Computers & Chemical Engineering最新文献

Digital twin framework with physics-informed neural networks for real-time monitoring of PEM electrolyzers in renewable microgrids 用于可再生微电网PEM电解槽实时监测的物理信息神经网络数字孪生框架

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-25 DOI: 10.1016/j.compchemeng.2026.109582

Hassan NAANANI, Meriem KAYSOUNY, Anas ABERHOUCH, Said SAIR, Abdessamad FAIK

The operation of electrolyzers for producing green hydrogen faces two key challenges: the rapidly increasing demand for hydrogen across diverse applications and the limited durability of electrochemical components. Digital twin (DT) technology offers a promising pathway to address these limitations by enabling real-time monitoring, fault detection, and predictive analysis. This study presents the development of a DT for a laboratory-scale proton exchange membrane (PEM) electrolyzer composed of two series-connected cells. The virtual counterpart synchronizes with experimental voltage and current data acquired under varying operating conditions, enabling continuous verification of the system’s behavior. To enhance predictive capability, a physicsinformed neural networks (PINN) is integrated into the DT, combining polarization data with electrochemical constraints, including Butler-Volmer activation kinetics and ohmic resistance. A complementary 3D representation of the developed system provides an interactive visualization of the electrolyzer and its operating state. The resulting framework supports real-time supervision, performance assessment, and degradation monitoring, offering a practical foundation for intelligent control and diagnostic strategies in PEM electrolysis systems.

生产绿色氢的电解槽的运行面临两个关键挑战：各种应用对氢的需求快速增长以及电化学元件的有限耐用性。数字孪生（DT）技术通过实现实时监控、故障检测和预测分析，为解决这些限制提供了一条很有前途的途径。本研究介绍了一个由两个串联电池组成的实验室规模质子交换膜（PEM）电解槽的DT的发展。虚拟对应体与在不同操作条件下获得的实验电压和电流数据同步，从而能够连续验证系统的行为。为了提高预测能力，将物理信息神经网络（PINN）集成到DT中，将极化数据与电化学约束（包括Butler-Volmer活化动力学和欧姆电阻）相结合。所开发系统的补充3D表示提供了电解槽及其运行状态的交互式可视化。由此产生的框架支持实时监督、性能评估和退化监测，为PEM电解系统的智能控制和诊断策略提供了实用基础。

{"title":"Digital twin framework with physics-informed neural networks for real-time monitoring of PEM electrolyzers in renewable microgrids","authors":"Hassan NAANANI, Meriem KAYSOUNY, Anas ABERHOUCH, Said SAIR, Abdessamad FAIK","doi":"10.1016/j.compchemeng.2026.109582","DOIUrl":"10.1016/j.compchemeng.2026.109582","url":null,"abstract":"<div><div>The operation of electrolyzers for producing green hydrogen faces two key challenges: the rapidly increasing demand for hydrogen across diverse applications and the limited durability of electrochemical components. Digital twin (DT) technology offers a promising pathway to address these limitations by enabling real-time monitoring, fault detection, and predictive analysis. This study presents the development of a DT for a laboratory-scale proton exchange membrane (PEM) electrolyzer composed of two series-connected cells. The virtual counterpart synchronizes with experimental voltage and current data acquired under varying operating conditions, enabling continuous verification of the system’s behavior. To enhance predictive capability, a physicsinformed neural networks (PINN) is integrated into the DT, combining polarization data with electrochemical constraints, including Butler-Volmer activation kinetics and ohmic resistance. A complementary 3D representation of the developed system provides an interactive visualization of the electrolyzer and its operating state. The resulting framework supports real-time supervision, performance assessment, and degradation monitoring, offering a practical foundation for intelligent control and diagnostic strategies in PEM electrolysis systems.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"208 ","pages":"Article 109582"},"PeriodicalIF":3.9,"publicationDate":"2026-01-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146075819","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Kolmogorov–Arnold network-assisted multi-objective approach for design and optimization of unbaffled stirred tanks 无挡板搅拌槽设计与优化的Kolmogorov-Arnold网络辅助多目标方法

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-25 DOI: 10.1016/j.compchemeng.2026.109583

Wentao Du , Tingting Liu , Zicheng Meng , Muhammad Waqas Yaqub , Baofeng Wang , Xizhong Chen

