Journal of Process Control最新文献_第5页

Model predictive control of pressure-swing distillation via closed-loop system identification 基于闭环系统辨识的变压蒸馏模型预测控制

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-11-10 DOI: 10.1016/j.jprocont.2025.103589

Daye Yang , Jingcheng Wang , Naiyi Ban , Yanjiu Zhong , David Shan-Hill Wong , Abolhassan Razminia , Chengtian Cui

Pressure-swing distillation (PSD) is a proven technique for separating azeotropic mixtures by exploiting pressure-dependent shifts in azeotropic composition. Despite its efficacy, PSD systems present significant control challenges due to inherent nonlinearities, complex multivariable interactions, and internal recycle loops. This study proposes a model predictive control (MPC) framework for PSD systems, founded on closed-loop system identification. A comprehensive plantwide nonlinear dynamic model of a PSD process for separating a maximum-boiling azeotrope of acetone and chloroform is developed using Aspen Dynamics and interfaced with MATLAB/Simulink for controller design and testing. To address the limitations of open-loop excitation in systems with recycles, pseudo-random binary sequence (PRBS) signals are applied under closed-loop operation to sufficiently excite the process. Subsequently, linear state-space models are identified using the prediction error method. Based on these models, two MPC configurations are developed: temperature control (TC) and composition–temperature cascade control (CC–TC). Simulation results demonstrate that the proposed MPC strategies quantitatively outperform proportional–integral (PI) controllers. Specifically, under the TC strategy, the total integral of absolute error (IAE) values of

X_{D 1, A C E}

and

X_{D 2, C H L}

are reduced by approximately 10% and 3%, respectively; while under the CC–TC strategy, the reductions reach about 26% and 55%. Moreover, across four disturbance scenarios, the steady convergence times of both composition purities are shortened by more than 5 h compared with PI controllers. These results highlight the advantages of the proposed MPC strategies in disturbance rejection and transient product quality regulation. These findings underscore the effectiveness of closed-loop system identification as a basis for advanced control of PSD processes.

变压蒸馏（PSD）是一种成熟的分离共沸混合物的技术，它利用了共沸成分的压力相关变化。尽管PSD系统很有效，但由于其固有的非线性、复杂的多变量相互作用和内部循环循环，PSD系统在控制方面存在重大挑战。本文提出了一种基于闭环辨识的PSD系统模型预测控制框架。利用Aspen Dynamics软件建立了丙酮和氯仿最高沸点共沸物PSD分离过程的全厂非线性动力学模型，并结合MATLAB/Simulink进行了控制器设计和测试。为了解决循环系统开环激励的局限性，在闭环操作下应用伪随机二值序列（PRBS）信号来充分激励过程。然后，利用预测误差法对线性状态空间模型进行识别。基于这些模型，提出了两种MPC配置：温度控制（TC）和成分-温度级联控制（CC-TC）。仿真结果表明，所提出的MPC策略在数量上优于比例积分（PI）控制器。具体而言，在TC策略下，XD1、ACE和XD2的绝对误差（IAE）值的总积分，CHL分别降低了约10%和3%；而在CC-TC战略下，减排幅度分别达到26%和55%左右。此外，在四种干扰情况下，与PI控制器相比，两种组合纯度的稳定收敛时间缩短了5小时以上。这些结果突出了所提出的MPC策略在干扰抑制和瞬态产品质量调节方面的优势。这些发现强调了闭环系统识别作为PSD过程高级控制基础的有效性。

{"title":"Model predictive control of pressure-swing distillation via closed-loop system identification","authors":"Daye Yang , Jingcheng Wang , Naiyi Ban , Yanjiu Zhong , David Shan-Hill Wong , Abolhassan Razminia , Chengtian Cui","doi":"10.1016/j.jprocont.2025.103589","DOIUrl":"10.1016/j.jprocont.2025.103589","url":null,"abstract":"<div><div>Pressure-swing distillation (PSD) is a proven technique for separating azeotropic mixtures by exploiting pressure-dependent shifts in azeotropic composition. Despite its efficacy, PSD systems present significant control challenges due to inherent nonlinearities, complex multivariable interactions, and internal recycle loops. This study proposes a model predictive control (MPC) framework for PSD systems, founded on closed-loop system identification. A comprehensive plantwide nonlinear dynamic model of a PSD process for separating a maximum-boiling azeotrope of acetone and chloroform is developed using Aspen Dynamics and interfaced with MATLAB/Simulink for controller design and testing. To address the limitations of open-loop excitation in systems with recycles, pseudo-random binary sequence (PRBS) signals are applied under closed-loop operation to sufficiently excite the process. Subsequently, linear state-space models are identified using the prediction error method. Based on these models, two MPC configurations are developed: temperature control (TC) and composition–temperature cascade control (CC–TC). Simulation results demonstrate that the proposed MPC strategies quantitatively outperform proportional–integral (PI) controllers. Specifically, under the TC strategy, the total integral of absolute error (IAE) values of <span><math><msub><mrow><mi>X</mi></mrow><mrow><mi>D</mi><mn>1</mn><mo>,</mo><mi>A</mi><mi>C</mi><mi>E</mi></mrow></msub></math></span> and <span><math><msub><mrow><mi>X</mi></mrow><mrow><mi>D</mi><mn>2</mn><mo>,</mo><mi>C</mi><mi>H</mi><mi>L</mi></mrow></msub></math></span> are reduced by approximately 10% and 3%, respectively; while under the CC–TC strategy, the reductions reach about 26% and 55%. Moreover, across four disturbance scenarios, the steady convergence times of both composition purities are shortened by more than 5 h compared with PI controllers. These results highlight the advantages of the proposed MPC strategies in disturbance rejection and transient product quality regulation. These findings underscore the effectiveness of closed-loop system identification as a basis for advanced control of PSD processes.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"156 ","pages":"Article 103589"},"PeriodicalIF":3.9,"publicationDate":"2025-11-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145520304","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

