Pub Date : 2024-11-09DOI: 10.1016/j.compchemeng.2024.108915
Hazem Damiri , Martin Steinberger , Lisa Kuchler , Atabak Azimi , Stefano Martinuzzi , Peter Sagmeister , Jason D. Williams , Stefan Koch , Markus Tranninger , Jakob Rehrl , Selma Celikovic , Stephan Sacher , C. Oliver Kappe , Martin Horn
In this work, real-time optimization (RTO) schemes are proposed and applied on a continuous pharmaceutical manufacturing process which consists of three units: synthesis unit, hot melt extrusion unit and direct compaction line. The developed RTO strategies calculate the operating conditions by optimizing the considered objective functions while satisfying the specific constraints. Moreover, the RTO schemes can cope with intentional changes in the process and unintentional changes such as disturbances. Results from simulations and experiments are presented in this work. An advantageous performance is achieved when using the developed schemes.
{"title":"Model-based real-time optimization in continuous pharmaceutical manufacturing","authors":"Hazem Damiri , Martin Steinberger , Lisa Kuchler , Atabak Azimi , Stefano Martinuzzi , Peter Sagmeister , Jason D. Williams , Stefan Koch , Markus Tranninger , Jakob Rehrl , Selma Celikovic , Stephan Sacher , C. Oliver Kappe , Martin Horn","doi":"10.1016/j.compchemeng.2024.108915","DOIUrl":"10.1016/j.compchemeng.2024.108915","url":null,"abstract":"<div><div>In this work, real-time optimization (RTO) schemes are proposed and applied on a continuous pharmaceutical manufacturing process which consists of three units: synthesis unit, hot melt extrusion unit and direct compaction line. The developed RTO strategies calculate the operating conditions by optimizing the considered objective functions while satisfying the specific constraints. Moreover, the RTO schemes can cope with intentional changes in the process and unintentional changes such as disturbances. Results from simulations and experiments are presented in this work. An advantageous performance is achieved when using the developed schemes.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108915"},"PeriodicalIF":3.9,"publicationDate":"2024-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142663976","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-11-09DOI: 10.1016/j.compchemeng.2024.108919
Yan Qi , Lifeng Zhao , Haiqiu Tang , Lei Zhang , Rafiqul Gani
Computer-aided formulation design is a methodology that utilizes domain knowledge and selected methods and tools suitable for computer-based applications to assist in formulation (product) design. In this paper, molecular dynamics simulation and Bayesian neural network algorithms are combined with well-known engineering models to help accelerate the development and optimization of formulation-based detergent products with a view to improve product quality and performance. In particular, the mechanism of the behavior of polymers (an active ingredient in the product) to improve the product quality in terms of the fragrance and its residence time is highlighted. Results from molecular dynamic simulation applied to study the molecular interaction mechanism show that the polymers have an attraction effect with fragrance molecules and could adsorb more to make them to stay on the surface of clothes. In addition, the polymer attenuates the diffusion of the fragrance molecules, lengthening the entire process of fragrance diffusion, which is the essence of the ability of the polymer to slow down the release of the fragrance. A Quantitative Structure-Property Relationship (QSPR) model between component proportions and fragrance diffusion is established through Bayesian Neural Network (BNN) and the product formulation is optimized based on this model. Keeping polymer and perfume ingredients unchanged, the surfactant amounts are optimized to provide improved product quality.