Stirred tanks are extensively employed in various industries to realize efficient mixing processes. In this work, an optimization framework integrating the Kolmogorov–Arnold network (KAN) surrogate model and the non-dominated sorting genetic algorithm (NSGA-II) is developed for the geometric design of unbaffled stirred-tank impellers. Three-dimensional computational fluid dynamics (CFD) simulations were conducted over broad parametric ranges of impeller blade length, width, number, and tilt angle to produce datasets of the flow fields. These datasets were subsequently employed to train the KAN surrogate model, enabling rapid and accurate prediction of the three-dimensional flow fields. The root-mean-square (RMS) of static pressure and mixing intensity (MI) were calculated from the surrogate-predicted data and served as dual objective functions for NSGA-II optimization. The optimal impeller geometry identified by the KAN–NSGA-II framework was further validated, revealing a significant reduction in RMS pressure and an enhancement in MI relative to the baseline design. The result shows that combining data-driven surrogate modeling with evolutionary optimization provides a robust and efficient strategy for the performance-driven geometric optimization of industrial mixing equipment.

搅拌槽广泛应用于各个行业，以实现高效的混合过程。在这项工作中，结合Kolmogorov-Arnold网络（KAN）代理模型和非支配排序遗传算法（NSGA-II）开发了一个优化框架，用于无挡板搅拌槽叶轮的几何设计。在较宽的叶轮叶片长度、宽度、叶片数和叶片倾角等参数范围内进行了三维计算流体力学（CFD）仿真，得到了流场数据集。随后利用这些数据集训练KAN代理模型，实现了对三维流场的快速准确预测。根据替代预测数据计算静压均方根（RMS）和混合强度（MI），并作为NSGA-II优化的双目标函数。KAN-NSGA-II框架确定的最佳叶轮几何形状得到了进一步验证，结果显示相对于基线设计，RMS压力显著降低，MI增强。结果表明，将数据驱动的代理建模与进化优化相结合，为性能驱动的工业混合设备几何优化提供了一种鲁棒高效的策略。

{"title":"Kolmogorov–Arnold network-assisted multi-objective approach for design and optimization of unbaffled stirred tanks","authors":"Wentao Du , Tingting Liu , Zicheng Meng , Muhammad Waqas Yaqub , Baofeng Wang , Xizhong Chen","doi":"10.1016/j.compchemeng.2026.109583","DOIUrl":"10.1016/j.compchemeng.2026.109583","url":null,"abstract":"<div><div>Stirred tanks are extensively employed in various industries to realize efficient mixing processes. In this work, an optimization framework integrating the Kolmogorov–Arnold network (KAN) surrogate model and the non-dominated sorting genetic algorithm (NSGA-II) is developed for the geometric design of unbaffled stirred-tank impellers. Three-dimensional computational fluid dynamics (CFD) simulations were conducted over broad parametric ranges of impeller blade length, width, number, and tilt angle to produce datasets of the flow fields. These datasets were subsequently employed to train the KAN surrogate model, enabling rapid and accurate prediction of the three-dimensional flow fields. The root-mean-square (RMS) of static pressure and mixing intensity (MI) were calculated from the surrogate-predicted data and served as dual objective functions for NSGA-II optimization. The optimal impeller geometry identified by the KAN–NSGA-II framework was further validated, revealing a significant reduction in RMS pressure and an enhancement in MI relative to the baseline design. The result shows that combining data-driven surrogate modeling with evolutionary optimization provides a robust and efficient strategy for the performance-driven geometric optimization of industrial mixing equipment.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"208 ","pages":"Article 109583"},"PeriodicalIF":3.9,"publicationDate":"2026-01-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146075821","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Modelling and optimization of extractive distillation for IPA-water separation using artificial neural network models 基于人工神经网络模型的ipa -水萃取精馏建模与优化

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-22 DOI: 10.1016/j.compchemeng.2026.109580

Ziba Valizadeh, Hanieh Shokrkar

The extraction of isopropyl alcohol (IPA) from water has long presented a considerable challenge in the chemical industry, mainly due to the formation of an azeotropic mixture and the close boiling points of the two substances. This research introduces a hybrid extractive distillation method utilizing dimethyl sulfoxide (DMSO) as an entrainer, which was simulated using Aspen Plus V11 (NRTL model). The comprehensive simulation attained an impressive IPA purity of 99.9%, which resulted in considerable energy costs. In order to tackle the balance between purity and energy consumption, a novel hybrid framework was created, and essential operational parameters such as feed stage, reflux ratio, an entrainer ratio, condenser duty, and reboiler duty were fine-tuned to enhance IPA purity while reducing energy consumption. The advancement is found in substituting conventional simulation methods with a trained artificial neural network (ANN) to enable rapid predictions, alongside the use of particle swarm optimization (PSO) for fine-tuning parameters. The ANN-PSO framework successfully pinpointed an optimal operating point that led to a 25% decrease in total energy consumption when compared to the baseline. While achieving a practically acceptable IPA purity of 95% in the first column and a water purity of 95% in the second column at reflux ratios of 0.77 and 0.4, respectively, this method demonstrates a notable decrease in computational effort (40% less time) and offers a reliable strategy for low-energy separation processes at an industrial level.