Layer-wise information-aggregation-decoupled convolutional self-attention network guided by process knowledge for quality-related process monitoring 基于过程知识的分层信息聚合解耦卷积自关注网络用于质量相关过程监控

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-11-10 DOI: 10.1016/j.jprocont.2025.103580

Yuguo Yang, Hongbo Shi, Bing Song, Yang Tao, Keyu Yao, Hongyu Tian

Large-scale industrial processes produce process data that have the characteristics of high dimensionality, nonlinearity, and strong coupling. Process monitoring plays a vital role in ensuring production safety and product quality. At present, for quality-related fault detection, the existing methods have the problems of weak robustness for the extracted features, large model parameters, and black-box modeling. To address these problems, this paper proposes the layer-wise information-aggregation-decoupled convolutional self-attention network (LIA-DCSA). First, a one-dimensional decoupled convolutional self-attention network (DCSA) is constructed to explicitly extract complex features between variables guided by process knowledge. Second, the rules of relationship inheritance and relationship elimination are proposed to construct different levels of quality-related variable relationship graphs (QR-VRGs). The QR-VRGs and DCSA are combined to achieve effective extraction and layer-by-layer aggregation of quality-related features. Then, based on Kullback–Leibler (KL) divergence, a distribution-constrained regression layer is designed to regularize the quality-related features gathered in neurons. Finally, the Tennessee Eastman process and the Cranfield multiphase flow process are used to show the effectiveness of the proposed method. The experimental results show that compared with the other methods, this method has the best fault monitoring performance while effectively reducing the number of model parameters. Furthermore, the quality-related features extracted by LIA-DCSA show the best robustness in the presence of interference from Gaussian noise and impulse noise.

大规模工业过程产生的过程数据具有高维、非线性和强耦合的特点。过程监控对保证生产安全和产品质量起着至关重要的作用。目前，对于质量相关故障检测，现有方法存在提取特征鲁棒性弱、模型参数大、黑箱建模等问题。为了解决这些问题，本文提出了分层信息聚合解耦卷积自关注网络（LIA-DCSA）。首先，构建一维解耦卷积自关注网络（DCSA），在过程知识的引导下显式提取变量之间的复杂特征；其次，提出关系继承和关系消除规则，构建不同层次的质量相关变量关系图（qr - vrg）。将qr - vrg和DCSA相结合，实现了质量相关特征的有效提取和逐层聚合。然后，基于Kullback-Leibler （KL）散度，设计分布约束回归层，对神经元中收集到的质量相关特征进行正则化。最后，以Tennessee Eastman过程和Cranfield多相流过程为例，验证了该方法的有效性。实验结果表明，与其他方法相比，该方法在有效减少模型参数数量的同时，具有最佳的故障监测性能。此外，LIA-DCSA提取的质量相关特征在高斯噪声和脉冲噪声干扰下具有最佳的鲁棒性。

{"title":"Layer-wise information-aggregation-decoupled convolutional self-attention network guided by process knowledge for quality-related process monitoring","authors":"Yuguo Yang, Hongbo Shi, Bing Song, Yang Tao, Keyu Yao, Hongyu Tian","doi":"10.1016/j.jprocont.2025.103580","DOIUrl":"10.1016/j.jprocont.2025.103580","url":null,"abstract":"<div><div>Large-scale industrial processes produce process data that have the characteristics of high dimensionality, nonlinearity, and strong coupling. Process monitoring plays a vital role in ensuring production safety and product quality. At present, for quality-related fault detection, the existing methods have the problems of weak robustness for the extracted features, large model parameters, and black-box modeling. To address these problems, this paper proposes the layer-wise information-aggregation-decoupled convolutional self-attention network (LIA-DCSA). First, a one-dimensional decoupled convolutional self-attention network (DCSA) is constructed to explicitly extract complex features between variables guided by process knowledge. Second, the rules of relationship inheritance and relationship elimination are proposed to construct different levels of quality-related variable relationship graphs (QR-VRGs). The QR-VRGs and DCSA are combined to achieve effective extraction and layer-by-layer aggregation of quality-related features. Then, based on Kullback–Leibler (KL) divergence, a distribution-constrained regression layer is designed to regularize the quality-related features gathered in neurons. Finally, the Tennessee Eastman process and the Cranfield multiphase flow process are used to show the effectiveness of the proposed method. The experimental results show that compared with the other methods, this method has the best fault monitoring performance while effectively reducing the number of model parameters. Furthermore, the quality-related features extracted by LIA-DCSA show the best robustness in the presence of interference from Gaussian noise and impulse noise.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"156 ","pages":"Article 103580"},"PeriodicalIF":3.9,"publicationDate":"2025-11-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145520305","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