{"title":"Computer aided formulation design based on molecular dynamics simulation: Detergents with fragrance","authors":"Yan Qi , Lifeng Zhao , Haiqiu Tang , Lei Zhang , Rafiqul Gani","doi":"10.1016/j.compchemeng.2024.108919","DOIUrl":"10.1016/j.compchemeng.2024.108919","url":null,"abstract":"<div><div>Computer-aided formulation design is a methodology that utilizes domain knowledge and selected methods and tools suitable for computer-based applications to assist in formulation (product) design. In this paper, molecular dynamics simulation and Bayesian neural network algorithms are combined with well-known engineering models to help accelerate the development and optimization of formulation-based detergent products with a view to improve product quality and performance. In particular, the mechanism of the behavior of polymers (an active ingredient in the product) to improve the product quality in terms of the fragrance and its residence time is highlighted. Results from molecular dynamic simulation applied to study the molecular interaction mechanism show that the polymers have an attraction effect with fragrance molecules and could adsorb more to make them to stay on the surface of clothes. In addition, the polymer attenuates the diffusion of the fragrance molecules, lengthening the entire process of fragrance diffusion, which is the essence of the ability of the polymer to slow down the release of the fragrance. A Quantitative Structure-Property Relationship (QSPR) model between component proportions and fragrance diffusion is established through Bayesian Neural Network (BNN) and the product formulation is optimized based on this model. Keeping polymer and perfume ingredients unchanged, the surfactant amounts are optimized to provide improved product quality.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108919"},"PeriodicalIF":3.9,"publicationDate":"2024-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142663975","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}
Pub Date : 2024-11-06DOI: 10.1016/j.compchemeng.2024.108912
Jing Wang , Christopher L.E. Swartz , Kai Huang
Classical reinforcement learning (RL) may suffer performance degradation when the environment deviates from training conditions, limiting its application in risk-averse supply chain management. This work explores using robust RL in supply chain operations to hedge against environment inconsistencies and changes. Two robust RL algorithms, -learning and -pessimistic -learning, are examined against conventional -learning and a baseline order-up-to inventory policy. Furthermore, this work extends RL applications from forward to closed-loop supply chains. Two case studies are conducted using a supply chain simulator developed with agent-based modeling. The results show that -learning can outperform the baseline policy under normal conditions, but notably degrades under environment deviations. By comparison, the robust RL models tend to make more conservative inventory decisions to avoid large shortage penalties. Specifically, fine-tuned -pessimistic -learning can achieve good performance under normal conditions and maintain robustness against moderate environment inconsistencies, making it suitable for risk-averse decision-making.
{"title":"Risk-averse supply chain management via robust reinforcement learning","authors":"Jing Wang , Christopher L.E. Swartz , Kai Huang","doi":"10.1016/j.compchemeng.2024.108912","DOIUrl":"10.1016/j.compchemeng.2024.108912","url":null,"abstract":"<div><div>Classical reinforcement learning (RL) may suffer performance degradation when the environment deviates from training conditions, limiting its application in risk-averse supply chain management. This work explores using robust RL in supply chain operations to hedge against environment inconsistencies and changes. Two robust RL algorithms, <span><math><mover><mrow><mi>Q</mi></mrow><mrow><mo>ˆ</mo></mrow></mover></math></span>-learning and <span><math><mi>β</mi></math></span>-pessimistic <span><math><mi>Q</mi></math></span>-learning, are examined against conventional <span><math><mi>Q</mi></math></span>-learning and a baseline order-up-to inventory policy. Furthermore, this work extends RL applications from forward to closed-loop supply chains. Two case studies are conducted using a supply chain simulator developed with agent-based modeling. The results show that <span><math><mi>Q</mi></math></span>-learning can outperform the baseline policy under normal conditions, but notably degrades under environment deviations. By comparison, the robust RL models tend to make more conservative inventory decisions to avoid large shortage penalties. Specifically, fine-tuned <span><math><mi>β</mi></math></span>-pessimistic <span><math><mi>Q</mi></math></span>-learning can achieve good performance under normal conditions and maintain robustness against moderate environment inconsistencies, making it suitable for risk-averse decision-making.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108912"},"PeriodicalIF":3.9,"publicationDate":"2024-11-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142663977","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}
Despite the recent stunning progress in large-scale deep neural network applications, our understanding of their microstructure, ‘energy’ functions, and optimal design remains incomplete. Here, we present a new game-theoretic framework, called statistical teleodynamics, that reveals important insights into these key properties. The optimally robust design of such networks inherently involves computational benefit–cost trade-offs that physics-inspired models do not adequately capture. These trade-offs occur as neurons and connections compete to increase their effective utilities under resource constraints during training. In a fully trained network, this results in a state of arbitrage equilibrium, where all neurons in a given layer have the same effective utility, and all connections to a given layer have the same effective utility. The equilibrium is characterized by the emergence of two lognormal distributions of connection weights and neuronal output as the universal microstructure of large deep neural networks. We call such a network the Jaynes Machine. Our theoretical predictions are shown to be supported by empirical data from seven large-scale deep neural networks. We also show that the Hopfield network and the Boltzmann Machine are the same special case of the Jaynes Machine.