长期以来，从水中提取异丙醇（IPA）一直是化学工业面临的一个相当大的挑战，主要是由于形成共沸混合物和两种物质的沸点接近。本文介绍了一种以二甲亚砜（DMSO）为夹带剂的混合萃取精馏方法，并采用Aspen Plus V11 （NRTL模型）进行了模拟。综合模拟获得了令人印象深刻的99.9%的IPA纯度，这导致了相当大的能源成本。为了解决纯度和能耗之间的平衡，创建了一个新的混合框架，并对进料级、回流比、夹带器比、冷凝器负荷和再锅炉负荷等基本操作参数进行了调整，以提高IPA纯度，同时降低能耗。这一进步是用训练有素的人工神经网络（ANN）代替传统的模拟方法来实现快速预测，同时使用粒子群优化（PSO）来微调参数。ANN-PSO框架成功地确定了一个最佳工作点，与基线相比，总能耗降低了25%。虽然在回流比分别为0.77和0.4的情况下，第一柱的IPA纯度为95%，第二柱的水纯度为95%，但该方法可以显著减少计算工作量（减少40%的时间），并为工业水平的低能量分离过程提供了可靠的策略。

{"title":"Modelling and optimization of extractive distillation for IPA-water separation using artificial neural network models","authors":"Ziba Valizadeh, Hanieh Shokrkar","doi":"10.1016/j.compchemeng.2026.109580","DOIUrl":"10.1016/j.compchemeng.2026.109580","url":null,"abstract":"<div><div>The extraction of isopropyl alcohol (IPA) from water has long presented a considerable challenge in the chemical industry, mainly due to the formation of an azeotropic mixture and the close boiling points of the two substances. This research introduces a hybrid extractive distillation method utilizing dimethyl sulfoxide (DMSO) as an entrainer, which was simulated using Aspen Plus V11 (NRTL model). The comprehensive simulation attained an impressive IPA purity of 99.9%, which resulted in considerable energy costs. In order to tackle the balance between purity and energy consumption, a novel hybrid framework was created, and essential operational parameters such as feed stage, reflux ratio, an entrainer ratio, condenser duty, and reboiler duty were fine-tuned to enhance IPA purity while reducing energy consumption. The advancement is found in substituting conventional simulation methods with a trained artificial neural network (ANN) to enable rapid predictions, alongside the use of particle swarm optimization (PSO) for fine-tuning parameters. The ANN-PSO framework successfully pinpointed an optimal operating point that led to a 25% decrease in total energy consumption when compared to the baseline. While achieving a practically acceptable IPA purity of 95% in the first column and a water purity of 95% in the second column at reflux ratios of 0.77 and 0.4, respectively, this method demonstrates a notable decrease in computational effort (40% less time) and offers a reliable strategy for low-energy separation processes at an industrial level.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"208 ","pages":"Article 109580"},"PeriodicalIF":3.9,"publicationDate":"2026-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146075820","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Hybrid approach for comprehensive recognition of line objects contained in high-density piping and instrumentation diagrams using deep learning and rules 基于深度学习和规则的高密度管道和仪表图中线对象综合识别的混合方法

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-20 DOI: 10.1016/j.compchemeng.2026.109572

Yoochan Moon , Seung-Tae Han , Ji-Beob Kim , Choongsub Yeom , Duhwan Mun

This study presents a hybrid approach for the automated recognition and classification of line objects in piping and instrumentation diagrams (P&IDs), with the goal of supporting the digital transformation of chemical process design and operation. By integrating Deep Learning (DL) techniques with rule-based methods, the proposed approach extracts flow and signal paths from legacy P&ID images, enabling applications such as process simulation, safety verification, and control logic validation. The approach consists of two stages. In the first stage, all line objects in a P&ID are detected and categorized into lines with special signs and continuous lines. A DL model identifies directional arrows and determines the overall flow structure. In the second stage, the continuous lines are further classified into dimension, extension, and leader lines using the rule-based algorithms, according to their functional characteristics. The method was tested on 30 P&ID sheets from Project A and two from Project B. Initially, the model trained on Project A data achieved precision and recall rates of 95.02% and 93.09%, respectively. On Project B, the performance dropped to 88.92% and 84.76% due to domain shift. After applying transfer learning using the four additional Project B sheets, the performance improved to 95.32% precision and 91.55% recall. These results demonstrate the potential of the proposed approach for accurate and scalable conversion of P&ID data into structured formats, contributing to smart plant design and engineering data integration.