Capturing stirring–diffusion and product inhibition in fed-batch fermentation process: A distributed parameter modeling framework 间歇发酵过程中搅拌扩散和产物抑制的捕获：一个分布式参数建模框架

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-11-06 DOI: 10.1016/j.jprocont.2025.103587

Bingke Zhou, Shunyi Zhao, Runsheng Guo, Fei Liu, Xiaoli Luan

Agitation intensity exerts a decisive influence on product yield in microbial fermentation, yet most published work treats mixing effects qualitatively and lacks a quantitative link between stirring conditions and product synthesis. This study develops a distributed parameter system (DPS) model that couples partial differential equations (PDEs) with ordinary differential equations (ODEs) to capture how agitation shapes axial substrate gradients and product formation. Building on the Birol kinetics, the model introduces an axial diffusion coefficient

D_{Z}

(m² s^$-1$) to translate stirring speed into mixing efficiency, and a product-inhibition constant to represent feedback from local product accumulation on microbial metabolism. Theoretical analysis establishes the well-posedness of the model, and numerical simulations illustrate that insufficient agitation leads to strong substrate gradients and reduced productivity, whereas stronger agitation promotes uniform mixing and higher yields. In addition, a laboratory visualization experiment confirmed the agitation-dependent attenuation of gradients, providing further support for the proposed model. Together, these results highlight the DPS model’s potential to guide the design of effective agitation strategies and improve the efficiency of large-scale fermentation processes.

在微生物发酵过程中，搅拌强度对产物产率有决定性的影响，但大多数已发表的研究都定性地看待混合效应，缺乏搅拌条件与产物合成之间的定量联系。本研究开发了一个分布式参数系统（DPS）模型，该模型将偏微分方程（PDEs）与常微分方程（ode）耦合在一起，以捕捉搅拌如何塑造轴向基底梯度和产物形成。该模型以Birol动力学为基础，引入轴向扩散系数DZ （m2 s$-1$）将搅拌速度转化为混合效率，并引入产物抑制常数来表示微生物代谢的局部产物积累反馈。理论分析证实了该模型的拟合性，数值模拟结果表明，搅拌不足导致底物梯度强，导致生产率降低，而较强的搅拌则促进混合均匀，提高产量。此外，实验室可视化实验证实了梯度随搅拌的衰减，为所提模型提供了进一步的支持。总之，这些结果突出了DPS模型在指导有效搅拌策略设计和提高大规模发酵过程效率方面的潜力。

{"title":"Capturing stirring–diffusion and product inhibition in fed-batch fermentation process: A distributed parameter modeling framework","authors":"Bingke Zhou, Shunyi Zhao, Runsheng Guo, Fei Liu, Xiaoli Luan","doi":"10.1016/j.jprocont.2025.103587","DOIUrl":"10.1016/j.jprocont.2025.103587","url":null,"abstract":"<div><div>Agitation intensity exerts a decisive influence on product yield in microbial fermentation, yet most published work treats mixing effects qualitatively and lacks a quantitative link between stirring conditions and product synthesis. This study develops a distributed parameter system (DPS) model that couples partial differential equations (PDEs) with ordinary differential equations (ODEs) to capture how agitation shapes axial substrate gradients and product formation. Building on the Birol kinetics, the model introduces an axial diffusion coefficient <span><math><msub><mrow><mi>D</mi></mrow><mrow><mi>Z</mi></mrow></msub></math></span> (m<sup>2</sup> <!-->s<sup>$-1$</sup>) to translate stirring speed into mixing efficiency, and a product-inhibition constant to represent feedback from local product accumulation on microbial metabolism. Theoretical analysis establishes the well-posedness of the model, and numerical simulations illustrate that insufficient agitation leads to strong substrate gradients and reduced productivity, whereas stronger agitation promotes uniform mixing and higher yields. In addition, a laboratory visualization experiment confirmed the agitation-dependent attenuation of gradients, providing further support for the proposed model. Together, these results highlight the DPS model’s potential to guide the design of effective agitation strategies and improve the efficiency of large-scale fermentation processes.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"156 ","pages":"Article 103587"},"PeriodicalIF":3.9,"publicationDate":"2025-11-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145468045","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

Fast decentralized multi-stage model predictive control using sensitivity-based path-following and a nonsmooth Newton method 基于灵敏度路径跟踪和非光滑牛顿法的快速分散多阶段模型预测控制

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-11-04 DOI: 10.1016/j.jprocont.2025.103570

Simen Bjorvand, Johannes Jäschke

This work addresses the challenge of computational complexity in Multistage Model Predictive Control (MPC). Multistage MPC is a robust control algorithm where uncertainty is accounted for by constructing a finite set of scenarios for different realizations of uncertainty. The resulting Multistage MPC problem becomes large and computationally expensive to solve. Several strategies have been proposed to reduce computation time and to make real-time implementation possible. A popular class of methods is decomposition methods, in which the Multistage MPC problem is decomposed into several smaller parametric subproblems that can be solved in parallel. The subproblems are repeatedly solved while a coordinator algorithm adjusts a parameter to recover the solution of the full problem. The main computational cost in this approach comes from (1) the coordinator algorithm requiring many iterations to converge, resulting in many resolves of the subproblems, and (2) the computational time to solve the subproblems. In this paper we propose a Predictor–Corrector based path-following algorithm to reduce the solution time of the subproblems for a primal-decomposition algorithm for Multistage MPC. A new Predictor–Corrector methodology based on nonsmooth equation solving is proposed with local superlinear/quadratic convergence. The algorithm path-follows along the parameter path given by the coordinator algorithm. Our path-following algorithm is combined with the Extended Newton algorithm from Bjorvand and Jäschke (2023) for reducing number of coordinator steps. The proposed algorithm is applied to a gas-lift system where the total number of iterations of the algorithm are significantly reduced compared to the standard primal decomposition algorithm for Multistage MPC.

