Computers & Chemical Engineering最新文献

{"title":"Leveraging machine learning algorithms to predict life cycle inventory assessments (LCIA) to facilitate sustainable process design","authors":"Emmanuel Aboagye ,&nbsp;Jared Longo ,&nbsp;Matt Conway ,&nbsp;John Pazik ,&nbsp;Milo Barkow ,&nbsp;Marcella McMahon ,&nbsp;Brendan Weil ,&nbsp;Harriet Dufie Appiah ,&nbsp;Robert Hesketh ,&nbsp;Kirti M. Yenkie","doi":"10.1016/j.compchemeng.2025.109217","DOIUrl":"10.1016/j.compchemeng.2025.109217","url":null,"abstract":"<div><div>Life Cycle Assessment (LCA) is the methodology used to evaluate the environmental impacts of materials and services throughout their life cycle, which includes the production/manufacturing (cradle-to-gate (C2G)) phase, use (gate-to-gate (G2G)) phase, and discard (gate-to-grave (G2G)) or recycle (gate-to-cradle (G2C)) phases. The overall LCA from the manufacturing to end-of-life (cradle-to-cradle (C2C)) impact assessment requires Life Cycle Inventory (LCI) in several impact categories such as human health, global warming potential, resource consumption, and ecosystem quality. The life cycle inventories (LCI) for new chemicals and process technologies are not readily available in the early stages of process design. However, these are required to assess if the novel chemicals and alternate technologies are eco-friendly, have long-term sustainability, and are safer when implemented at large scales. Determining these via conventional methods of laboratory experimentation, pilot-scale studies, computationally intensive molecular simulations, and group contribution methods can be time-consuming. To this end, we propose Machine Learning (ML) approaches whereby LCIs can be predicted from data collected in the exploratory stage of process development such as physiochemical, molecular, and structural properties of chemicals as input features. These input features are readily available in well-established chemical databases or can be measured via laboratory experimental techniques. Through ML-based predictive algorithms such as neural networks and extreme gradient boosting, we can generate the LCI data in different categories as output labels. To demonstrate the utility of these ML-predicted LCI values in enabling a full cradle-to-cradle LCA, a case study is presented. This methodology could be easily adapted to other case studies for conducting an early-stage LCA, which shall enable the design of inherently sustainable processes.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"201 ","pages":"Article 109217"},"PeriodicalIF":3.9,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144231882","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}

{"title":"Accelerating process synthesis with reinforcement learning: Transfer learning from multi-fidelity simulations and variational autoencoders","authors":"Qinghe Gao,&nbsp;Haoyu Yang,&nbsp;Maximilian F. Theisen,&nbsp;Artur M. Schweidtmann","doi":"10.1016/j.compchemeng.2025.109192","DOIUrl":"10.1016/j.compchemeng.2025.109192","url":null,"abstract":"<div><div>Reinforcement learning has shown some success in automating process design by integrating data-driven models that interact with process simulators to learn to build process flowsheets iteratively. However, one major challenge in the learning process is that the reinforcement learning agent demands numerous process simulations in rigorous process simulators, thereby requiring long simulation times and expensive computational power. We propose employing transfer learning to enhance the reinforcement learning process in process design. This study examines two transfer learning strategies: (i) transferring knowledge from shortcut process simulators to rigorous simulators, and (ii) transferring knowledge from process variational autoencoders (VAEs). Our findings reveal that appropriate transfer learning can significantly improve both learning efficiency and convergence scores. However, transfer learning can also negatively impact the learning process when there are substantial discrepancies in decision range and reward function. This suggests that pre-trained process data should match the complexity of the fine-tuning task.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"201 ","pages":"Article 109192"},"PeriodicalIF":3.9,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144240999","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}