本研究提出了一种混合方法，用于管道和仪表图（P&IDs）中线对象的自动识别和分类，目的是支持化学过程设计和操作的数字化转型。通过将深度学习（DL）技术与基于规则的方法相结合，该方法可以从遗留的P&；ID图像中提取流程和信号路径，从而实现过程仿真、安全验证和控制逻辑验证等应用。该方法包括两个阶段。在第一阶段，检测P&；ID中的所有线对象，并将其分类为具有特殊符号的线和连续线。DL模型识别方向箭头并确定整体流结构。第二阶段，根据连续线的功能特征，采用基于规则的算法将连续线进一步划分为维线、延长线和先导线。该方法在项目A的30张P&；ID表和项目b的2张P&；ID表上进行了测试。最初，在项目A数据上训练的模型的准确率和召回率分别达到95.02%和93.09%。在项目B中，由于域移位，性能下降到88.92%和84.76%。在使用额外的四张B项目表应用迁移学习后，性能提高到95.32%的准确率和91.55%的召回率。这些结果证明了所提出的方法将P&；ID数据精确且可扩展地转换为结构化格式的潜力，有助于智能工厂设计和工程数据集成。

{"title":"Hybrid approach for comprehensive recognition of line objects contained in high-density piping and instrumentation diagrams using deep learning and rules","authors":"Yoochan Moon , Seung-Tae Han , Ji-Beob Kim , Choongsub Yeom , Duhwan Mun","doi":"10.1016/j.compchemeng.2026.109572","DOIUrl":"10.1016/j.compchemeng.2026.109572","url":null,"abstract":"<div><div>This study presents a hybrid approach for the automated recognition and classification of line objects in piping and instrumentation diagrams (P&IDs), with the goal of supporting the digital transformation of chemical process design and operation. By integrating Deep Learning (DL) techniques with rule-based methods, the proposed approach extracts flow and signal paths from legacy P&ID images, enabling applications such as process simulation, safety verification, and control logic validation. The approach consists of two stages. In the first stage, all line objects in a P&ID are detected and categorized into lines with special signs and continuous lines. A DL model identifies directional arrows and determines the overall flow structure. In the second stage, the continuous lines are further classified into dimension, extension, and leader lines using the rule-based algorithms, according to their functional characteristics. The method was tested on 30 P&ID sheets from Project A and two from Project B. Initially, the model trained on Project A data achieved precision and recall rates of 95.02% and 93.09%, respectively. On Project B, the performance dropped to 88.92% and 84.76% due to domain shift. After applying transfer learning using the four additional Project B sheets, the performance improved to 95.32% precision and 91.55% recall. These results demonstrate the potential of the proposed approach for accurate and scalable conversion of P&ID data into structured formats, contributing to smart plant design and engineering data integration.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"208 ","pages":"Article 109572"},"PeriodicalIF":3.9,"publicationDate":"2026-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146075818","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Online alarm flood classification via interpretable template extraction and structured convolutional matching 基于可解释模板提取和结构化卷积匹配的洪水报警在线分类

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-19 DOI: 10.1016/j.compchemeng.2026.109570

Yashar Rahimi , Harikrishna Rao Mohan Rao , Jing Zhou , Tongwen Chen

Alarm flood classification in industrial alarm systems is a challenging task due to variability in fault durations, process noise, and the volume of overlapping alarms. However, alarm floods triggered by similar faults often exhibit recurring structural patterns, which, if identified effectively, can support the root cause diagnosis and informed decision-making by operators. Existing classification methods often rely on opaque models, extensive retraining, or lack integration with operator-facing tools. Motivated by this practical problem, a unified visual analytics-based methodology for the real-time classification of alarm floods is proposed in this paper. The contributions are threefold: (1) The existing High-Density Alarm Plot (HDAP) is extended into a structured matrix representation to encode alarm activity over time; (2) a 2D convolution-based alignment technique is developed to extract representative templates from historical alarm floods, enabling category-specific pattern generation; and (3) a dynamic matrix representation is introduced to support real-time alarm monitoring, where similarity matching against pre-learned templates facilitates online classification with minimal delay. The proposed method is interpretable, operator-friendly, and seamlessly integrates with existing visual tools. The effectiveness of the proposed method is validated on the Tennessee Eastman Process benchmark, demonstrating robust and accurate early-stage classification.