这项工作解决了多阶段模型预测控制（MPC）中计算复杂性的挑战。多级MPC是一种鲁棒控制算法，它通过构建一组有限的场景来考虑不确定性的不同实现。由此产生的多阶段MPC问题变得庞大且计算成本高。提出了几种策略来减少计算时间并使实时实现成为可能。一类流行的方法是分解方法，其中多阶段MPC问题被分解成几个较小的参数子问题，这些子问题可以并行求解。通过协调器算法调整参数恢复整个问题的解，重复求解子问题。该方法的主要计算成本来自(1)协调器算法需要多次迭代才能收敛，从而导致子问题的多次求解；(2)求解子问题的计算时间。本文提出了一种基于预测校正器的路径跟踪算法，以减少多阶段MPC原始分解算法子问题的求解时间。提出了一种局部超线性/二次收敛的基于非光滑方程求解的预测校正方法。算法路径遵循协调器算法给出的参数路径。我们的路径跟踪算法与Bjorvand和Jäschke（2023）的扩展牛顿算法相结合，以减少协调器步骤的数量。该算法应用于气举系统，与多级MPC的标准原始分解算法相比，该算法的总迭代次数显著减少。

{"title":"Fast decentralized multi-stage model predictive control using sensitivity-based path-following and a nonsmooth Newton method","authors":"Simen Bjorvand, Johannes Jäschke","doi":"10.1016/j.jprocont.2025.103570","DOIUrl":"10.1016/j.jprocont.2025.103570","url":null,"abstract":"<div><div>This work addresses the challenge of computational complexity in Multistage Model Predictive Control (MPC). Multistage MPC is a robust control algorithm where uncertainty is accounted for by constructing a finite set of scenarios for different realizations of uncertainty. The resulting Multistage MPC problem becomes large and computationally expensive to solve. Several strategies have been proposed to reduce computation time and to make real-time implementation possible. A popular class of methods is decomposition methods, in which the Multistage MPC problem is decomposed into several smaller parametric subproblems that can be solved in parallel. The subproblems are repeatedly solved while a coordinator algorithm adjusts a parameter to recover the solution of the full problem. The main computational cost in this approach comes from (1) the coordinator algorithm requiring many iterations to converge, resulting in many resolves of the subproblems, and (2) the computational time to solve the subproblems. In this paper we propose a Predictor–Corrector based path-following algorithm to reduce the solution time of the subproblems for a primal-decomposition algorithm for Multistage MPC. A new Predictor–Corrector methodology based on nonsmooth equation solving is proposed with local superlinear/quadratic convergence. The algorithm path-follows along the parameter path given by the coordinator algorithm. Our path-following algorithm is combined with the Extended Newton algorithm from Bjorvand and Jäschke (2023) for reducing number of coordinator steps. The proposed algorithm is applied to a gas-lift system where the total number of iterations of the algorithm are significantly reduced compared to the standard primal decomposition algorithm for Multistage MPC.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"156 ","pages":"Article 103570"},"PeriodicalIF":3.9,"publicationDate":"2025-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145468044","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

Prior-informed adaptive multi-objective graph reinforcement learning for lysine fed-batch fermentation process 赖氨酸分批补料发酵过程的先验信息自适应多目标图强化学习

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-11-04 DOI: 10.1016/j.jprocont.2025.103578

Dazi Li, Ziqi Yang

The optimization and control of modern industrial processes are increasingly complex, often necessitating fine trade-offs among multiple conflicting objectives. Multi-Objective Reinforcement Learning (MORL) offers a highly promising paradigm for addressing such problems. However, existing MORL methods commonly face challenges in effectively integrating domain prior knowledge and exhibit low data efficiency. To tackle these challenges, a novel MORL framework named Pri-AG is proposed in this paper. Pri-AG utilizes Graph Convolutional Networks (GCNs) to effectively integrate an initial graph topology derived from domain knowledge. Furthermore, a hypernetwork is employed to dynamically adjust graph edge weights according to preferences, thereby enhancing learning efficiency and improving model interpretability. The proposed Pri-AG framework was systematically validated on a typical lysine fed-batch fermentation process simulation. Extensive experimental results demonstrate that Pri-AG enhances sample utilization efficiency, accelerates policy convergence, and surpasses other MORL benchmark algorithms across most performance metrics.

