Batch production scheduling of resources to meet fluctuating product demand is a critical topic in the process industry. Existing optimisation approaches, based on heuristic and exact methods, trade off solution optimality and scalability to large problems. In this work, we investigate deep reinforcement learning as a powerful alternative in order to learn heuristics for batch scheduling. We formulate the batch scheduling problem as a Markov decision process operating on a state-task network representation encoded using graph neural networks, capturing relevant structural inductive biases. We propose a centralised training with decentralised execution architecture, in which agents placed on machines individually choose which tasks to complete using a global view of the network, cooperating towards task schedules that optimise the final production quantity. Preliminary results demonstrate that the proposed end-to-end framework learns to construct task schedules comparable to the optimal... [more]

This work aims to systematically generate intensified process designs by integrating reinforcement learning (RL)-driven process synthesis and phenomena-based modeling via Generalized Modular Framework (GMF). Rather than considering flowsheet synthesis with conventional unit-operations, GMF utilizes fundamental building blocks, also known as mass and heat exchange modules, to describe the physiochemical phenomena and to enhance novel process discovery. At its core are driving forces which characterize the mass transfer feasibility based on the total change in Gibbs free energy of the system. RL is integrated with this phenomena-based modeling strategy to drive flowsheet generation by exploring much of the total action space and minimizing pre-postulation of stream connections. All possible inlets, outlets, and interconnections between modules are contained in a stream matrix. Deep Q-Network is used as the RL agent, which contains a multi-layer convolution neural network followed by a mu... [more]

Although solvent-based crystallization is widely adopted for separation and purification of crystalline pharmaceutical products, solvent choice and utilisation critically influence product quality, manufacturing cost, and the environmental performance of the pharmaceutical process. Escalating demands to reduce process mass intensity (PMI), together with increasing vulnerabilities in the supply chains, necessitate the development of more efficient and resilient process designs, incorporating solvent and active pharmaceutical ingredient (API) recycling. The conceptual design of crystallization processes offers a viable route to identify flowsheets with substantially reduced solvent consumption. In this paper we present a computer-aided molecular and process design (CAMPD) formulation to explore the benefits of solvent/API recycle for two processes/APIs: (i) a continuous cooling crystallization process for mefenamic acid (MA) employing a binary solvent mixture and (ii) a batch cooling cry... [more]

In recent years, research for systematic process design approaches has gained traction, especially with the rise in popularity of generative machine learning models and reinforcement learning. However, works from the literature will often focus on proof-of-concept studies, limited to a specific process synthesis problem. Despite showing promising results, it is not clear how easily these methodologies could be transposed to new applications, and whether they would be successful. In this context, this work evaluates the possibility of using a Natural Language Processing model, which has already proven itself for thermodynamic cycle generation, for another different case: seawater desalination. The processes generated by this model will initially be those using reverse osmosis processes aimed at desalinating a seawater solution containing 25000 ppm of NaCl. Results show that the model has been successful in designing structural reverse osmosis desalination processes without defining asse... [more]

Heat Pumps (HPs) can play a vital role in the decarbonization of heating in industry. The performance of a HP strongly depends on the refrigerant, the working fluid within the HP. In order to maximize HP performance, systematic selection of the refrigerant is key. Refrigerant choice affects the very feasibility of employing a HP to deliver heating to a process. A flexible and robust method is required to select refrigerants that are the best fit for a given heating application. A computer-aided molecular & process design (CAMPD) method is developed to design the optimal refrigerant that is tailored to process needs. The method is applied to three case studies across which the HP performance objectives and constraints, and heat source and heat sink temperatures are varied. In addition, the design of refrigerants with low (<150) global warming potentials and zero ozone depletion potentials is investigated. For all applications across all case studies, the CAMPD approach successfully iden... [more]

An optimized flowsheet can be generated by numerous approaches. Process optimization via superstructure is one of the methods used to provide solutions that consider the interactions between different decision layers. A process simulator-based optimization is considered in this work, as it offers a reliable and rigorous modeling environment. It is then coupled with a P-graph-based framework to reduce the tedious mathematical writing of the logical constraints to guarantee the structural coherence of a sequence of unit operations.The developed framework consists of three algorithms. The first algorithm transforms the superstructure flowsheet into a P-graph. The second algorithm gets process sub-structures from the superstructure by searching for active units corresponding to a set of decisions made, for example, by an optimizer. The third one checks structural feasibility by verifying that the resulting structure satisfies the five axioms of the original P-graph framework and two additi... [more]