由于故障持续时间、过程噪声和重叠报警量的变化，工业报警系统中的报警洪水分类是一项具有挑战性的任务。然而，由类似故障触发的报警洪水通常表现出重复的结构模式，如果有效识别，可以支持根本原因诊断和运营商的明智决策。现有的分类方法通常依赖于不透明的模型，大量的再培训，或者缺乏与面向操作人员的工具的集成。针对这一实际问题，本文提出了一种统一的基于可视化分析的洪水报警实时分类方法。贡献有三个方面：(1)将现有的高密度报警图（HDAP）扩展为结构化矩阵表示，以编码随时间变化的报警活动；(2)开发了一种基于二维卷积的对齐技术，从历史报警洪水中提取代表性模板，实现了特定类别模式的生成；(3)引入动态矩阵表示来支持实时报警监控，其中针对预学习模板的相似度匹配有助于以最小的延迟在线分类。该方法具有可解释性、操作友好性和与现有可视化工具无缝集成的特点。在田纳西州伊士曼过程基准上验证了该方法的有效性，证明了早期分类的鲁棒性和准确性。

{"title":"Online alarm flood classification via interpretable template extraction and structured convolutional matching","authors":"Yashar Rahimi , Harikrishna Rao Mohan Rao , Jing Zhou , Tongwen Chen","doi":"10.1016/j.compchemeng.2026.109570","DOIUrl":"10.1016/j.compchemeng.2026.109570","url":null,"abstract":"<div><div>Alarm flood classification in industrial alarm systems is a challenging task due to variability in fault durations, process noise, and the volume of overlapping alarms. However, alarm floods triggered by similar faults often exhibit recurring structural patterns, which, if identified effectively, can support the root cause diagnosis and informed decision-making by operators. Existing classification methods often rely on opaque models, extensive retraining, or lack integration with operator-facing tools. Motivated by this practical problem, a unified visual analytics-based methodology for the real-time classification of alarm floods is proposed in this paper. The contributions are threefold: (1) The existing High-Density Alarm Plot (HDAP) is extended into a structured matrix representation to encode alarm activity over time; (2) a 2D convolution-based alignment technique is developed to extract representative templates from historical alarm floods, enabling category-specific pattern generation; and (3) a dynamic matrix representation is introduced to support real-time alarm monitoring, where similarity matching against pre-learned templates facilitates online classification with minimal delay. The proposed method is interpretable, operator-friendly, and seamlessly integrates with existing visual tools. The effectiveness of the proposed method is validated on the Tennessee Eastman Process benchmark, demonstrating robust and accurate early-stage classification.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"208 ","pages":"Article 109570"},"PeriodicalIF":3.9,"publicationDate":"2026-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146025889","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Bayesian optimization and temporal attention-enhanced deep neural network for accurate and reliable state of health estimation of lithium-ion batteries 基于贝叶斯优化和时间注意力增强的深度神经网络的锂离子电池健康状态准确可靠估计

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-19 DOI: 10.1016/j.compchemeng.2026.109571

Zhiyu Chen , Hanfei Wang , Siquan Li , Kena Chen , Jinjie Wang , Ping Wang , Lijun Yang

The capacity degradation trend can indirectly reflect the health status of a battery, and accurate state of health (SOH) estimation can reduce the risk of failure and ensure stable operation. However, the limited feature learning capability of traditional models and the random combination of hyperparameters often lead to large estimation errors. To address the issues, this research proposes a deep neural network (DNN) enhanced by Bayesian optimization (BO) and the temporal attention (TA) mechanism to achieve accurate and reliable SOH estimation for lithium-ion batteries. First, direct aging features of the battery are extracted based on the constant-current charging phase and further fused using principal component analysis (PCA). Then, a mapping model between the aging features and capacity is constructed, in which the TA mechanism is employed to enhance the feature learning capability of the DNN, and BO is used to determine the optimal combination of key hyperparameters. Finally, twelve single-battery under different operating conditions and seven multi-battery capacity estimation experiments are conducted. The estimation performance of the proposed model is evaluated using metrics such as mean absolute error (MAE). The experimental results show that the BO-TADNN model achieves capacity estimation errors within ±3% for single-battery experiments, representing an improvement of approximately 70% in stability compared to the DNN. Furthermore, BO-TADNN achieves the best performance across all evaluation metrics in the multi-battery experiments, which provides a theoretical foundation for future applications in battery management systems.