现代工业过程的优化和控制越来越复杂，往往需要在多个相互冲突的目标之间进行精细的权衡。多目标强化学习（MORL）为解决这些问题提供了一个非常有前途的范例。然而，现有的MORL方法在有效整合领域先验知识方面存在挑战，数据效率较低。为了解决这些问题，本文提出了一种新的MORL框架Pri-AG。Pri-AG利用图卷积网络（GCNs）有效地整合由领域知识派生的初始图拓扑。利用超网络根据用户偏好动态调整图边权重，提高学习效率，提高模型可解释性。在一个典型的赖氨酸补料间歇发酵过程模拟中，系统地验证了所提出的Pri-AG框架。大量的实验结果表明，Pri-AG提高了样本利用率，加速了策略收敛，并且在大多数性能指标上超过了其他MORL基准算法。

{"title":"Prior-informed adaptive multi-objective graph reinforcement learning for lysine fed-batch fermentation process","authors":"Dazi Li, Ziqi Yang","doi":"10.1016/j.jprocont.2025.103578","DOIUrl":"10.1016/j.jprocont.2025.103578","url":null,"abstract":"<div><div>The optimization and control of modern industrial processes are increasingly complex, often necessitating fine trade-offs among multiple conflicting objectives. Multi-Objective Reinforcement Learning (MORL) offers a highly promising paradigm for addressing such problems. However, existing MORL methods commonly face challenges in effectively integrating domain prior knowledge and exhibit low data efficiency. To tackle these challenges, a novel MORL framework named Pri-AG is proposed in this paper. Pri-AG utilizes Graph Convolutional Networks (GCNs) to effectively integrate an initial graph topology derived from domain knowledge. Furthermore, a hypernetwork is employed to dynamically adjust graph edge weights according to preferences, thereby enhancing learning efficiency and improving model interpretability. The proposed Pri-AG framework was systematically validated on a typical lysine fed-batch fermentation process simulation. Extensive experimental results demonstrate that Pri-AG enhances sample utilization efficiency, accelerates policy convergence, and surpasses other MORL benchmark algorithms across most performance metrics.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"156 ","pages":"Article 103578"},"PeriodicalIF":3.9,"publicationDate":"2025-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145468043","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

Adaptive ensemble reinforcement learning for industrial process control 工业过程控制中的自适应集成强化学习

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-11-01 DOI: 10.1016/j.jprocont.2025.103575

Soroush Rastegarpour, Hamid Reza Feyzmahdavian, Alf J. Isaksson

Reinforcement learning (RL) has shown significant promise in optimizing control strategies for industrial processes. However, maintaining robustness and stability remains a critical challenge, especially in continuous-time systems affected by uncertainties and nonlinearities. This paper introduces an Adaptive Ensemble RL framework that improves control performance and robustness by dynamically integrating multiple process models. Instead of relying on a fixed model ensemble, we employ a deep neural network to predict optimal weighting factors for combining models based on input-state conditions during training. This adaptive strategy refines the representation of system dynamics, improving robustness against disturbances and model mismatches. We validate the proposed framework using a continuous stirred tank reactor (CSTR) benchmark, showing improved robustness, faster convergence, and reduced performance degradation under varying uncertainties compared to existing methods.

强化学习（RL）在优化工业过程控制策略方面显示出巨大的前景。然而，保持鲁棒性和稳定性仍然是一个关键的挑战，特别是在受不确定性和非线性影响的连续时间系统中。本文介绍了一种自适应集成RL框架，该框架通过动态集成多个过程模型来提高控制性能和鲁棒性。在训练过程中，我们使用深度神经网络来预测基于输入状态条件的组合模型的最优权重因子，而不是依赖于固定的模型集合。这种自适应策略改进了系统动力学的表示，提高了对干扰和模型不匹配的鲁棒性。我们使用连续搅拌槽反应器（CSTR）基准验证了所提出的框架，与现有方法相比，在不同的不确定性下，该框架具有更好的鲁棒性，更快的收敛速度和更少的性能退化。

引用次数: 0

Sensor fault characteristics and fault detection in wastewater treatment plants: Current status and trend analysis 污水处理厂传感器故障特征与故障检测：现状与趋势分析

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-11-01 DOI: 10.1016/j.jprocont.2025.103574

Shanshan Chen , Xiaodong Wang , Xuejun Bi , Zakhar Maletskyi

Sensor faults in wastewater treatment plants (WWTPs) significantly impact the data quality of online monitoring and further affect process operation. The reliability of online sensor data remains the key barrier which obstacles the digitalization of the water sector. Advances in machine learning (ML) and artificial intelligence (AI) offer new opportunities to improve fault detection and diagnosis. Using CiteSpace, this review analyzes literature from 2008 to 2024, highlighting the increasing adoption of hybrid fault detection models that integrate statistical, model-based, and data-driven methods. It categorizes sensor faults, examines their impact on WWTP monitoring, and evaluates mathematical approaches used for fault detection. While AI-driven models enhance detection accuracy, challenges persist in real-time implementation and adaptability to dynamic WWTP conditions. The review further explores strategies for enhancing fault resilience, emphasizing hybrid models, soft sensors, and advanced sensor networks as effective solutions for maintaining system functionality and ensuring continuous monitoring.

