Accurately predicting chemical reaction yields can accelerate reaction optimization by prioritizing promising conditions; however, the complexity of multicomponent reactions and the limited availability of high-quality datasets remain significant challenges. While machine learning has achieved substantial progress in molecular property prediction, reaction-level modeling requires representations that capture three-dimensional structures, intercomponent interactions, and mechanistically critical features. In this study, we investigate the effective extension of molecular pretraining to reaction yield prediction from two complementary perspectives. First, we apply a reaction-aware architectural design based on molecular representation models pretrained via partial denoising to multicomponent reactions. Each reaction component is encoded as a three-dimensional stereoisomer embedding and concatenated into a reaction-level representation, with multi-head attention modeling intercomponent de... [more]

The availability of temperature-dependent physicochemical property data forms the cornerstone of process simulation, optimization, and sustainable molecular and product design. However, a critical data gap persists, as experimental measurements are accessible for only a small subset of known chemicals. This renders experimental characterization resource-prohibitive, often compelling reliance on empirical estimation methods. Moreover, although many models offer single-point predictions at fixed temperatures, accurately modeling continuous temperature-dependent behavior remains challenging. Conventional methods frequently overlook intermediate variations, resulting in limited extrapolation capability. To overcome these limitations, we introduce a mechanism-guided hybrid modeling framework that integrates physical insights into data-driven models. This framework is built on two strategies. Strategy ? targets trend correction by generating a continuous representation from discrete single-p... [more]

A recurrent neural network (RNN) model for a CO2 methanation reactor was developed based on synthetic data generated from a validated mechanistic model of the same unit. The model was used to predict the main properties of the reactor - methane productivity and hotspot temperature - during a dynamic operation of the unit. The dynamic profile of feedstock availability was simulated taking into account the H2 flow that can be produced from PV-powered water electrolysis using solar irradiation profiles over a year in Milan, Italy. The dataset therefore consists of 366 data instances (one for each day), each composed of one datapoint per minute of sunlight. The best agreement between the predictions from the RNN and the target output values from the mechanistic model was found using a shallow RNN of 20 hidden-layer neurons, trained with a batch size of 10 and an 80/20 training-testing split. This showed that RNNs can constitute a reliable tool for dynamic surrogate modelling of energy conv... [more]

As the demand of freshwater continuous to rise, the water desalination can act as prominent solution to address the global water scarcity. However, significant environmental concerns arise as the process is mainly powered by fossil fuel, which contributes to greenhouse gas emissions and thus leading to global warming. This research intends to explore the potential of renewable energy sources to effectively reduce freshwater production cost. Specifically, it intends to estimate the power generated and freshwater production cost of a multi-effect distillation (MED) desalination process, powered by solar and wind energy sources in addition to a comparison against the performance of an MED process powered by fossil fuel based conventional steam boiler. The comparison of energy efficiency and freshwater production cost is conducted in two different UK locations: Wick in the north of Scotland and Watchet in the south of England. MED process model with both wind power and solar energy models... [more]

The conceptual design and performance of a novel Methylcyclohexane-Toluene-Hydrogen based chemical heat pump was studied using steady state simulations. The distillation operating parameters of the chemical heat pump were optimized to maximize the Coefficient of Performance based on heat quantity (COP) and its corresponding Coefficient of Performance based on electric work input (COPW) was calculated. The best operating temperature ranges of the endothermic and exothermic reactor are 200°C-225°C and 250°C-275°C respectively. An endothermic temperature of 200°C and an exothermic temperature of 250°C results in a COP of 0.1357 and a COPW of 13.3. By integrating this chemical heat pump with a vapor compression heat pump COP increased to 0.1445 while COPW reduced to 4.9.