{"title":"Jaynes machine: The universal microstructure of deep neural networks","authors":"Venkat Venkatasubramanian , N. Sanjeevrajan , Manasi Khandekar , Abhishek Sivaram , Collin Szczepanski","doi":"10.1016/j.compchemeng.2024.108908","DOIUrl":"10.1016/j.compchemeng.2024.108908","url":null,"abstract":"<div><div>Despite the recent stunning progress in large-scale deep neural network applications, our understanding of their microstructure, ‘energy’ functions, and optimal design remains incomplete. Here, we present a new game-theoretic framework, called statistical teleodynamics, that reveals important insights into these key properties. The optimally robust design of such networks inherently involves computational benefit–cost trade-offs that physics-inspired models do not adequately capture. These trade-offs occur as neurons and connections compete to increase their effective utilities under resource constraints during training. In a fully trained network, this results in a state of arbitrage equilibrium, where all neurons in a given layer have the same effective utility, and all connections to a given layer have the same effective utility. The equilibrium is characterized by the emergence of two lognormal distributions of connection weights and neuronal output as the universal microstructure of large deep neural networks. We call such a network the Jaynes Machine. Our theoretical predictions are shown to be supported by empirical data from seven large-scale deep neural networks. We also show that the Hopfield network and the Boltzmann Machine are the same special case of the Jaynes Machine.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108908"},"PeriodicalIF":3.9,"publicationDate":"2024-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142663974","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-11-02DOI: 10.1016/j.compchemeng.2024.108907
Jaeyeon Kim , Luthfan Adhy Lesmana , Muhammad Aziz
This study focuses on the impacts of particle's sphericity on the properties of porous materials crucial to electrochemical devices. Three-dimensional structures with spherical and cylindrical particles were generated to simulate porous granular and fibrous materials. The constructed particle geometries are as follows: a sphere and cylinders with different aspect ratios (height-to-diameter) of 0.1, 0.5, 1.0, 2.5, 5.0, 10, and 20. Every model exhibits a porosity of 0.500 ± 0.001 to exclude the effects of porosity. The structures were binarized with 200×200×200 dimensionless voxels, which were analyzed with the specific surface area, grain and pore size distributions, geometrical tortuosity, conductivity, and diffusivity across the through- and in-planes. As a result, the particle geometry significantly impacts on tortuosity, conductivity, and diffusivity, with the absolute value of Spearman's correlation coefficient of up to 1. It may imply the necessity to consider particle geometry as an ex-situ characterization for better electrochemical performance.
{"title":"Impact analysis of particle sphericity on the properties of porous materials via particle packing method for hydrogen fuel and electrolysis cells","authors":"Jaeyeon Kim , Luthfan Adhy Lesmana , Muhammad Aziz","doi":"10.1016/j.compchemeng.2024.108907","DOIUrl":"10.1016/j.compchemeng.2024.108907","url":null,"abstract":"<div><div>This study focuses on the impacts of particle's sphericity on the properties of porous materials crucial to electrochemical devices. Three-dimensional structures with spherical and cylindrical particles were generated to simulate porous granular and fibrous materials. The constructed particle geometries are as follows: a sphere and cylinders with different aspect ratios (height-to-diameter) of 0.1, 0.5, 1.0, 2.5, 5.0, 10, and 20. Every model exhibits a porosity of 0.500 ± 0.001 to exclude the effects of porosity. The structures were binarized with 200×200×200 dimensionless voxels, which were analyzed with the specific surface area, grain and pore size distributions, geometrical tortuosity, conductivity, and diffusivity across the through- and in-planes. As a result, the particle geometry significantly impacts on tortuosity, conductivity, and diffusivity, with the absolute value of Spearman's correlation coefficient of up to 1. It may imply the necessity to consider particle geometry as an <em>ex-situ</em> characterization for better electrochemical performance.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108907"},"PeriodicalIF":3.9,"publicationDate":"2024-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142573013","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-11-02DOI: 10.1016/j.compchemeng.2024.108898