容量退化趋势可以间接反映电池的健康状态，准确的健康状态（SOH）估计可以降低电池故障的风险，保证电池的稳定运行。然而，传统模型有限的特征学习能力和超参数的随机组合往往导致估计误差较大。为了解决这一问题，本研究提出了一种基于贝叶斯优化（BO）和时间注意（TA）机制的深度神经网络（DNN）来实现锂离子电池SOH的准确可靠估计。首先，基于恒流充电阶段提取电池的直接老化特征，并利用主成分分析（PCA）进行融合；然后，构建了老化特征与容量的映射模型，其中利用TA机制增强深度神经网络的特征学习能力，利用BO确定关键超参数的最优组合。最后进行了12个不同工况下的单电池和7个多电池容量估算实验。使用平均绝对误差（MAE）等度量来评估所提出模型的估计性能。实验结果表明，BO-TADNN模型在单电池实验中实现了±3%的容量估计误差，与DNN相比，稳定性提高了约70%。此外，在多电池实验中，BO-TADNN在所有评估指标中都取得了最佳性能，为未来在电池管理系统中的应用提供了理论基础。

{"title":"Bayesian optimization and temporal attention-enhanced deep neural network for accurate and reliable state of health estimation of lithium-ion batteries","authors":"Zhiyu Chen , Hanfei Wang , Siquan Li , Kena Chen , Jinjie Wang , Ping Wang , Lijun Yang","doi":"10.1016/j.compchemeng.2026.109571","DOIUrl":"10.1016/j.compchemeng.2026.109571","url":null,"abstract":"<div><div>The capacity degradation trend can indirectly reflect the health status of a battery, and accurate state of health (SOH) estimation can reduce the risk of failure and ensure stable operation. However, the limited feature learning capability of traditional models and the random combination of hyperparameters often lead to large estimation errors. To address the issues, this research proposes a deep neural network (DNN) enhanced by Bayesian optimization (BO) and the temporal attention (TA) mechanism to achieve accurate and reliable SOH estimation for lithium-ion batteries. First, direct aging features of the battery are extracted based on the constant-current charging phase and further fused using principal component analysis (PCA). Then, a mapping model between the aging features and capacity is constructed, in which the TA mechanism is employed to enhance the feature learning capability of the DNN, and BO is used to determine the optimal combination of key hyperparameters. Finally, twelve single-battery under different operating conditions and seven multi-battery capacity estimation experiments are conducted. The estimation performance of the proposed model is evaluated using metrics such as mean absolute error (MAE). The experimental results show that the BO-TADNN model achieves capacity estimation errors within ±3% for single-battery experiments, representing an improvement of approximately 70% in stability compared to the DNN. Furthermore, BO-TADNN achieves the best performance across all evaluation metrics in the multi-battery experiments, which provides a theoretical foundation for future applications in battery management systems.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"208 ","pages":"Article 109571"},"PeriodicalIF":3.9,"publicationDate":"2026-01-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146006595","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Corrigendum to Hybrid Modelling of Chemical Processes: A Unified Framework Based on Deductive, Inductive, and Abductive Inference 化学过程混合模型的勘误表：基于演绎、归纳和溯因推理的统一框架

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-17 DOI: 10.1016/j.compchemeng.2026.109549

Raymoon Hwang , Jae Hyun Cho , Il Moon , Min Oh

引用次数: 0

Variable-horizon economic MPC for cyclic industrial air dryers using hybrid models and state estimation 基于混合模型和状态估计的循环工业空气干燥机变视距经济MPC

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-16 DOI: 10.1016/j.compchemeng.2026.109569

Sida Chai , Ece Serenat Köksal , Xiangyin Kong , Winston S.K. Tang , Erdal Aydın , Mehmet Mercangöz

This paper introduces a variable horizon economic model predictive control (EMPC) framework for a twin bed industrial desiccant air drying plant. Hybrid mechanistic and machine learning models are employed to simulate the drying and regeneration processes, providing a realistic representation of system dynamics. A moving horizon state estimation framework, integrated with hybrid models, is utilized to estimate the adsorbed water content in the beds. Based on these estimated values, an algorithm is implemented to estimate the end time of the regeneration process. The EMPC framework uses this end time as the prediction horizon to optimize the manipulated variable trajectories for the drying process. Simulation results show that the proposed EMPC reduces cooling-energy consumption by increasing the average temperature of the inlet wet air by approximately 2°C. At the same time, it improves system performance by increasing the moisture adsorbed in the bed by approximately

6 – 10 %

. Under these new operating conditions, the overall energy consumption is estimated to decrease by about 6.5%, thereby enhancing process profitability.