{"title":"Enhanced PEMFC parameter estimation using a hybrid gorilla troops optimizer and honey badger algorithm","authors":"Mohammad Aljaidi ,&nbsp;Sunilkumar P. Agrawal ,&nbsp;Toshika R. Agrawal ,&nbsp;Sundaram B. Pandya ,&nbsp;Anil Parmar ,&nbsp;Pradeep Jangir ,&nbsp;Arpita ,&nbsp;G. Gulothungan ,&nbsp;Muteb Alshammari ,&nbsp;Reena Jangid","doi":"10.1016/j.compchemeng.2025.109216","DOIUrl":"10.1016/j.compchemeng.2025.109216","url":null,"abstract":"<div><div>The optimization of proton exchange membrane fuel cell (PEMFC) parameters is crucial for improving performance, durability, and efficiency across various applications. However, existing metaheuristic algorithms face limitations such as slow convergence, susceptibility to local optima, and high computational costs. To address these issues, this study proposes a hybrid Artificial Gorilla Troops Optimizer and Honey Badger Algorithm (GTOHBA) to enhance PEMFC parameter estimation. The proposed algorithm integrates the efficient global search of GTO with the adaptive foraging behavior of HBA, ensuring better exploration-exploitation balance. The performance of GTOHBA was evaluated on six different PEMFC models: BCS 500 W, SR-12, Horizon H-12, Nedstack 600 W PS6, STD 250 W and Ballard Mark V. The results demonstrate that GTOHBA achieved the lowest sum of squared errors (<em>SSE_min</em> = 0.025493) with an exceptionally low standard deviation (<em>SSE_SD</em> = 8.35×10⁻⁶), outperforming other state-of-the-art optimization methods. Moreover, GTOHBA exhibited superior computational efficiency with the shortest runtime (0.179909 s), significantly reducing the computational burden compared to PSO and MFO. The theoretical analysis confirms that GTOHBA effectively mitigates issues of premature convergence and enhances parameter extraction accuracy through its hybrid structure. Practically, the algorithm enables precise modeling of PEMFC behavior, improving power output predictions with minimal absolute error (AE = 0.012913) and percentage relative error ( %RE = 0.061363 %). The key benefit for readers is the demonstration of a robust, fast, and highly accurate optimization framework for PEMFC parameter estimation. However, further research is needed to assess GTOHBA’s adaptability under dynamic real-time conditions and large-scale fuel cell stacks.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"201 ","pages":"Article 109216"},"PeriodicalIF":3.9,"publicationDate":"2025-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144231883","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}

{"title":"A novel deep graph learning-based multivariate clustering method for time series forecasting of complex chemical systems","authors":"Jing Huang ,&nbsp;Rui Qin","doi":"10.1016/j.compchemeng.2025.109214","DOIUrl":"10.1016/j.compchemeng.2025.109214","url":null,"abstract":"<div><div>Many coupled monitoring variables in complex chemical systems form a dynamic coupling relationship network that can reflect the operating status of the system. The existing data-driven methods based on neural network for mining big data information lack specific learning mechanisms. To this end, this paper proposes a multivariate time series prediction method based on graph convolutional neural (GCN) and long short-term memory (LSTM) according to multivariate directed coupling relations. Specifically, the coupling Granger causality measure was first used to analyze the nonlinear causal relationship between multiple variables. Then, the edge-weight connection relationship was embedded in the training process of GCN. This hierarchical information transmission method achieves precise division of the network community structure for monitoring variables. Finally, prediction was accomplished by using LSTM to obtain temporal characteristics of non-targeted monitoring variables in the communities where the target monitoring variables were located. The effectiveness and reliability of the proposed method have been verified on the actual monitoring series of the compressor unit in the chemical production system. The experimental results show that the proposed method outperforms existing methods in terms of comprehensive prediction accuracy and computational complexity. Also, the proposed method can still maintain strong predictive ability under abnormal system conditions.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"200 ","pages":"Article 109214"},"PeriodicalIF":3.9,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144194991","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}

{"title":"Dynamics of batch protein precipitation","authors":"Amirkiarash Ehtesham ,&nbsp;Abhishek Sivaram ,&nbsp;Sara Danielle Siegel ,&nbsp;John van Zanten ,&nbsp;Seyed Soheil Mansouri","doi":"10.1016/j.compchemeng.2025.109193","DOIUrl":"10.1016/j.compchemeng.2025.109193","url":null,"abstract":"<div><div>Precipitation is a valuable alternative to recovery processes like chromatography. While the process is a plausible alternative, without experimentation, there is a lack of sufficient understanding of how process operating conditions lead to expected chord length distributions (also referred to as particle size distributions). In this work, a modeling framework is proposed through which, first, the system is specified, then the population balance model (PBM) is constructed. Data is then gathered from experiments and subsequently processed. Afterwards, an optimization problem is defined to regress the PBM parameters to the experimental data. Finally, a surrogate model correlating the PBM parameters with process operating conditions is developed. Thereby, avoiding future experiments and regression of the parameters. The application of the modeling framework is demonstrated using 15 batch experiments for lysozyme precipitation and particle size distribution measurements using focused beam reflectance measurement (FBRM). The current approach helps identify the correlations between process operating conditions and process model (PBM) parameters such as maximum collision efficiency and breakage rate coefficient, which in turn helps in the estimation of the population distribution in the system. The work also proposes the limitations with modeling and data acquisition that pave the way for future research.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"200 ","pages":"Article 109193"},"PeriodicalIF":3.9,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144194990","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}

