Deep learning faces significant challenges in efficiently training large-scale models. These issues are closely linked, as efficient training often depends on precise and computationally feasible gradient calculations. This work introduces innovative methodologies to improve deep learning network (DLN) training in complex systems. A novel approach to DLN training is proposed by adapting the block coordinate descent (BCD) method, which optimizes individual layers sequentially. This is combined with traditional batch-based training to create a hybrid method that harnesses the strengths of both techniques. Additionally, the study explores Iterated Control Random Search (ICRS) for initializing parameters and applies quasi-Newton methods like L-BFGS with restricted iterations to enhance optimization. By tackling DLN training efficiency, this contribution offers a comprehensive framework to address key challenges in modern machine learning. The proposed methods improve scalability and effect... [more]

The efficient separation of carbon dioxide (CO2) and methane (CH4) is crucial for chemical processes, including biogas upgrading and natural gas purification. Metal-organic frameworks (MOFs) have gained significant attention as promising adsorbents for these processes due to their high porosity and tunable structures. Estimating the adsorption capacity of MOFs is essential for screening high performing adsorbents. While molecular simulations are commonly used to estimate the adsorption capacities, their computational intensity acts as a bottleneck in screening MOF adsorbents. In this study, we propose a machine learning (ML)-based framework for the high-throughput prediction of adsorption isotherms for CO2 and CH4 in MOFs. A graph neural network (GNN) model was developed to predict adsorption capacities, effectively replacing the time-consuming molecular simulations. The GNN model processes the structural graphs of MOFs, capturing their spatial configurations, such as surface structure... [more]

Physics-Informed Symbolic Regression (PISR) offers a pathway to discover human-interpretable solutions to partial differential equations (PDEs). This work investigates three fitness metrics within a PISR framework: PDE fitness, Bayesian Information Criterion (BIC), and a fitness metric proportional to the probability of a model given the data. Through experiments with Laplace’s equation, Burgers’ equation, and a nonlinear wave equation, we demonstrate that incorporating information theoretic criteria like BIC can yield higher fidelity models while maintaining interpretability. Our results show that BIC-based PISR achieved the best performance, identifying an exact solution to Laplace’s equation and finding solutions with R2-values of 0.998 for Burgers’ equation and 0.957 for the nonlinear wave equation. The inclusion of the Bayes D-optimality criterion in estimating model probability strongly constrained solution complexity, limiting models to 3-4 parameters and reducing accuracy. Thes... [more]

Balancing economic and environmental goals in industrial applications is critical amid challenges like climate change. Multi-objective optimization (MOO) and multi-criteria decision analysis (MCDA) are key tools for addressing conflicting objectives. MOO generates viable solutions, while MCDA selects the optimal option based on key performance indicators such as profitability, environmental impact, safety, and efficiency. However, large datasets pose a challenge in selecting the preferred solution during the MCDA process This study introduces a novel machine learning-enhanced MCDA framework and applies the method to analyze decarbonization solutions for a European refinery. A stage-wise dimensionality reduction method, combining AutoEncoders and Principal Component Analysis (PCA), is applied to simplify high-dimensional datasets while preserving key spatial features. Geometric analysis techniques, including Intrinsic Shape Signatures (ISS), are employed to refine the identification of... [more]

The accumulation of heavy polynuclear aromatics (HPNAs) in hydrocracking units (HCUs) poses significant challenges to catalyst performance and process efficiency. This study proposes an integrated machine learning framework that combines ridge regression, K-means, and long short-term memory (LSTM) neural networks to predict HPNA formation, enabling proactive process management. For the training phase, weighted average bed temperature (WABT), catalyst deactivation phase—clustered using unsupervised K-means clustering—and hydrocracker feed (HCU feed) parameters obtained from laboratory analyses are utilized to capture the complex nonlinear relationships influencing HPNA formation. In the simulation phase, forecasted WABT values are generated using a ridge regression model, and future HCU feed changes are derived from planned crude oil blend data provided by the planning department. These forecasted WABT values, predicted catalyst deactivation phases, and anticipated HCU feed parameters s... [more]

The rapid advancement of computational drug discovery has enabled the generation of vast virtual libraries of promising drug candidates. However, evaluating the synthetic accessibility (SA) of these compounds remains a critical bottleneck. While computer-aided synthesis planning (CASP) tools can provide synthesis routes to the candidate, their computational demands make them impractical for large-scale screening. Existing rapid SA scoring methods, struggle to generalize to out-of-distribution molecules and do not account for economic viability. To address these challenges, we present MolPrice, an accurate and reliable price prediction tool. By introducing a novel self-supervised learning approach, MolPrice achieves robust generalization to diverse molecular structures of various complexities. Our comprehensive analysis of model architectures and molecular representations reveals that substructure-based features strongly correlate with market prices, supporting the relationship between... [more]