Lu Zhang , Junyao Xie , Qingqing Xu , Charles Robert Koch , Stevan Dubljevic
Hydrogen energy, as one of the promising future energy forms, has attracted attentions from academia and industry due to its cost-effective and low-carbon nature. Compared with oil and gas transportation, its transportation is more challenging due to its complex blending mechanism. Inferring the internal states during transportation is essential for condition monitoring and operational planning of hydrogen-blending natural gas pipelines. Considering the nonlinear spatiotemporal dynamics and limited sensor information, reconstructing infinite-dimensional pipeline state variables is challenging. This paper addresses the state reconstruction of nonlinear infinite-dimensional hydrogen-blending natural gas pipeline systems using physics-informed neural networks. The proposed design combines neural networks with nonlinear partial differential equations that govern the pipeline systems. With limited measurements, the trained model is capable of predicting the state evolutions of pressure, flow, and mass flux ratio of hydrogen during transient transportation at any location. The proposed design is demonstrated through detailed numerical simulations and sensitivity analyses.
{"title":"Physics-informed neural networks for state reconstruction of hydrogen energy transportation systems","authors":"Lu Zhang , Junyao Xie , Qingqing Xu , Charles Robert Koch , Stevan Dubljevic","doi":"10.1016/j.compchemeng.2024.108898","DOIUrl":"10.1016/j.compchemeng.2024.108898","url":null,"abstract":"<div><div>Hydrogen energy, as one of the promising future energy forms, has attracted attentions from academia and industry due to its cost-effective and low-carbon nature. Compared with oil and gas transportation, its transportation is more challenging due to its complex blending mechanism. Inferring the internal states during transportation is essential for condition monitoring and operational planning of hydrogen-blending natural gas pipelines. Considering the nonlinear spatiotemporal dynamics and limited sensor information, reconstructing infinite-dimensional pipeline state variables is challenging. This paper addresses the state reconstruction of nonlinear infinite-dimensional hydrogen-blending natural gas pipeline systems using physics-informed neural networks. The proposed design combines neural networks with nonlinear partial differential equations that govern the pipeline systems. With limited measurements, the trained model is capable of predicting the state evolutions of pressure, flow, and mass flux ratio of hydrogen during transient transportation at any location. The proposed design is demonstrated through detailed numerical simulations and sensitivity analyses.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108898"},"PeriodicalIF":3.9,"publicationDate":"2024-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142663973","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-11-02DOI: 10.1016/j.compchemeng.2024.108902
Marcello Di Martino , Patrick Linke , Efstratios N. Pistikopoulos
The food–energy–water nexus (FEWN) postulates that sustainable decision-making regarding the interconnected resources food, energy and water must consider all involved resources holistically. Due to its multi-scale complexity, modeling challenges and computational intractability regarding the interconnected FEWN optimization remain. To overcome these challenges, this work proposes employing surrogate models based on data-driven and model optimization techniques, while quantifying the introduced errors due to both the selected approximation and optimization methods. In turn, we derive a mixed-integer linear FEWN planning and scheduling optimization model based on a greenhouse farming, a renewable energy and a reverse osmosis desalination water supply system, which is initially computationally intractable. This computational complexity is first discussed and overcome for the energy–water nexus supply system, before solving the complete FEWN supply system by utilizing strategies such as relaxation, modularization and convex hull reformulation.