{"title":"Caffeine, riboflavin and curcumin amended clays for PFAS binding","authors":"Xenophon Xenophontos ,&nbsp;Johnson O. Oladele ,&nbsp;Meichen Wang ,&nbsp;Kendall Lilly ,&nbsp;Laura Martinez ,&nbsp;Timothy D. Phillips ,&nbsp;Phanourios Tamamis","doi":"10.1016/j.compchemeng.2025.109215","DOIUrl":"10.1016/j.compchemeng.2025.109215","url":null,"abstract":"<div><div>Per- and polyfluoroalkyl substances (PFAS) are usually found in mixtures with other toxic compounds. Therefore, the study and design of broad acting sorbents, such as clays, is an attractive sorption solution. We previously demonstrated that clays amended with choline and carnitine could enhance PFAS sorption properties. Here, we used computations to screen from a pool of chemical compounds, which are either supplements or generally recognized as safe, and identified particular supplements that can be amended to clay and potentially improve its sorbing capacity for PFAS in acidic conditions. Simulations were initially used as a tool to identify promising amendments to the clay, while subsequently, simulations evaluated which selected amendments could potentially bind PFAS. Our results showed that caffeine-, riboflavin- and curcumin-amended clays can, in particular instances, enhance the binding of different PFAS compared to parent clays. Experiments investigated the sorption properties of the designed systems. Notably, caffeine-amended clay significantly enhanced GenX binding when compared to parent clay, with its binding capacity being increased from 0.15 mol/kg to 1.17 mol/kg. Caffeine-amended clay also enhanced binding for PFOS by 125%, compared to the parent clay, and for PFOA to a lesser extent. Additionally, riboflavin-amended clay enhanced binding for GenX, PFOA and PFOS by 120%, 23%, and 70%, respectively, compared to the parent clay. Our studies provide atomistic details into their mechanisms of action. Both the novel computational library of chemical compound-amended clays and the approach utilized, combining computations and experiments, could enhance the future design of novel amended clays for other toxins.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"201 ","pages":"Article 109215"},"PeriodicalIF":3.9,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144254867","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}

{"title":"Multi-objective optimization design of heat exchanger networks with simultaneous evaluation of economic performance and complexity via a novel two-step optimization approach","authors":"Kai Zhao ,&nbsp;Li Zhao ,&nbsp;Weizhuo Lu ,&nbsp;Qiao Q. Tang ,&nbsp;Chang He ,&nbsp;Qing L. Chen ,&nbsp;Bing J. Zhang","doi":"10.1016/j.compchemeng.2025.109212","DOIUrl":"10.1016/j.compchemeng.2025.109212","url":null,"abstract":"<div><div>Heat exchanger networks (HENs) are critical for energy integration in industrial processes, but their optimization often results in highly complex topologies that hinder operability. To address this challenge, this study proposes a novel engineering complexity index (ECI) to quantitatively evaluate HEN structural complexity, incorporating key features such as variations in the number of streams splitting and mixing repeatedly, the number of heat exchangers used, and the minimum temperature difference. The entropy weight method, calibrated with industrial data, is used to objectively determine the weight of each factor. For the first time, ECI is integrated with total annual cost (TAC) in a multi-objective optimization framework that balances economic and structural objectives. Computational efficiency is enhanced through temperature-difference parameterization and a two-step optimization strategy that generates a well-defined Pareto frontier. Two representative examples are investigated to validate the proposed method. In Example 1, the TAC is reduced by 41.2% with a corresponding 176.7% increase in ECI from point A to F. In Example 2, a 23.8% decrease in TAC is observed alongside an 89.8% increase in ECI from point A to E. Compared with previous single-objective results, the proposed approach achieves a 1.8% further reduction in TAC in Example 1, and only a marginal 1.4% TAC increase under complexity constraints relative to our earlier work in Example 2. These results confirm the method’s effectiveness in generating cost-effective and operable HEN designs, providing a robust decision-support framework for industrial applications.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"200 ","pages":"Article 109212"},"PeriodicalIF":3.9,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144194992","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}