污水处理厂传感器故障严重影响在线监测数据质量，进而影响工艺运行。在线传感器数据的可靠性仍然是阻碍水务部门数字化的主要障碍。机器学习（ML）和人工智能（AI）的进步为改进故障检测和诊断提供了新的机会。本文利用CiteSpace分析了2008年至2024年的文献，强调了越来越多地采用混合故障检测模型，该模型集成了统计、基于模型和数据驱动的方法。它对传感器故障进行分类，检查它们对污水处理厂监测的影响，并评估用于故障检测的数学方法。虽然人工智能驱动的模型提高了检测精度，但在实时实施和对动态污水处理厂条件的适应性方面仍然存在挑战。本文进一步探讨了提高故障恢复能力的策略，强调混合模型、软传感器和先进的传感器网络是维持系统功能和确保持续监测的有效解决方案。

{"title":"Sensor fault characteristics and fault detection in wastewater treatment plants: Current status and trend analysis","authors":"Shanshan Chen , Xiaodong Wang , Xuejun Bi , Zakhar Maletskyi","doi":"10.1016/j.jprocont.2025.103574","DOIUrl":"10.1016/j.jprocont.2025.103574","url":null,"abstract":"<div><div>Sensor faults in wastewater treatment plants (WWTPs) significantly impact the data quality of online monitoring and further affect process operation. The reliability of online sensor data remains the key barrier which obstacles the digitalization of the water sector. Advances in machine learning (ML) and artificial intelligence (AI) offer new opportunities to improve fault detection and diagnosis. Using CiteSpace, this review analyzes literature from 2008 to 2024, highlighting the increasing adoption of hybrid fault detection models that integrate statistical, model-based, and data-driven methods. It categorizes sensor faults, examines their impact on WWTP monitoring, and evaluates mathematical approaches used for fault detection. While AI-driven models enhance detection accuracy, challenges persist in real-time implementation and adaptability to dynamic WWTP conditions. The review further explores strategies for enhancing fault resilience, emphasizing hybrid models, soft sensors, and advanced sensor networks as effective solutions for maintaining system functionality and ensuring continuous monitoring.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"155 ","pages":"Article 103574"},"PeriodicalIF":3.9,"publicationDate":"2025-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145416725","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

Dynamic compensation of the threading speed drop in rolling processes: Bayesian optimization of the roughing and finishing mill 轧制过程中螺纹速度下降的动态补偿：粗精轧机的贝叶斯优化

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-10-31 DOI: 10.1016/j.jprocont.2025.103579

Johannes Reinhard , Klaus Löhe , Sebastian Kallabis , Knut Graichen

This paper introduces an advanced approach for dynamic speed drop compensation during threading in rolling processes. The approach combines a data-driven machine learning procedure with a recently presented flatness-based feedforward control to robustly compensate for the speed drop. The feedforward control design accelerates both the rolls and the drivetrain, ensuring that the acceleration torque matches the rolling torque during threading, while maintaining the roll at the desired target speed. Ideally, this prevents the speed drop and enhances the quality and stability of the rolling process. The dynamic speed drop compensation approach is extended in this paper to optimize all stands of a rolling mill, finishing mill as well as roughing mill. To achieve this, the flatness-based feedforward trajectories are adapted to account for uncertainties in the threading event. Moreover, a cost function dependent on optimization parameters is established to optimize the dynamic speed drop compensation. This optimization is carried out using Bayesian Optimization with a Gaussian Process as surrogate model. Both the feedforward control and the Bayesian Optimization run in real-time on an industrial Programmable Logic Controller (PLC). Extensive experimental validation on a hot strip finishing mill, including both the roughing and finishing mill, demonstrates the superior performance of this approach across various key performance indicators in comparison to standard compensation methods.

介绍了一种轧钢螺纹加工过程中动态降速补偿的先进方法。该方法将数据驱动的机器学习过程与最近提出的基于平面度的前馈控制相结合，以鲁棒地补偿速度下降。前馈控制设计使滚道和传动系统同时加速，确保加速扭矩与滚道扭矩匹配，同时使滚道保持在预期的目标速度。理想情况下，这可以防止速度下降，提高轧制过程的质量和稳定性。本文将动态降速补偿方法推广到轧机、精轧机和粗轧机的全机架优化中。为了实现这一点，基于平面度的前馈轨迹适应于考虑线程事件中的不确定性。建立了依赖于优化参数的代价函数，对动态降速补偿进行了优化。该优化是用高斯过程作为代理模型的贝叶斯优化来实现的。前馈控制和贝叶斯优化都在工业可编程控制器（PLC）上实时运行。在热轧带钢精轧机（包括粗轧和精轧机）上进行的大量实验验证表明，与标准补偿方法相比，该方法在各种关键性能指标上具有优越的性能。