Industrial decarbonization is crucial to reducing global emissions. Efficient processes lower energy use and reduce the environmental impacts, such as material use and waste, decreasing the overall industrial footprint. In this context, the present study explores the impact of reducing steam consumption (thermal energy) during the ammonia regeneration process in the production of sodium carbonate. A key feature of the Solvay process is ammonia recycling, which significantly reduces raw material consumption and ensures both economic and environmental sustainability. However, this stage is highly energy-intensive. To enhance energy efficiency in soda ash production, a study was conducted to analyze variation in temperature, pressure, and steam flow introduced into the ammonia regeneration system. The objective is to understand its impact on both ammonia recovery and the process's energy consumption. Variations in steam pressure do not impact on energy consumption of the process. By reduc... [more]

Industrial ammonia synthesis remains a high-impact decarbonization target due to the combined effect of large production volumes and reliance on CO2-intensive fossil-based hydrogen. Replacing it partially with renewable-based (green) electrolytic hydrogen can minimize direct emissions; however, the intermittency of solar and wind complicates retrofit strategies for existing continuous plants. This paper addresses a techno-economic study centered on a retrofit-oriented hybrid concept: a conventional natural-gas ammonia plant is operated within a pre-defined "hybridization envelope", where part of the process hydrogen is supplied by a renewable Power-to-Hydrogen subsystem. A steady-state process model (gray/blue baselines) is coupled with an hourly energy dispatch and a sizing optimization of solar/wind park, electrolyzer, battery, and hydrogen storage. A case study based on 2024 California data for renewables and carbon pricing illustrates how hybridization can reduce carbon intensity w... [more]

Grid-connected electrolyzers with intermediate hydrogen storage offer significant potential for reducing electricity costs through flexible operation under dynamic pricing. A threshold-based scheduling optimization approach is developed that derives interpretable on/off production rules from electricity price signals. The method identifies local price thresholds separating high-price from low-price periods, yielding binary production schedules. Adaptive horizon partitioning-subdividing the scheduling horizon when constant thresholds become infeasible-is combined with a receding horizon strategy that implements only a portion of each optimized schedule before re-optimization. This procedure enables systematic investigation of how characteristics of Integrated Electrolyzer-Storage Systems (IESS) influence cost reduction potential while maintaining computational tractability for both offline analysis and online implementation. A case study applying the approach to historical German electr... [more]

To effectively distribute power to a system of multiple electrolyzer stack units, control strategies have been developed that now need to be applied to heterogeneous electrolysis systems. These are the 'segment principle', the 'slow start principle' and the 'start-stop principle'. As there are many possible combinations to the system composition of a modular heterogeneous electrolysis system together with the most suitable control strategy, a discrete multi-criteria optimisation problem can be formulated. To solve this discrete multi-criteria optimisation problem, two discrete decision variables are introduced. One is the electrolysis system composition, represented by the power ratio/configuration (C). A total of 17 different configurations were used for this, consisting of different proportions of alkaline electrolysis (AEL) and proton exchange membrane electrolysis (PEMEL). The other one are the control strategies (R). For the control strategies, the conventional strategies, mention... [more]

Today, beverage production plants are planned and designed from the material-handling context as a packaged-goods production facility, not as a process plant. Therefore, a lot of potential for optimization exists. This paper presents a new approach to the design of beverage production plants according to the design of process plants. A component library for the simple creation of beverage production plant process models is developed. All steps in the plant design process can be accelerated and automated to be used for the high number of existing and new installations around the world. As first use case an energy optimization upgrade for existing Carbonated-Soft-Drink production lines is described to save cooling and heating energy in warm climates.

In the pursuit of decarbonisation, process industries are turning to electrification as a solution to avoid fossil fuels for heating and processing raw material. Transitioning to renewable electricity couples the processes to varying electricity availability and requires more consideration for production timing and scheduling to support grid stability and avoid high electricity prices. However, practical challenges limit the capability for unforeseen rescheduling for large processes. This paper explores the idea of auxiliary flexibility in an electrified steel production process, where only the auxiliary systems can react to changing conditions. We model an H2-DRI-EAF inspired process with controllable Air-Separation unit, water electrolysis, pressurized hydrogen storage, gas liquefaction units, and a battery energy storage system to react to a production related demand delay. First, we compare hourly and 15-minute DA pricing and observe that without fast flexibility the cost differenc... [more]

For process simulation, Python and Aspen Plus can be combined to leverage the advantages of both. This work utilizes the integration of Python and Aspen Plus for the design and optimization of a lycopene production process from tomato waste. Three production pathways are studied: acetone and hexane as solvents, enzymes with ethyl acetate, and supercritical CO2 with ethanol. This allows for the scaling of laboratory-scale studies into industrial-scale analyses. Genetic algorithms are used for optimization, enabling the determination of the optimal process design, costs, and operating conditions, while minimizing the total annual cost. The process with acetone and hexane yields a final production of 0.21 kg/h, the process with enzymes and ethyl acetate, 5.13 kg/h, and the process with supercritical CO2 and ethanol, 0.13 kg/h. It is shown that the process with ethyl acetate has a higher production and the process with supercritical CO2 and ethanol results in lower production and higher co... [more]