粮食-能源-水关系（FEWN）假定，有关相互关联的粮食、能源和水资源的可持续决策必须全面考虑所有相关资源。由于其多尺度的复杂性，相互关联的 FEWN 优化仍面临建模挑战和计算难点。为了克服这些挑战，本研究提出采用基于数据驱动和模型优化技术的代用模型，同时量化因所选近似和优化方法而引入的误差。反过来，我们基于温室种植、可再生能源和反渗透海水淡化供水系统，推导出一个混合整数线性 FEWN 规划和调度优化模型，该模型最初在计算上是难以实现的。在利用松弛、模块化和凸壳重构等策略求解完整的 FEWN 供水系统之前，首先讨论并克服了能源-水关系供应系统的计算复杂性。
{"title":"Overcoming modeling and computational complexity challenges in food–energy–water nexus optimization","authors":"Marcello Di Martino , Patrick Linke , Efstratios N. Pistikopoulos","doi":"10.1016/j.compchemeng.2024.108902","DOIUrl":"10.1016/j.compchemeng.2024.108902","url":null,"abstract":"<div><div>The food–energy–water nexus (FEWN) postulates that sustainable decision-making regarding the interconnected resources food, energy and water must consider all involved resources holistically. Due to its multi-scale complexity, modeling challenges and computational intractability regarding the interconnected FEWN optimization remain. To overcome these challenges, this work proposes employing surrogate models based on data-driven and model optimization techniques, while quantifying the introduced errors due to both the selected approximation and optimization methods. In turn, we derive a mixed-integer linear FEWN planning and scheduling optimization model based on a greenhouse farming, a renewable energy and a reverse osmosis desalination water supply system, which is initially computationally intractable. This computational complexity is first discussed and overcome for the energy–water nexus supply system, before solving the complete FEWN supply system by utilizing strategies such as relaxation, modularization and convex hull reformulation.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108902"},"PeriodicalIF":3.9,"publicationDate":"2024-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142578938","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}
Pub Date : 2024-11-01DOI: 10.1016/j.compchemeng.2024.108901
Oscar Generoso Gutierrez , Clara Simón de Blas , Ana E. Garcia Sipols
Improving prediction computation for time series analysis is still a challenge. Finding a method that combines the benefits of different methodologies is still an open problem. Besides the very efficient prediction combination techniques proposed, there is still a lack of procedures that jointly consider error measure combinations and model constraints. In this work, we propose a new forecast combination procedure based on multi-criteria methods that allows the assignment of weights to different error measures in the objective function and the incorporation of constraints. A real case from the pharmaceutical industry for the sale of a probiotic product is presented to illustrate the performance of the proposal. This method is capable of considering different error measures and non distance based errors, is enriched by the consideration of constraints that consider desirable properties of the solution and is robust with respect to different time series characteristics such as trends, seasonality, etc. Results shows similar accuracy to the best known forecasting methods to date.
{"title":"Multi-criteria Forecast Combination Method with Nonlinear Programming for time series prediction models","authors":"Oscar Generoso Gutierrez , Clara Simón de Blas , Ana E. Garcia Sipols","doi":"10.1016/j.compchemeng.2024.108901","DOIUrl":"10.1016/j.compchemeng.2024.108901","url":null,"abstract":"<div><div>Improving prediction computation for time series analysis is still a challenge. Finding a method that combines the benefits of different methodologies is still an open problem. Besides the very efficient prediction combination techniques proposed, there is still a lack of procedures that jointly consider error measure combinations and model constraints. In this work, we propose a new forecast combination procedure based on multi-criteria methods that allows the assignment of weights to different error measures in the objective function and the incorporation of constraints. A real case from the pharmaceutical industry for the sale of a probiotic product is presented to illustrate the performance of the proposal. This method is capable of considering different error measures and non distance based errors, is enriched by the consideration of constraints that consider desirable properties of the solution and is robust with respect to different time series characteristics such as trends, seasonality, etc. Results shows similar accuracy to the best known forecasting methods to date.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108901"},"PeriodicalIF":3.9,"publicationDate":"2024-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142587399","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-10-30DOI: 10.1016/j.compchemeng.2024.108899
Mehmet Velioglu , Song Zhai , Sophia Rupprecht , Alexander Mitsos , Andreas Jupke , Manuel Dahmen
In chemical engineering, process data are expensive to acquire, and complex phenomena are difficult to fully model. We explore the use of physics-informed neural networks (PINNs) for modeling dynamic processes with incomplete mechanistic semi-explicit differential–algebraic equation systems and scarce process data. In particular, we focus on estimating states for which neither direct observational data nor constitutive equations are available. We propose an easy-to-apply heuristic to assess whether estimation of such states may be possible. As numerical examples, we consider a continuously stirred tank reactor and a liquid–liquid separator. We find that PINNs can infer immeasurable states with reasonable accuracy, even if respective constitutive equations are unknown. We thus show that PINNs are capable of modeling processes when relatively few experimental data and only partially known mechanistic descriptions are available, and conclude that they constitute a promising avenue that warrants further investigation.