介绍了一种双床工业干燥剂风干装置的变水平经济模型预测控制（EMPC）框架。采用混合机械和机器学习模型来模拟干燥和再生过程，提供了系统动力学的真实表示。采用结合混合模型的移动层位状态估计框架对床层中吸附水分进行了估计。基于这些估计值，实现了一种算法来估计再生过程的结束时间。EMPC框架使用该结束时间作为预测范围来优化干燥过程的可操纵变量轨迹。仿真结果表明，该方法可使入口湿空气平均温度提高约2℃，从而降低冷却能耗。同时，它通过增加床层中吸附的水分约6-10%来改善系统性能。在这些新的操作条件下，预计总能耗将降低约6.5%，从而提高工艺盈利能力。

{"title":"Variable-horizon economic MPC for cyclic industrial air dryers using hybrid models and state estimation","authors":"Sida Chai , Ece Serenat Köksal , Xiangyin Kong , Winston S.K. Tang , Erdal Aydın , Mehmet Mercangöz","doi":"10.1016/j.compchemeng.2026.109569","DOIUrl":"10.1016/j.compchemeng.2026.109569","url":null,"abstract":"<div><div>This paper introduces a variable horizon economic model predictive control (EMPC) framework for a twin bed industrial desiccant air drying plant. Hybrid mechanistic and machine learning models are employed to simulate the drying and regeneration processes, providing a realistic representation of system dynamics. A moving horizon state estimation framework, integrated with hybrid models, is utilized to estimate the adsorbed water content in the beds. Based on these estimated values, an algorithm is implemented to estimate the end time of the regeneration process. The EMPC framework uses this end time as the prediction horizon to optimize the manipulated variable trajectories for the drying process. Simulation results show that the proposed EMPC reduces cooling-energy consumption by increasing the average temperature of the inlet wet air by approximately 2°C. At the same time, it improves system performance by increasing the moisture adsorbed in the bed by approximately <span><math><mrow><mn>6</mn><mtext>–</mtext><mn>10</mn><mtext>%</mtext></mrow></math></span>. Under these new operating conditions, the overall energy consumption is estimated to decrease by about 6.5%, thereby enhancing process profitability.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"207 ","pages":"Article 109569"},"PeriodicalIF":3.9,"publicationDate":"2026-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146034929","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

HOFLON: Hybrid Offline Learning and Online Optimization for process start-up and grade-transition control HOFLON：过程启动和等级转换控制的混合离线学习和在线优化

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-16 DOI: 10.1016/j.compchemeng.2026.109566

Alex Durkin , Jasper Stolte , Mehmet Mercangöz

Start-ups and product grade-changes are critical steps in continuous-process plant operation, because any misstep immediately affects product quality and drives operational losses. These transitions have long relied on supervision by a handful of expert operators, but the progressive retirement of that workforce is leaving plant owners without the tacit know-how needed to execute them consistently. In the absence of a process model, offline reinforcement learning (RL) promises to capture — and even surpass — human expertise by mining historical start-up and grade-change logs, yet standard offline RL struggles with distribution-shift and value-overestimation whenever a learned policy ventures outside the data envelope. We introduce HOFLON (Hybrid Offline Learning ＋ Online Optimization) to overcome those limitations. Offline, HOFLON learns (i) a latent data manifold that represents the feasible region spanned by past transitions and (ii) a long-horizon Q-critic that predicts the cumulative reward from state–action pairs. Online, it solves a one-step optimization problem that maximizes the Q-critic while penalizing deviations from the learned manifold and excessive rates of change in the manipulated variables. We test HOFLON on two industrial case studies—a polymerization reactor start-up and a paper-machine grade-change problem—and benchmark it against Implicit Q-Learning (IQL), a leading offline-RL algorithm. In both plants HOFLON not only surpasses IQL but also delivers on average better cumulative rewards compared to the best start-up or grade-change ever observed in the historical data, demonstrating its potential to automate transition operations beyond current expert capability.