{"title":"Dynamic compensation of the threading speed drop in rolling processes: Bayesian optimization of the roughing and finishing mill","authors":"Johannes Reinhard , Klaus Löhe , Sebastian Kallabis , Knut Graichen","doi":"10.1016/j.jprocont.2025.103579","DOIUrl":"10.1016/j.jprocont.2025.103579","url":null,"abstract":"<div><div>This paper introduces an advanced approach for dynamic speed drop compensation during threading in rolling processes. The approach combines a data-driven machine learning procedure with a recently presented flatness-based feedforward control to robustly compensate for the speed drop. The feedforward control design accelerates both the rolls and the drivetrain, ensuring that the acceleration torque matches the rolling torque during threading, while maintaining the roll at the desired target speed. Ideally, this prevents the speed drop and enhances the quality and stability of the rolling process. The dynamic speed drop compensation approach is extended in this paper to optimize all stands of a rolling mill, finishing mill as well as roughing mill. To achieve this, the flatness-based feedforward trajectories are adapted to account for uncertainties in the threading event. Moreover, a cost function dependent on optimization parameters is established to optimize the dynamic speed drop compensation. This optimization is carried out using Bayesian Optimization with a Gaussian Process as surrogate model. Both the feedforward control and the Bayesian Optimization run in real-time on an industrial Programmable Logic Controller (PLC). Extensive experimental validation on a hot strip finishing mill, including both the roughing and finishing mill, demonstrates the superior performance of this approach across various key performance indicators in comparison to standard compensation methods.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"156 ","pages":"Article 103579"},"PeriodicalIF":3.9,"publicationDate":"2025-10-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145420506","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 Bayesian Optimisation with segmentation approach to optimising liquid handling parameters 带分割的贝叶斯优化方法优化液体处理参数

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-10-30 DOI: 10.1016/j.jprocont.2025.103571

Estefania Yap , Viet Huynh , Calvin Vong , Peter Vogel , Viv Louzado , Thomas Barnes , Buser Say , Michael Burke , Dana Kulić , Aldeida Aleti

The automation of liquid handling has become integral in speeding up pharmaceutical development for faster drug development and more affordable treatments. However, the optimal parameters which define the aspirate and dispense procedures vary between liquids and liquid volumes, limiting transfer accuracy and precision. Even state-of-the-art liquid handling devices offer predefined parameters for only a handful of liquids and volumes, resulting in novel parameter sets being defined via a manual, time-consuming process. In this study, we propose an experimental framework for automating the optimisation of liquid class parameters for arbitrary liquids. Within our framework, we propose an optimisation and segmentation algorithm, OptAndSeg, to identify the optimal parameters by automatically grouping volumes into volume ranges and optimising parameters for these volume range subsets. Our method was validated on three live experiments: glycerol, a solution of 25% purified human serum albumin, and human serum. The results showed that OptAndSeg outperformed existing benchmarks for glycerol and human serum. By optimising in non-overlapping volume range segments, we were also able to increase the accuracy and precision of liquid transfer for the 25% purified human serum albumin solution and human serum, achieving relative errors of 5% and 6% or less for volumes as small as 30

μ

L. This methodology can be rapidly applied to any arbitrary liquid, therefore enhancing efficiency and throughput of liquid handling in research and development settings.

液体处理的自动化已经成为加速药物开发的组成部分，以更快的药物开发和更实惠的治疗。然而，定义抽吸和分配程序的最佳参数因液体和液体体积而异，限制了转移的准确性和精度。即使是最先进的液体处理设备也只能为少数液体和体积提供预定义参数，因此需要通过手动、耗时的过程来定义新的参数集。在这项研究中，我们提出了一个实验框架，用于自动优化任意液体的液体类参数。在我们的框架内，我们提出了一种优化和分割算法，OptAndSeg，通过自动将卷分组到卷范围并优化这些卷范围子集的参数来识别最佳参数。我们的方法在三个活体实验中得到验证：甘油，25%纯化的人血清白蛋白溶液和人血清。结果表明，OptAndSeg优于甘油和人血清的现有基准。通过优化非重叠体积范围段，我们还能够提高25%纯化的人血清白蛋白溶液和人血清的液体转移的准确性和精密度，对于小至30 μL的体积，相对误差分别为5%和6%或更小。该方法可以快速应用于任何任意液体，从而提高研究和开发环境中液体处理的效率和吞吐量。

{"title":"A Bayesian Optimisation with segmentation approach to optimising liquid handling parameters","authors":"Estefania Yap , Viet Huynh , Calvin Vong , Peter Vogel , Viv Louzado , Thomas Barnes , Buser Say , Michael Burke , Dana Kulić , Aldeida Aleti","doi":"10.1016/j.jprocont.2025.103571","DOIUrl":"10.1016/j.jprocont.2025.103571","url":null,"abstract":"<div><div>The automation of liquid handling has become integral in speeding up pharmaceutical development for faster drug development and more affordable treatments. However, the optimal parameters which define the aspirate and dispense procedures vary between liquids and liquid volumes, limiting transfer accuracy and precision. Even state-of-the-art liquid handling devices offer predefined parameters for only a handful of liquids and volumes, resulting in novel parameter sets being defined via a manual, time-consuming process. In this study, we propose an experimental framework for automating the optimisation of liquid class parameters for arbitrary liquids. Within our framework, we propose an optimisation and segmentation algorithm, OptAndSeg, to identify the optimal parameters by automatically grouping volumes into volume ranges and optimising parameters for these volume range subsets. Our method was validated on three live experiments: glycerol, a solution of 25% purified human serum albumin, and human serum. The results showed that OptAndSeg outperformed existing benchmarks for glycerol and human serum. By optimising in non-overlapping volume range segments, we were also able to increase the accuracy and precision of liquid transfer for the 25% purified human serum albumin solution and human serum, achieving relative errors of 5% and 6% or less for volumes as small as 30 <span><math><mi>μ</mi></math></span>L. This methodology can be rapidly applied to any arbitrary liquid, therefore enhancing efficiency and throughput of liquid handling in research and development settings.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"156 ","pages":"Article 103571"},"PeriodicalIF":3.9,"publicationDate":"2025-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145420504","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