Model adaptation to production environment is critical for reliable Machine Learning Operations (MLOps), less attention is paid to developing systematic framework for updating the ML models when they fail under drift. This paper compares the transfer learning enabled model update strategies including ensemble transfer learning (ETL), all-layers transfer learning (ALTL), and last-layer transfer learning (LLTL) for updating the failed feedforward artificial neural network (ANN) model. The flue gas differential pressure across the air pre-heater unit installed in a 660 MW thermal power plant is analyzed as a case study since it mimics the batch processes due to load cycling in the power plant. Updating the failed ANN model by three transfer learning techniques reveals that ETL provides relatively higher predictive accuracy for the batch size of 5 days than those of LLTL and ALTL. However, ALTL is found to be suitable for effective update of the model trained on large batch size (8 days).... [more]

Brazed aluminum heat exchangers (BAHX) are key cryogenic equipment, but their simulation is sensitive to phase change modelling. This work benchmarks ProSec, CO-ProSec Reaction, and Aspen EDR on an ethylene-chiller BAHX using ethane in thermosiphon mode. ProSec's interpolation scheme of tabulated data is validated against CO-ProSec Reaction's approach that employs a thermodynamic model for the evaluation of thermodynamic and physical properties; an initial 25% pressure drop gap is traced to different void-fraction models (slip vs homogeneous). Against Aspen EDR, default duty/quality are ~15% higher; differences are mainly due to nucleate-boiling treatment and distributor pressure drop modelling. Harmonizing options reduces duty mismatch to ~5%.

Bioprocess modelling, optimisation and scale-up are central components for improving sustainable manufacturing within pharmaceutical and chemical industries. However, developing accurate bioprocess digital twins remains a challenging process. Conventional mechanistic models are difficult to construct because of limited mechanistic understanding and large complexity of cellular metabolisms. While data-driven models have gained popularity, they often require large amounts of experimental data that is often time consuming to obtain and lack any quantitative description of the process. Hybrid modelling methods have emerged as promising alternatives however fail to provide physical insight to the root cause of model error. This work therefore presents a promising solution by developing a data-efficient symbolic regression (SR) based framework to enable the automated discovery of interpretable bioprocess models. A universal kinetic model backbone was used to capture overall process behaviour... [more]

The objective of the present work is to optimize the simulation of an electrodialysis reactor for the desalination and regeneration of multi-ionic wastewater with high salt contents and conductivities, within the framework in the Sustainable Development Goal 6 (clean water and sanitation) and remarking the Electrodialysis (ED) as a highly energy-efficient and sustainable technology. The mathematical modelling has been carried out by using a semiempirical model that involves an algebraic system of differential equations, including mass and charge balances (taking into account the ions present in the wastewater: Na?, Ca²?, Mg²?, Cl?, SO4²?, and HCO3?), and the total electrodialysis stack voltage considering ohmic drops (in the dilute and concentrate compartments), the potential of membrane in each cell pair, and the electrode potentials. In the simulation process, different theoretical and experimental parameters are necessary such as number of cells, membrane working areas, efficiency,... [more]

Geologic hydrogen has emerged as a promising low-carbon energy vector, but its sustainable recovery requires effective stimulation and production strategies. This study presents an integrated process-modeling framework for evaluating hydrogen extraction from hydraulically stimulated reservoirs. The framework combines fracture propagation, damage evolution, Darcy-scale multiphase flow, permeability-aperture dynamics, and a dual-porosity dual-permeability (DPDP) representation to simulate hydrogen production in fractured source rock systems. In addition to production dynamics, the framework tracks operational water and energy inputs and incorporates a simplified reaction-extent formulation to represent hydrogen generation under data-limited conditions. The model was benchmarked against published shale gas production results to evaluate its ability to reproduce fracture-controlled production behavior and was subsequently applied to a representative multi-well hydrogen development scenario... [more]