{"title":"Multi-objective dispatch of integrated renewable power systems leveraging robust optimization in deep reinforcement learning","authors":"Aoqun Ma ,&nbsp;Zhi Li ,&nbsp;Feifei Shen ,&nbsp;Xin Peng ,&nbsp;Yurong Liu ,&nbsp;Weimin Zhong ,&nbsp;Feng Qian","doi":"10.1016/j.compchemeng.2025.109173","DOIUrl":"10.1016/j.compchemeng.2025.109173","url":null,"abstract":"<div><div>The time-varying nature of wind speed and solar radiation introduces significant intermittency and uncertainty to the grid integration of renewable energy. We propose a robust optimization method with a dynamic framework based on reinforcement learning to address this challenge. Our approach considers economic cost and carbon emissions as objective functions in a multi-objective robust optimization model. Day-ahead scheduling results and real-time renewable energy forecasting are used to dynamically adjust the uncertainty set in dispatch, employing support vector clustering and Deep Q-Network. The dynamic framework aims to achieve feasible and cost-effective dispatch by accounting for real-time prediction errors and penalty costs associated with energy spillage and load curtailment risks. A practical industrial system case study shows that the proposed algorithm significantly reduces average day-ahead costs by 21.43% compared to static solutions and exhibits better environmental performance compared to the counterparts.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"201 ","pages":"Article 109173"},"PeriodicalIF":3.9,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144223633","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}

{"title":"Adaptive model-based predictive control with changing operation points","authors":"Hugo A. Pipino ,&nbsp;Eduardo J. Adam","doi":"10.1016/j.compchemeng.2025.109188","DOIUrl":"10.1016/j.compchemeng.2025.109188","url":null,"abstract":"<div><div>Most industrial processes are nonlinear and experience frequent variations in the operating point, which can make them impractical for real-time Model-based Predictive Control (MPC) implementation. This research explores the design and analysis of MPC formulations developed within the context of Linear Parameter Varying (LPV) model framework. These methods take into account the scheduling parameters of the multi-model and perform online process-model adaptation, obtaining a linear prediction model that allows representing the nonlinear process at each instant. Additionally, necessary conditions are established to guarantee the asymptotic stability of the feasible equilibrium set for all models contained in the LPV model. This enables the consideration of changes in operating points that occur during the normal operation of the process. The article concludes with realistic simulation results of two typical unit operations in the process industry, comparing the analyzed MPC techniques with a linear MPC present in the literature. Discussions are presented on the results in terms of performance, effectiveness, computational effort and disturbance rejection, in the presence of changing operating points.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"200 ","pages":"Article 109188"},"PeriodicalIF":3.9,"publicationDate":"2025-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144178772","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}

{"title":"Digital twins for accurate prediction beyond routine operation","authors":"Bryan Li ,&nbsp;Isaac Severinsen ,&nbsp;Wei Yu ,&nbsp;Timothy Walmsley ,&nbsp;Brent Young","doi":"10.1016/j.compchemeng.2025.109211","DOIUrl":"10.1016/j.compchemeng.2025.109211","url":null,"abstract":"<div><div>Non-routine operating conditions in process plants can hinder safe operation and effective control due to limited understanding of system behavior. Therefore, it is desirable that process digital twins are capable of accurately extrapolating system behavior beyond the scenarios typically encountered. Physics-informed neural networks, which integrate data-driven models with physics constraints, offer a promising solution to this challenge. However, the constraints often fail to accurately represent the physics of chemical engineering unit operations and rigorously enforcing these constraints throughout training may compromise predictive accuracy. Furthermore, many studies focus on complex scenarios that require detailed equations, which may not be available for industrial plants or may be in fact unnecessary depending on the specific behavior being modeled. This study presents a novel approach to training digital twins for extrapolation by combining physics-informed neural networks with meta-learning, enabling the model to dynamically change the weight term of each governing differential equation throughout training. This method is applied to data from an industrial geothermal heat exchanger, using only the governing mass and energy balance partial differential equations as a part of the loss functions and excluding detailed heat transfer processes. The proposed approach eliminates the need for complex heat transfer models in predicting the outlet temperatures of geothermal heat exchangers. The meta-learning physics-informed neural network performed significantly better for the case study situations tested in this work (<span><math><mrow><msup><mrow><mi>R</mi></mrow><mn>2</mn></msup><mo>=</mo><mn>0.97</mn></mrow></math></span>), at the cost of nearly 30 times higher training time. The results highlight the potential of this approach to create high-fidelity digital twins for industrial applications.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"201 ","pages":"Article 109211"},"PeriodicalIF":3.9,"publicationDate":"2025-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144205134","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}