonstrates superior performance, with RMSE reductions of 3.43% – 34.9% compared to mainstream methods across three operating conditions, while achieving an average <em>R</em><sup>2</sup> of 0.745. The proposed architecture provides a deployable solution for online MLF monitoring, enabling adaptive quality control in smart continuous casting systems through its interpretable causal reasoning and multi-condition adaptability.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"169 ","pages":"Article 106759"},"PeriodicalIF":4.6,"publicationDate":"2026-01-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980024","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Robust jumping control of quadruped robots under body and leg disturbances using reinforcement learning with H∞ regularization 基于H∞正则化强化学习的四足机器人身体和腿部扰动鲁棒跳跃控制

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-10 DOI: 10.1016/j.conengprac.2025.106751

Shuo Xue , Liang Yuan , Kai Lv , Teng Ran , Wendong Xiao , Jianbo Zhang

Quadruped robots rely on jumping ability to traverse discontinuous terrain and complex obstacles, allowing them to perform challenging tasks in dynamic and unpredictable environments. However, jumping control faces multiple challenges, including high-dimensional nonlinear dynamics, underactuation during flight, external disturbances, which make it difficult to ensure stability and robustness. Current reinforcement learning methods usually apply random disturbances only on the body of robot during training to improve jumping robustness. However, due to the high randomness, low accuracy and narrow application scope, the training effect is limited. To address these issues, this study proposes a reinforcement learning framework that systematically introduces disturbances on both the body and legs to improve overall stability. An adaptation module is designed to predict environmental and disturbance information, enhancing the quality of training data. Furthermore, a learnable disturbance generator based on H_∞ regularization is introduced to dynamically generate appropriate disturbances according to the performance of the robot. The learned policy is deployed on a real quadruped robot and evaluated in complex indoor and outdoor environments. Experimental results demonstrate strong robustness and generalization of the learned policy, enabling stable and reliable jumping performance.

四足机器人依靠跳跃能力穿越不连续的地形和复杂的障碍物，使它们能够在动态和不可预测的环境中执行具有挑战性的任务。然而，跳跃控制面临着高维非线性动力学、飞行欠驱动、外部扰动等诸多挑战，难以保证其稳定性和鲁棒性。目前的强化学习方法通常只在训练过程中对机器人身体施加随机扰动，以提高跳跃的鲁棒性。但由于随机性大、准确率低、适用范围窄，训练效果有限。为了解决这些问题，本研究提出了一个强化学习框架，系统地在身体和腿部引入干扰，以提高整体稳定性。设计了自适应模块来预测环境和干扰信息，提高了训练数据的质量。在此基础上，引入了基于H∞正则化的可学习扰动发生器，根据机器人的性能动态产生适当的扰动。将学习到的策略部署在一个真实的四足机器人上，并在复杂的室内和室外环境中进行评估。实验结果表明，该策略具有较强的鲁棒性和泛化性，实现了稳定可靠的跳跃性能。

{"title":"Robust jumping control of quadruped robots under body and leg disturbances using reinforcement learning with H∞ regularization","authors":"Shuo Xue , Liang Yuan , Kai Lv , Teng Ran , Wendong Xiao , Jianbo Zhang","doi":"10.1016/j.conengprac.2025.106751","DOIUrl":"10.1016/j.conengprac.2025.106751","url":null,"abstract":"<div><div>Quadruped robots rely on jumping ability to traverse discontinuous terrain and complex obstacles, allowing them to perform challenging tasks in dynamic and unpredictable environments. However, jumping control faces multiple challenges, including high-dimensional nonlinear dynamics, underactuation during flight, external disturbances, which make it difficult to ensure stability and robustness. Current reinforcement learning methods usually apply random disturbances only on the body of robot during training to improve jumping robustness. However, due to the high randomness, low accuracy and narrow application scope, the training effect is limited. To address these issues, this study proposes a reinforcement learning framework that systematically introduces disturbances on both the body and legs to improve overall stability. An adaptation module is designed to predict environmental and disturbance information, enhancing the quality of training data. Furthermore, a learnable disturbance generator based on <em>H</em><sub>∞</sub> regularization is introduced to dynamically generate appropriate disturbances according to the performance of the robot. The learned policy is deployed on a real quadruped robot and evaluated in complex indoor and outdoor environments. Experimental results demonstrate strong robustness and generalization of the learned policy, enabling stable and reliable jumping performance.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"169 ","pages":"Article 106751"},"PeriodicalIF":4.6,"publicationDate":"2026-01-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145928901","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Contraction-based active disturbance rejection controller for an active ankle foot orthosis 基于收缩的主动干扰抑制控制器用于主动踝足矫形器

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-09 DOI: 10.1016/j.conengprac.2026.106757