Reliable kinetic models are essential for the design, optimisation and operation of methanol synthesis reactors in power-to-X applications. However, parameter estimation is frequently performed without prior assessment of parametric identifiability or the information content of experimental conditions, often resulting in poorly constrained parameters and inefficient experimental campaigns. This study introduces a systematic, pre-calibration, information-driven framework for identifiability analysis and experimental design. The framework integrates local sensitivity analysis, structural and practical identifiability metrics and a Sequential Information-Driven Experimental Selection (SIDeS) strategy to guide experiment selection prior to parameter estimation. The methodology is applied to four literature kinetic models for methanol synthesis, spanning varying levels of mechanistic detail. A Sobol-sampled design space is first screened for feasibility and local information content, follow... [more]

In this work, artificial neural networks are used to represent advanced oxidation processes, such as electrochemical oxidation, electro-Fenton, and photoelectro-Fenton, for the degradation of two dyes. The effect of treatment time, initial dye concentration, and current density on the degradation percentage is studied. Additionally, a network is developed to include discrete variables, such as treatment type and dye type, as input features, enabling it to predict the system's performance across different technologies and pollutants. Operating conditions are optimized using the universal ANN as a surrogate model and the adaptive differential evolution algorithm to maximize dye removal efficiency. According to the results, after optimizing the architecture of the artificial neural networks using Bayesian optimization, deviations of 3.9% or less are obtained for purple RL removal predictions, while for Green A, deviations of 8% or less are obtained. These models may serve as a basis for m... [more]

This work presents the dynamic optimization of glucose feed in cell cultivation considering the balance between productivity and impurity generation. We first developed a mechanistic model considering cell growth promotion by glucose and cell growth inhibition by osmolarity for a newly developed, high-productivity CHO-MK cell line. For model development, fed-batch cultivation experiments were conducted at a 250 mL scale under three different glucose feeding profiles. Results from a single-objective dynamic optimization, using the glucose feed profile as a design variable, were compared to those from multi-objective problem settings with varying weights assigned to productivity and final impurity concentrations. Simulation results suggested different glucose feed profiles depending on the priority given to the mAb and impurities, where the main difference was in the generated viable cell density profiles. Productivity-focused profiles employed a low-high-intermediate feeding strategy, i... [more]

In the field of biochemical process design, the accurate modeling of microbial growth is essential for the development and optimization of biological reactors used in the production of high-value compounds. Achieving this objective requires a detailed understanding of how environmental factors-such as pH and nutrient availability-influence microbial dynamics across the four distinct growth phases: lag, exponential, stationary, and death. Traditionally, reactor design relies heavily on the Monod model, which provides a simplified representation of microbial growth, focusing primarily on the exponential phase under constant operating conditions (1). However, this model presents substantial limitations when applied to dynamic environments where key parameters vary over time. To overcome these constraints, the present study proposes a data-driven modeling approach using a multilayer perceptron (MLP) artificial neural network for the prediction of microbial growth trajectories under varying... [more]

Magnetic Inductive Swing Adsorption (MISA) is a carbon dioxide capture process similar to Temperature Swing Adsorption that uses direct electromagnetic heating instead of classic heating systems for the regeneration step of the process. However, the lack of validated dynamic models hinders process optimization. This work introduces an open-source MISA model in the IDAES framework, incorporating Specific Absorption Rate (SAR) physics (SAR ? B²) to capture electromagnetic heating. Binary Sips isotherm parameters for Fe3O4@HKUST-1 were fitted to experimental data, achieving high statistical agreement (R2 > 0.996, RMSE < 0.022 mol/kg). Comprehensive validation was performed against adsorption isotherms, dynamic breakthrough curves, and desorption profiles. The model predicts breakthrough time with only 9% error and saturation time with 6% error. Crucially, the coupled thermal transport and SAR heating model capture temperature evolution during desorption within 5% error across all field st... [more]