{"title":"Physics-informed neural networks for dynamic process operations with limited physical knowledge and data","authors":"Mehmet Velioglu , Song Zhai , Sophia Rupprecht , Alexander Mitsos , Andreas Jupke , Manuel Dahmen","doi":"10.1016/j.compchemeng.2024.108899","DOIUrl":"10.1016/j.compchemeng.2024.108899","url":null,"abstract":"<div><div>In chemical engineering, process data are expensive to acquire, and complex phenomena are difficult to fully model. We explore the use of physics-informed neural networks (PINNs) for modeling dynamic processes with incomplete mechanistic semi-explicit differential–algebraic equation systems and scarce process data. In particular, we focus on estimating states for which neither direct observational data nor constitutive equations are available. We propose an easy-to-apply heuristic to assess whether estimation of such states may be possible. As numerical examples, we consider a continuously stirred tank reactor and a liquid–liquid separator. We find that PINNs can infer immeasurable states with reasonable accuracy, even if respective constitutive equations are unknown. We thus show that PINNs are capable of modeling processes when relatively few experimental data and only partially known mechanistic descriptions are available, and conclude that they constitute a promising avenue that warrants further investigation.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108899"},"PeriodicalIF":3.9,"publicationDate":"2024-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142706028","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-10-30DOI: 10.1016/j.compchemeng.2024.108903
Uthraa K. Ramesh , Styliani Avraamidou , Hari S. Ganesh
Climate control in buildings involves multiple conflicting objectives, such as energy consumption and occupant comfort, which have to be considered simultaneously during the operation of the climate control system. In this work, the Multi-Objective Model Predictive Control (MOMPC) solution method is further developed through the multiparametric programming approach (mpMOMPC). The MOMPC optimal control problem is reformulated according to the -constraint method, and the vector is treated as unknown parameters to generate the control law expressions offline. This reduces online calculations to point location followed by function evaluation, enabling the controller to be implemented through a chip or low-cost hardware. To demonstrate the potential and versatility of the developed mpMOMPC algorithm, three case studies are conducted. Numerical simulation results show that the extreme-value case is the same as the rule-based MPC case and the preference function case results in maximum energy reduction by 20.1% compared to the rule-based MPC case.
{"title":"Energy and temperature management in buildings through Multi-Objective Model Predictive Control on a chip","authors":"Uthraa K. Ramesh , Styliani Avraamidou , Hari S. Ganesh","doi":"10.1016/j.compchemeng.2024.108903","DOIUrl":"10.1016/j.compchemeng.2024.108903","url":null,"abstract":"<div><div>Climate control in buildings involves multiple conflicting objectives, such as energy consumption and occupant comfort, which have to be considered simultaneously during the operation of the climate control system. In this work, the Multi-Objective Model Predictive Control (MOMPC) solution method is further developed through the multiparametric programming approach (mpMOMPC). The MOMPC optimal control problem is reformulated according to the <span><math><mi>ϵ</mi></math></span>-constraint method, and the <span><math><mi>ϵ</mi></math></span> vector is treated as unknown parameters to generate the control law expressions offline. This reduces online calculations to point location followed by function evaluation, enabling the controller to be implemented through a chip or low-cost hardware. To demonstrate the potential and versatility of the developed mpMOMPC algorithm, three case studies are conducted. Numerical simulation results show that the extreme-value case is the same as the rule-based MPC case and the preference function case results in maximum energy reduction by 20.1% compared to the rule-based MPC case.</div></div>","PeriodicalId":286,"journal":{"name":"Computers & Chemical Engineering","volume":"192 ","pages":"Article 108903"},"PeriodicalIF":3.9,"publicationDate":"2024-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142578859","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}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号