Rami Jradi , Hala Rifaï , José Fermi Guerrero-Castellanos , Samer Mohammed

In this paper, a control strategy for an Actuated Ankle Foot Orthosis (AAFO) is proposed to provide the assistance needed by the wearer at the ankle joint level. The control scheme is based on a contraction-based active disturbance rejection controller (Cont-ADRC). It includes an estimation of the human muscular torque and difficult-to-capture external torques affecting the AAFO-wearer system at the ankle joint level alongside unmodeled dynamics by means of a nonlinear disturbance observer (NDOB). A contraction-based variable gain controller determines the amount of assistance to be provided by the AAFO to perform the movement in complement to the aforementioned muscular torque. The variable gain controller provides a compromise between the low frequency disturbance rejection and the high frequency measurement noise attenuation. Using a contraction-based differential Lyapunov analysis, the trajectories of the AAFO-wearer system subject to the proposed active disturbance rejection controller are proved to be incrementally bounded, which is considered to be a stronger form of boundedness with respect to the uniform one. To demonstrate the efficiency of the Cont-ADRC, it has been applied in real-time experiments with robustness tests, involving three healthy subjects during walking activities. The outcomes revealed its superiority over other ADRCs developed for wearable robotics where it showed improved tracking accuracy compared to PID and Control Lyapunov Functions-based ADRC and reduced computational efforts compared to adaptive-based ADRC.

本文提出了一种驱动式踝足矫形器（AAFO）的控制策略，以在踝关节水平提供穿戴者所需的辅助。该控制方案基于基于收缩的自抗扰控制器（control - adrc）。它包括通过非线性干扰观测器（NDOB）对影响aafo -穿戴者系统的踝关节水平的人体肌肉扭矩和难以捕获的外部扭矩的估计，以及未建模的动力学。基于收缩的可变增益控制器决定了AAFO执行运动所需的辅助量，以补充上述肌肉扭矩。可变增益控制器提供了低频干扰抑制和高频测量噪声衰减之间的折衷。利用基于收缩的微分Lyapunov分析，证明了aafo -穿戴者系统在主动抗扰控制器控制下的轨迹是增量有界的，这被认为是相对于均匀有界的一种更强的有界形式。为了验证控制自适应控制的有效性，我们将其应用于3名健康受试者步行活动的实时实验中，并进行了鲁棒性测试。结果显示其优于其他为可穿戴机器人开发的ADRC，与基于PID和控制Lyapunov函数的ADRC相比，它具有更高的跟踪精度，并且与基于自适应的ADRC相比减少了计算量。

{"title":"Contraction-based active disturbance rejection controller for an active ankle foot orthosis","authors":"Rami Jradi , Hala Rifaï , José Fermi Guerrero-Castellanos , Samer Mohammed","doi":"10.1016/j.conengprac.2026.106757","DOIUrl":"10.1016/j.conengprac.2026.106757","url":null,"abstract":"<div><div>In this paper, a control strategy for an Actuated Ankle Foot Orthosis (AAFO) is proposed to provide the assistance needed by the wearer at the ankle joint level. The control scheme is based on a contraction-based active disturbance rejection controller (Cont-ADRC). It includes an estimation of the human muscular torque and difficult-to-capture external torques affecting the AAFO-wearer system at the ankle joint level alongside unmodeled dynamics by means of a nonlinear disturbance observer (NDOB). A contraction-based variable gain controller determines the amount of assistance to be provided by the AAFO to perform the movement in complement to the aforementioned muscular torque. The variable gain controller provides a compromise between the low frequency disturbance rejection and the high frequency measurement noise attenuation. Using a contraction-based differential Lyapunov analysis, the trajectories of the AAFO-wearer system subject to the proposed active disturbance rejection controller are proved to be incrementally bounded, which is considered to be a stronger form of boundedness with respect to the uniform one. To demonstrate the efficiency of the Cont-ADRC, it has been applied in real-time experiments with robustness tests, involving three healthy subjects during walking activities. The outcomes revealed its superiority over other ADRCs developed for wearable robotics where it showed improved tracking accuracy compared to PID and Control Lyapunov Functions-based ADRC and reduced computational efforts compared to adaptive-based ADRC.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"169 ","pages":"Article 106757"},"PeriodicalIF":4.6,"publicationDate":"2026-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145928904","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

ODKP: A multivariate time series predictor based on orthogonal dynamic Koopman operator for blast furnace permeability index prediction 基于正交动态Koopman算子的多变量时间序列预测器用于高炉渗透率指数预测

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-08 DOI: 10.1016/j.conengprac.2025.106749

Mingyang Gao , Lina Wang , Xiangrong Song , Zhuoqing Li , Dazhong Ma

In blast furnace (BF) smelting, the permeability index (PI) is a key metric for evaluating whether the process is progressing toward optimal operating conditions. However, developing an effective prediction method for PI is challenging due to the influence of entangled temporal dependencies and the intrinsic non-stationarity of BF time series data. To address these challenges, a multivariate time series predictor based on orthogonal dynamic Koopman for BF PI prediction is proposed in this paper. Firstly, a feature decoupling (FD) module is designed, which uses a data-adaptive transformation based on an orthogonal matrix to effectively mitigate the impact of redundancy introduced by coupled variables. The feature decoupling module can be incorporated into the predictor to more effective the encoding and decoding within the decorrelated feature space. Subsequently, the multivariate BF time series, segmented using a sliding-window strategy, are projected into the Koopman embedding space via a multi-layer perceptron to extract interpretable dynamic modes. Furthermore, context-aware Koopman operator calculations are adaptively performed using extended dynamic mode decomposition (eDMD) across different temporal windows. This approach enables the approximation of non-stationary dynamics as locally linear, capturing the temporal evolution within each segment. Finally, comparative simulations with state-of-the-art models demonstrate that the proposed method achieves superior PI prediction performance.