Recent advances in computer-aided retrosynthesis (CAR), flow chemistry, and continuous manufacturing collectively offer new opportunities to enable environmentally sustainable development and manufacturing practices across the pharmaceutical development and manufacturing value chain. However, the implementation of these methods and technologies remains scattered and fragmented, preventing full realization of their potential to address one of the most urgent needs in the pharmaceutical and related sectors. This work introduces a holistic digital framework for the design and optimization of an end-to-end manufacturing process for paracetamol (acetaminophen). The framework integrates Green-by-Design synthetic and purification routes of the active pharmaceutical ingredient (API) aims to deliver cost efficiency and robust quality, safety, and environmental sustainability assurance. The approach integrates AI-driven CAR with plant wide modelling, Techno-Economic Analysis (TEA), and prospecti... [more]

Decisions taken throughout a bioprocess lifecycle are often guided by heuristic knowledge that is difficult to summarize and sort, scattered across heterogeneous tools and documents, and partly retained as tacit expert mental models alongside fragmented computational models. This fragmentation remains a central barrier to reproducibility, transparent provenance, and systematic reuse of prior learning across comparable development projects. In this paper, it is argued that a key missing link toward Bioprocessing 5.0 is the digitalization of FAIR knowledge through a Cognitive Digital Thread that couples semantic knowledge graphs with AI methods to connect experimental data, protocols, workflows, and decision rationale with mathematical models and digital twins in a machine-actionable and auditable manner. A digitalization roadmap is outlined as a sequence of capability stages-from local device and data integration, to reproducible workflow execution and metadata capture, to semantic know... [more]

Single-cell protein (SCP) fermentation efficiently converts carbohydrates into high-protein food products but typically relies on purified glucose. Better understanding of SCP growth on mixed sugar substrates could allow for use of lower-cost, less processed sugars and waste-derived feedstocks. Glucose-fructose mixtures are particularly relevant, as these are the main sugars in sucrose hydrolysates and are also common in many food and beverage waste streams. In this study, the growth of Fusarium venenatum on glucose-fructose mixtures was investigated experimentally and modelled using a lagged dual-substrate Monod framework incorporating inhibition of fructose growth by glucose. Batch fermentations were conducted at a fixed total sugar concentration (15 g/L) with four different initial substrate compositions. Model parameters were estimated using both single-experiment and multi-experiment fitting strategies using differential evolution and wild bootstrap uncertainty analysis. A shared... [more]

The formulation of oral solid dosage forms can have a significant impact on drug bioavailability, particularly for poorly soluble drugs. However, traditional formulation development relies heavily on extensive experimental testing, which limits its efficiency and effectiveness in oral drug product design. In this study, we present an integrated framework to support rational formulation design and exploration of optimal dosage regimens. This framework combines population balance-based tablet disintegration and dissolution modeling with physiologically based pharmacokinetic (PBPK) modeling to link critical material attributes (CMAs) with the pharmacokinetic response. The anticoagulant drug rivaroxaban is selected as a model compound for calibration and deployment of the framework, enabling systematic investigation of the effects of crystal size distribution (CSD) and tablet porosity on in vivo performance. The results demonstrate that CSD has a pronounced impact on in pharmacokinetics, w... [more]

Probabilistic design spaces enable pharmaceutical manufacturers to balance regulatory compliance and operational efficiency. In this context, the "edge-of-failure" separating compliant from non-compliant operation is not a fixed line, but an uncertain region driven by model parameter uncertainty. Traditional methods typically map this probability of failure across the whole design space, which is a computationally expensive task. We propose a more direct approach, reformulating the problem using quantile functions to search for a deterministic boundary at a desired confidence level. These functions are approximated via Gaussian process surrogates using a novel adaptive sampling strategy. An industrial peptide acylation case study identifies the probabilistic design space using 13 experiments versus 18 from a traditional DoE; a 28% reduction. The framework quantifies trade-offs between operational robustness and yield optima.