Implementing quality-by-design latent-variable model predictive control (QbD-LV-MPC) for batch processes: An updating policy for batch profiles 为批处理实现质量设计潜变量模型预测控制（QbD-LV-MPC）：批配置文件的更新策略

IF 3.9 2区计算机科学 Q2 AUTOMATION & CONTROL SYSTEMS

Journal of Process Control

Pub Date : 2025-10-30 DOI: 10.1016/j.jprocont.2025.103576

Qiang Zhu, Zhonggai Zhao, Fei Liu

Ensuring on-spec product quality that satisfies both customer and regulatory requirements is a fundamental objective in batch manufacturing. Over the years, various data-driven strategies have been proposed for batch quality control, involving batch-to-batch and within-batch approaches. While the former is often implemented using offline optimization, maintaining consistent product quality within a batch remains challenging due to unanticipated disturbances that can lead to off-spec products. Existing within-batch strategies, such as latent-variable-based tracking control, mainly address disturbances that affect batch trajectories, potentially overlooking quality-related variations that do not manifest in the trajectory. To address this gap, we proposed a new within-batch control strategy, quality-by-design latent-variable model predictive control (QbD-LV-MPC), which extends the conventional LV-MPC framework. This strategy dynamically updates the reference trajectories within a predefined design space (DS), ensuring all adjustments remain quality-compliant. Two latent variable models, namely principal component analysis and partial least-squares, are calibrated in parallel to construct the LV-MPC and calculate the DS. Upon detecting quality-related disturbances, QbD-LV-MPC promptly adjusts the reference profiles within the DS and computes optimal inputs using LV-MPC. By confining control actions to the DS, the strategy ensures product quality and enhances process flexibility. The proposed strategy has been validated using a benchmark simulator, IndPensim, and the case study results show that it outperforms the conventional LV-MPC in reducing quality deviations.

确保符合规格的产品质量，满足客户和法规要求是批量生产的基本目标。多年来，已经提出了各种数据驱动的批量质量控制策略，包括批对批和批内方法。虽然前者通常使用离线优化实现，但由于意外干扰可能导致产品不合规格，因此在批内保持一致的产品质量仍然具有挑战性。现有的批内策略，如基于潜在变量的跟踪控制，主要处理影响批轨迹的干扰，潜在地忽略了在轨迹中未显示的质量相关变化。为了解决这一差距，我们提出了一种新的批内控制策略，即基于设计的质量潜变量模型预测控制（QbD-LV-MPC），它扩展了传统的LV-MPC框架。该策略在预定义的设计空间（DS）内动态更新参考轨迹，确保所有调整都符合质量要求。平行校准两个潜变量模型，即主成分分析和偏最小二乘，构建LV-MPC并计算DS。在检测到与质量相关的干扰后，QbD-LV-MPC迅速调整DS内的参考轮廓，并使用LV-MPC计算最佳输入。通过将控制行动限制在DS，该策略确保了产品质量并增强了过程灵活性。采用基准模拟器IndPensim对该策略进行了验证，实例研究结果表明，该策略在减少质量偏差方面优于传统的LV-MPC。

{"title":"Implementing quality-by-design latent-variable model predictive control (QbD-LV-MPC) for batch processes: An updating policy for batch profiles","authors":"Qiang Zhu, Zhonggai Zhao, Fei Liu","doi":"10.1016/j.jprocont.2025.103576","DOIUrl":"10.1016/j.jprocont.2025.103576","url":null,"abstract":"<div><div>Ensuring on-spec product quality that satisfies both customer and regulatory requirements is a fundamental objective in batch manufacturing. Over the years, various data-driven strategies have been proposed for batch quality control, involving batch-to-batch and within-batch approaches. While the former is often implemented using offline optimization, maintaining consistent product quality within a batch remains challenging due to unanticipated disturbances that can lead to off-spec products. Existing within-batch strategies, such as latent-variable-based tracking control, mainly address disturbances that affect batch trajectories, potentially overlooking quality-related variations that do not manifest in the trajectory. To address this gap, we proposed a new within-batch control strategy, quality-by-design latent-variable model predictive control (QbD-LV-MPC), which extends the conventional LV-MPC framework. This strategy dynamically updates the reference trajectories within a predefined design space (DS), ensuring all adjustments remain quality-compliant. Two latent variable models, namely principal component analysis and partial least-squares, are calibrated in parallel to construct the LV-MPC and calculate the DS. Upon detecting quality-related disturbances, QbD-LV-MPC promptly adjusts the reference profiles within the DS and computes optimal inputs using LV-MPC. By confining control actions to the DS, the strategy ensures product quality and enhances process flexibility. The proposed strategy has been validated using a benchmark simulator, IndPensim, and the case study results show that it outperforms the conventional LV-MPC in reducing quality deviations.</div></div>","PeriodicalId":50079,"journal":{"name":"Journal of Process Control","volume":"156 ","pages":"Article 103576"},"PeriodicalIF":3.9,"publicationDate":"2025-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145420503","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