在高炉冶炼中，渗透系数（PI）是衡量冶炼过程是否朝着最佳运行状态发展的关键指标。然而，由于BF时间序列数据的纠缠时间依赖性和固有非平稳性的影响，开发一种有效的PI预测方法是具有挑战性的。为了解决这些问题，本文提出了一种基于正交动态库普曼的多元时间序列预测器用于BF PI预测。首先，设计了特征解耦（FD）模块，采用基于正交矩阵的数据自适应变换，有效减轻了耦合变量引入的冗余影响；在预测器中加入特征解耦模块，可以在去相关特征空间内更有效地进行编码和解码。然后，使用滑动窗口策略分割多元BF时间序列，通过多层感知器将其投影到Koopman嵌入空间中以提取可解释的动态模式。此外，使用扩展动态模式分解（eDMD）跨不同时间窗口自适应地执行上下文感知的Koopman算子计算。这种方法使非平稳动态近似为局部线性，捕获每个段内的时间演变。最后，与最先进的模型进行了对比仿真，结果表明该方法具有较好的PI预测性能。

{"title":"ODKP: A multivariate time series predictor based on orthogonal dynamic Koopman operator for blast furnace permeability index prediction","authors":"Mingyang Gao , Lina Wang , Xiangrong Song , Zhuoqing Li , Dazhong Ma","doi":"10.1016/j.conengprac.2025.106749","DOIUrl":"10.1016/j.conengprac.2025.106749","url":null,"abstract":"<div><div>In blast furnace (BF) smelting, the permeability index (PI) is a key metric for evaluating whether the process is progressing toward optimal operating conditions. However, developing an effective prediction method for PI is challenging due to the influence of entangled temporal dependencies and the intrinsic non-stationarity of BF time series data. To address these challenges, a multivariate time series predictor based on orthogonal dynamic Koopman for BF PI prediction is proposed in this paper. Firstly, a feature decoupling (FD) module is designed, which uses a data-adaptive transformation based on an orthogonal matrix to effectively mitigate the impact of redundancy introduced by coupled variables. The feature decoupling module can be incorporated into the predictor to more effective the encoding and decoding within the decorrelated feature space. Subsequently, the multivariate BF time series, segmented using a sliding-window strategy, are projected into the Koopman embedding space via a multi-layer perceptron to extract interpretable dynamic modes. Furthermore, context-aware Koopman operator calculations are adaptively performed using extended dynamic mode decomposition (eDMD) across different temporal windows. This approach enables the approximation of non-stationary dynamics as locally linear, capturing the temporal evolution within each segment. Finally, comparative simulations with state-of-the-art models demonstrate that the proposed method achieves superior PI prediction performance.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"169 ","pages":"Article 106749"},"PeriodicalIF":4.6,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145928900","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Realistic process simulator for control strategy evaluation in continuous direct compaction tablet manufacturing 面向连续直接压片生产控制策略评价的逼真过程模拟器

IF 4.6 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Control Engineering Practice

Pub Date : 2026-01-08 DOI: 10.1016/j.conengprac.2025.106712

Dana Copot, Bora Ayvaz, Erhan Yumuk

This study introduces a mechanistic, high-fidelity simulation environment for continuous direct compaction (CDC) tablet manufacturing, integrating feeders, blender, hopper, and tablet press models into a unified framework. The platform realistically captures material flow dynamics, residence time effects, and multivariable interactions, providing a robust virtual environment for in silico control design and testing. A disturbance-aware feedforward-feedback control architecture is evaluated across realistic scenarios, including upstream flowrate and concentration disturbances, raw material variability (±10-20% bulk density changes), and model-plant mismatch. Results demonstrate the platform’s capability to systematically assess control performance and robustness, supporting safe optimization of manufacturing processes. This work lays the foundation for scalable, model-informed control development and de-risking of future plant-wide control strategies, aligning with the Pharma 4.0 vision of predictive and adaptive continuous pharmaceutical manufacturing.

本研究为连续直接压实（CDC）片剂生产引入了一个机械的、高保真的仿真环境，将给料机、搅拌器、料斗和压片机模型集成到一个统一的框架中。该平台逼真地捕捉了物料流动动力学、停留时间效应和多变量相互作用，为硅控制设计和测试提供了一个强大的虚拟环境。干扰感知的前馈-反馈控制架构在现实场景中进行评估，包括上游流量和浓度干扰、原材料可变性（±10-20%体积密度变化）以及模型-工厂不匹配。结果表明，该平台能够系统地评估控制性能和鲁棒性，支持制造过程的安全优化。这项工作为可扩展的、模型知情的控制开发和未来全厂控制策略的降低风险奠定了基础，与预测和自适应连续制药制造的制药4.0愿景保持一致。

引用次数: 0