In pharmaceutical process engineering, accurate prediction of tablet properties is crucial for ensuring product quality, optimizing manufacturing efficiency, and advancing sustainable production practices. This study presents a physics-informed neural network (PINN) framework for predicting the physical properties of pharmaceutical tablets from near-infrared (NIR) spectra. The PINN framework integrates revised Kubelka-Munk theory and physical constraints to ensure physically consistent predictions while requiring less training data than conventional artificial neural networks. Tablets were manufactured using acetaminophen and microcrystalline cellulose formulations with varying compositions and compression settings. The PINN framework successfully predicts critical quality attributes, including tensile strength, porosity, and density. It offers a data-efficient, interpretable solution for pharmaceutical tablet quality control.

Roller compaction is widely used in pharmaceutical manufacturing to improve powder flowability and enable robust tablet production. Although often treated as producing a homogeneous granule population, ribbons may undergo splitting during compaction, generating structurally distinct granules that affect downstream tableting. This study investigates the impact of ribbon splitting on tablet critical quality attributes (CQAs) for 10% and 20% acetaminophen (APAP) formulations. Reduced-order models (ROMs) proposed by Bachawala et al. [1] were applied to predict tablet density, elastic recovery, tensile strength, and tablet weight under split and non-split conditions. Although ribbon splitting alters the granule size distribution (GSD) and ribbon density, tablet CQAs such as tensile strength, elastic recovery, and tablet density are accurately predicted by the existing ROM framework, provided that GSD and ribbon density are known. In contrast, tablet weight predictions deteriorate when split... [more]

Real-time monitoring of bioprocesses is hindered by sparse, heterogeneous measurements of key biological states, such as biomass, substrate, and product concentrations. Extended Kalman Filter (EKF)-based soft sensors offer a physics-grounded solution but are sensitive to limited observability, sensor bias, and process-model mismatch-conditions common in industrial fermentations. This work proposes a Levelized Hybrid Estimation Architecture (LHEA) that systematically enhances physics-based state estimation through increasing robustness and adaptivity while preserving model transparency and regulatory interpretability. The approach is evaluated using the KTB1 benchmark simulation model for continuous lovastatin production under an industrially realistic, cost-constrained sensor configuration combining dissolved oxygen, biomass proxy, volume measurements, and sparse HPLC product assays. Three estimator levels are investigated: (L1) a baseline EKF, (L2) a bias-augmented EKF for sensor-drif... [more]

The decarbonization of pharmaceutical manufacturing is critical for achieving the industry's net-zero targets, and electrochemistry is emerging as a promising green technology that could play a key role in this transition. This work evaluates a continuous-flow electrochemical route for ibuprofen synthesis through electrochemical carboxylation of 1-chloro-(4-isobutylphenyl) ethane as a low-carbon alternative that can be directly coupled with renewable electricity. Experimental studies have demonstrated the selective formation of ibuprofen using a silver cathode in the ionic liquid N-methyl-N-propylpiperidinium bis(trifluoromethanesulfonyl)imide (PP13 TFSI). While the reaction mechanism is based on laboratory-scale, batch experiments, this study develops a conceptual design and scale-up methodology for the continuous route to provide an evaluation of the industrial feasibility of this electrochemical pathway through a rigorous plant-wide simulation in AVEVA® Process Simulation. Global se... [more]

The volumetric gas-liquid mass transfer coefficient (kLa) is a key performance parameter in stirred tank reactors and is commonly determined through extensive experiments across the operational space. This work presents an automatic, closed-loop framework for kLa determination based on model-based design of experiments (MBDoE), in which agitation and aeration inputs are adapted in real time.During each experiment, dissolved oxygen data is collected and used to estimate the parameters of a Van't Riet kLa relation. The parameter uncertainty is quantified using the covariance matrix, and the experiments are iteratively selected based on D-optimality or E-optimality MBDoE, until a threshold of RSEi < 0.15 is reached for all parameters. The MBDoE approach is evaluated through repeated runs and compared against random designs, full factorial (FF) design, and a full grid design.The results demonstrate that the closed-loop MBDoE framework can significantly reduce the number of experiments requ... [more]