Manufacturing disruptions can be a major driving factor in the wastage of resources and delays which result in spiralling costs and cancelled orders. Operational decision making should therefore consider the potential for disruptions from as many sources as possible, encouraging improvements to operational resilience and agility. Our work presents a scheduling and rescheduling framework formulated as a rolling horizon problem for the emulation of real time decision making within a dynamically changing scenario. The framework is applied to a complex multistage problem with parallel lines susceptible to disruptions as a result of process or equipment failures, or ineffective inventory management that results in material shortages. The framework is demonstrated for a simple example case which highlights the impact of disruptions on the time taken to complete orders and the associated costs. It is observed that the inclusion of disruptions can alter equipment congestion, shifting focus for... [more]

Decarbonizing ammonia by replacing grey with green hydrogen directly affects the operation of the Haber-Bosch (HB) process. When directly coupled to green hydrogen production from renewable energy, the HB process should be able operate flexibly to match variable hydrogen supply. This study presents a structured approach for designing a load-flexible HB plant, supported by a rigorous process model. First, we screen 2, 000 designs at high (100%) and low (10%) hydrogen loads to assess operability. Only 1, 100 designs are feasible for both loads, underscoring the need to account for multivariable interactions during design. Next, we assess the economic feasibility of a base design, comparing HB operation under constant and flexible loads. Flexible operation reduces the levelized cost of ammonia (LCOA) by about 5.8%, primarily by lowering green hydrogen production costs. This cost reduction results from downregulating hydrogen production during periods of high electricity prices. By contras... [more]

The integrated recovery of heat and mechanical work has gained increasing importance in process integration due to the strong thermodynamic coupling between temperature and pressure changes in many industrial systems. This work presents a rigorous framework for the simultaneous synthesis of Work and Heat Exchanger Networks (WHEN), in which heating, cooling, compression, expansion, throttling, and pumping are optimized in a coordinated manner. The problem is formulated using Generalized Disjunctive Programming (GDP), allowing the explicit representation of alternative thermodynamic paths, phase-dependent behavior, and logical equipment choices. Process streams are defined by supply and target states, while only bounds are imposed on intermediate pressures, temperatures, and flow rates. Streams may change classification between hot and cold multiple times and may undergo several phase transitions.Rigorous thermodynamic correlations obtained from Aspen HYSYS are embedded in the optimizati... [more]

Black-box optimization (BBO) deals with problems involving functions that are either unknown, imprecise, or costly to evaluate. Current BBO methods encounter multiple challenges, such as high computational demands from excessive function evaluations, difficulties in handling complex constraints, lack of theoretical convergence guarantees, and unstable performance due to significant variations in solution quality. This work presents a machine learning-powered feasible path (MLFP) framework for general BBO problems involving complex constraints. An adaptive sampling strategy is first proposed to explore optimal regions and pre-filter potentially infeasible points, thereby reducing the number of evaluations. Machine learning algorithms are utilized to build surrogates for black-box functions. The feasible path algorithm is integrated to accelerate theoretical convergence by updating only independent variables instead of all variables. Computational experiments demonstrate that MLFP can ra... [more]

Deep Kernel Learning (DKL) has emerged as a powerful framework for Bayesian Optimization (BO), via combining expressive representation learning models with typical Gaussian Processes (GPs) surrogate models. However, conventional DKL typically relies on weight-based feature extractors (e.g., multilayer perceptrons (MLPs)), which often lack interpretability and may suffer from overfitting under data scarcity or training instability, potentially leading to a degraded uncertainty quantification in GP models. Grounded in the Kolmogorov-Arnold representation theorem, this paper proposes a novel DKL-KAN framework that employs Kolmogorov-Arnold Networks (KANs) as adaptive feature extractors, formulating a DKL-KAN surrogate model. Unlike MLPs, the KANs learn data-driven univariate functions, yielding more sample-efficient and stable representations for regression under limited data regimes. Followed by the GP, the DKL-KAN facilitates end-to-end learning of expressive latent representations whil... [more]

Cryogenic air separation units (ASUs) are the primary industrial technology for producing high purity oxygen, nitrogen, and argon gases at commercial scale. Cryogenic ASUs are large consumers of electricity, making them ideal candidates for market-driven operation research in today's volatile and uncertain manufacturing environments. To maximize profitability, ASU operation must dynamically adapt to changing market conditions as they evolve. This work explores the implementation of a rolling-horizon scheduling (RHS) strategy for the real-time market-driven operation of a high-dimensional ASU model with inventory, responding to uncertainty in future plant demand and electricity price forecasts by periodically rescheduling in response to updated market information. A dynamic latent variable-based surrogate model (LV-SM) is used within the scheduling framework as a computationally efficient substitute for an existing first-principles-based ASU model. Results show that RHS and plant invent... [more]