启动和产品等级变化是连续工艺工厂操作的关键步骤，因为任何失误都会立即影响产品质量并导致运营损失。长期以来，这些转型依赖于少数专业操作人员的监督，但这些员工的逐步退休，使工厂所有者缺乏持续执行这些转型所需的隐性知识。在没有过程模型的情况下，离线强化学习（RL）有望通过挖掘历史启动和等级变化日志来捕获甚至超越人类的专业知识，然而，每当学习到的策略冒险超出数据范围时，标准的离线强化学习就会与分布转移和价值高估作斗争。我们引入HOFLON（混合式离线学习+在线优化）来克服这些限制。离线时，HOFLON学习(i)一个潜在的数据流形，它表示由过去的过渡跨越的可行区域；（ii）一个长视界Q-critic，它预测状态-行为对的累积奖励。在线，它解决了一个一步优化问题，最大限度地提高了Q-critic，同时惩罚了从学习流形的偏差和被操纵变量的过度变化率。我们在两个工业案例研究中测试了HOFLON——聚合反应器启动和纸机等级变化问题——并将其与隐式q -学习（IQL）（一种领先的离线rl算法）进行了基准测试。在这两个工厂中，HOFLON不仅超过了IQL，而且与历史数据中观察到的最佳启动或等级变更相比，平均累积回报更高，证明了其超越当前专家能力的自动化过渡操作的潜力。

{"title":"HOFLON: Hybrid Offline Learning and Online Optimization for process start-up and grade-transition control","authors":"Alex Durkin , Jasper Stolte , Mehmet Mercangöz","doi":"10.1016/j.compchemeng.2026.109566","DOIUrl":"10.1016/j.compchemeng.2026.109566","url":null,"abstract":"<div><div>Start-ups and product grade-changes are critical steps in continuous-process plant operation, because any misstep immediately affects product quality and drives operational losses. These transitions have long relied on supervision by a handful of expert operators, but the progressive retirement of that workforce is leaving plant owners without the tacit know-how needed to execute them consistently. In the absence of a process model, offline reinforcement learning (RL) promises to capture — and even surpass — human expertise by mining historical start-up and grade-change logs, yet standard offline RL struggles with distribution-shift and value-overestimation whenever a learned policy ventures outside the data envelope. We introduce HOFLON (Hybrid Offline Learning ＋ Online Optimization) to overcome those limitations. Offline, HOFLON learns (i) a latent data manifold that represents the feasible region spanned by past transitions and (ii) a long-horizon Q-critic that predicts the cumulative reward from state–action pairs. Online, it solves a one-step optimization problem that maximizes the Q-critic while penalizing deviations from the learned manifold and excessive rates of change in the manipulated variables. We test HOFLON on two industrial case studies—a polymerization reactor start-up and a paper-machine grade-change problem—and benchmark it against Implicit Q-Learning (IQL), a leading offline-RL algorithm. In both plants HOFLON not only surpasses IQL but also delivers on average better cumulative rewards compared to the best start-up or grade-change ever observed in the historical data, demonstrating its potential to automate transition operations beyond current expert capability.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"207 ","pages":"Article 109566"},"PeriodicalIF":3.9,"publicationDate":"2026-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145974181","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Big data-driven predictive control for nonlinear systems based on kernel density estimation of data trajectories 基于数据轨迹核密度估计的非线性系统大数据驱动预测控制

IF 3.9 2区工程技术 Q2 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Chemical Engineering

Pub Date : 2026-01-15 DOI: 10.1016/j.compchemeng.2026.109565

Shuangyu Han , Yitao Yan , Jie Bao , Biao Huang

A big data-driven predictive control approach for nonlinear systems is proposed based on the kernel density estimation of data trajectories (KDE-BDPC) in the behavioural systems framework, which aims to control the nonlinear process in the regions where only limited data are available. The nonlinear process behaviour (a set of input–output variable trajectories) can be partitioned into linear sub-behaviours (trajectory clusters) offline via multi-view clustering of collected data trajectories. To operate the nonlinear process behaviour outside the existing linear sub-behaviours, we propose a data-driven system behaviour approximation approach that can interpolate linear sub-behaviours based on the density estimation of existing data trajectories and linear subspace distance. Based on online interpolated linear sub-behaviours, an online big data-driven predictive controller is designed, which includes a path search to minimise uncertainty. The proposed approach is illustrated by a vanadium flow battery control problem.

提出了一种基于行为系统框架中数据轨迹核密度估计（KDE-BDPC）的非线性系统大数据驱动预测控制方法，旨在控制数据有限区域的非线性过程。通过对收集到的数据轨迹进行多视图聚类，可以将非线性过程行为（一组输入输出变量轨迹）离线划分为线性子行为（轨迹簇）。为了在现有的线性子行为之外操作非线性过程行为，我们提出了一种数据驱动的系统行为近似方法，该方法可以基于现有数据轨迹和线性子空间距离的密度估计来插值线性子行为。基于在线插值线性子行为，设计了一种在线大数据驱动的预测控制器，其中包括最小化不确定性的路径搜索。以钒液流电池的控制问题为例说明了该方法的可行性。

引用次数: 0