Simulated Moving Bed (SMB) chromatography is widely used for a variety of separations, yet, when applicable, these systems are typically operated using offline optimization strategies. Over time, process degradation and unforeseen disturbances may cause SMB units to deviate from the calculated optimal conditions, reducing overall performance. Real-Time Optimization (RTO) offers a promising solution by continuously monitoring and adjusting operating conditions to maintain optimal performance, despite such perturbations. However, experimental implementation of RTO in industrial SMB processes is costly and requires significant interdisciplinary coordination.To address this challenge, a virtual framework is proposed for the preliminary development of a model-based RTO system. The methodology employs a virtual plant-model pair, in which a representative plant model generates in silico experimental data, while a structurally distinct predictive model reproduces these results. Structural mism... [more]

Efficient deployment of carbon capture, utilisation, and storage (CCUS) within industrial clusters requires coordinated CO2 allocation under economic, technical, and environmental constraints, particularly when CO2 availability is limited. This paper presents a centralised optimisation framework for allocating captured CO2 from nine industrial sources to six utilisation and storage sinks within an industrial park in Qatar. A multi-objective mixed-integer linear programming (MILP) model is developed to minimise total system cost while accounting for capture, purification, transport, and utilisation processes, and enforcing an environmental feasibility constraint to ensure net CO2 reduction. The model is evaluated under four scenarios: a baseline case with sufficient CO2 to satisfy all sink demands, and three scarcity scenarios in which 15%, 25%, and 35% of total source emissions are available. Results show that under scarcity, allocations prioritise large EOR sinks supplied by high-volu... [more]

In this work, the synthesis, design and optimization of a quaternary double dividing wall distillation column (QDDWC) is presented. The effect of the feed composition over the performance of this intensified configuration is studied. The synthesis and design of the QDDWC takes place using as basis a conventional direct sequence for the separation of a n-butane/n-pentane/n-hexane/n-heptane mixture. The column is tested for three molar feed compositions: 40/10/10/40, 25/25/25/25, and 10/40/40/10. The configurations are optimized through a multiobjective genetic algorithm to simultaneously minimize the total heat duty and the number of stages. According to the results, the proposed structure allows savings in heat duty up to 59% but requiring up to 28% more stages than the conventional sequences.

Machine learning (ML) reduces reliance on computationally expensive first-principles simulation while capturing complex nonlinear behaviors. However, poor extrapolation, overfitting, limited interpretability, and lack of strict consistency with governing laws limit the use of ML models in process applications. Current methods for learning optimization policies also struggle with constraint satisfaction and optimality guarantees. Approaches such as physics-informed neural networks (PINNs) incorporate constraints "softly" and do not ensure strict constraint enforcement-an issue that can be particularly detrimental in safety-critical applications, where even minor violations may lead to unsafe or infeasible decisions. To resolve these issues, we develop an ML framework with a differential projection layer that allows computationally efficient process modeling, parameter estimation, and nonlinear optimization with feasibility and optimality guarantees. The framework is general in a sense t... [more]

Distributed deployment of process systems can benefit from modularity and shared components across multiple variants, reducing both manufacturing costs and engineering effort. Process family design formalizes this idea by simultaneously optimizing a family of process variants while determining a shared platform of common components. This results in a large-scale mixed-integer nonlinear program (MINLP) that couples nonlinear process models with discrete platform-allocation decisions. In this work, we solve the process family design MINLP using a progressive hedging (PH)-based decomposition strategy that exploits its block-angular structure. To improve convergence for this nonconvex problem, we introduce dynamic gradient-based penalty updates, a decoupled primal-dual strategy via separate PH runs, and parallel optimization-based bounds tightening of first-stage variables. Computational results on a water desalination case study demonstrate that the proposed approach improves solution qua... [more]

Dimethyl ether (DME) is a promising clean fuel and chemical intermediate, yet its synthesis from synthesis gas remains highly sensitive to both catalyst formulation and operating conditions. In this work, a data-driven framework is developed that combines machine learning surrogate modeling with multi-objective optimization to support systematic decision-making in DME synthesis. The novelty lies in the systematic comparison of different optimization approaches applied to an identical machine learning surrogate model for DME synthesis, thereby highlighting their respective strengths and limitations as decision-support tools under limited-data conditions. A dataset compiled from published literature includes catalyst composition, preparation methods, physicochemical descriptors, and operating conditions, with CO conversion and DME selectivity as performance indicators. After data preprocessing, feature analysis using correlation analysis and principal component analysis (PCA) is applied... [more]