This study tackles the challenge of accurate yield prediction in fluid catalytic cracking (FCC) units by comparing conventional supervised regression with time series forecasting methods using industrial data collected from the distributed control system (DCS) of an FCC plant. We introduce a shifted forecast paradigm that preserves temporal relationships between predictors and targets. Our preprocessing pipeline, which employs trimmed mean smoothing, addresses common industrial data challenges. Results demonstrate that the forecasting approach significantly outperforms supervised regression, achieving a mean absolute percentage error (MAPE) of 1.56% for 3-hour shifted predictions compared to 6.20% for supervised regression. The model maintains robust performance even with extended shifts during predictions, showing an MAPE of 3.55% for 14-day forecasts. This research provides valuable insights for implementing predictive analytics in industrial FCC operations, demonstrating the superio... [more]

Accurate mechanistic models provide valuable physical insight and are crucial for efficient process scale-up and optimisation, but their identification requires lengthy experimental data collection, model construction, validation and discrimination. Traditional black-box machine learning transfer methods leverage prior knowledge but lack interpretability and physical insights. To address this, we propose a novel approach using artificial neural network feature attribution to automatically locate corrections and symbolic regression to make structural modifications to an inaccurate or low-fidelity mechanistic model. In a comprehensive in-silico case study, the framework adapted a kinetic model from one biochemical system to a different but related one, enhancing predictive accuracy. Integrated within an iterative model-based design of experiments routine, it minimised the number of new experiments required. The study also discusses the impact of the inductive bias trade-off and alternati... [more]

As large language models have shown remarkable capabilities in conversing via natural language, the question arises in which way LLMs could potentially assist chemical engineers in research and industry with domain-specific tasks. We generate dynamic chemical reactor models in Modelica code format from textual descriptions as user input. We fine-tune Llama 3.1 8B Instruct on synthetically generated Modelica code for different reactor scenarios. We compare the performance of our fine-tuned model to the baseline Llama 3.1 8B Instruct model as well as GPT4o. We manually assess the models' predictions regarding the syntactic and semantic accuracy of the generated dynamic models. We find that considerable improvements are achieved by the fine-tuned model with respect to both the semantic and the syntactic accuracy of the Modelica models. However, the fine-tuned model lacks a satisfactory ability to generalize to unseen scenarios compared to GPT4o.

Many small and medium enterprises (SME) often fail to fully utilize the data they collect due to a lack of technical expertise. The ecoKI platform, a low-code solution that simplifies machine learning application for SMEs, showed a promising answer to the challenge. This study explores the application of ecoKI platform to design process monitoring tools for waste sorting plants. NIR separator data were processed through ecoKI’s building blocks to train two neural network architectures—MLP and LSTM—for predicting NIR separation efficiency. The results showed that the models accurately predicted NIR output and effectively identified regions where NIR separation performance declined, demonstrating the potential of data-driven approaches for real-time performance monitoring. This work highlights how SMEs can leverage existing data for operational efficiency and decision-making, offering an accessible solution for industries with limited machine learning expertise. The approach is adaptable... [more]

This study focuses on biomanufacturing case study, i.e. Lovastatin production, employing a hybrid modeling framework that combines mechanistic and data-driven approaches. A time-series dataset was generated using the KT-Biologics I (KTB1) plantwide model, a dynamic simulation of continuous biomanufacturing. The dataset captures critical parameters such as nutrient concentrations and API production. The AI-DARWIN framework was used to develop interpretable machine learning models with constrained functional forms, ensuring both accuracy and clarity. The resulting polynomial-based models reveal key relationships between process variables and system performance, bridging mechanistic insights with data-driven predictions. The models demonstrated reasonable accuracy showing minimal difference between the training and testing errors, highlighting their strong generalization. This work advances hybrid modeling in biomanufacturing by integrating plant-wide mechanistic simulations with interpre... [more]

As the demand for resources continues to grow, pipelines have become critical for transporting water, fossil fuels, and chemicals. Monitoring pipeline systems is essential, as leaks can lead to severe environmental damage and safety hazards. This study aims to develop a pipeline leakage detection system based on digital twin technology and Physics-Informed Neural Networks (PINNs). By embedding physical principles, such as the continuity and momentum equations derived from the Navier-Stokes equation, into the neural network's loss function, the model can predict pressure and flow dynamics with high accuracy while adhering to physical constraints. PINNs are particularly advantageous as they require minimal data, maintain physical consistency, and provide reliable interpretations, making them well-suited for addressing pipeline safety challenges. The model is designed to simulate fluid dynamics under normal operating conditions, with deviations in prediction errors signaling potential lea... [more]

Detailed assessment of fuel production processes at an early stage of a project is crucial to identify potential technical challenges, optimize efficiency and minimize costs and environmental impact. While process simulations often are either very rigid and accurate or very flexible and unprecise, informed decision making can only be maintained by establishing a detailed process model as early as possible within the project lifecycle while keeping relevant aspects of the process flexible enough. In this work, we present the development of a framework based on a dynamic interface between AspenPlus® process simulations and Python, enabling enhanced flexibility and automation for process modeling and optimization. This integration leverages the powerful simulation capabilities of AspenPlus® with the versatility of Python for data analysis and optimization, delivering significant improvements in workflow efficiency and process control. By utilizing the dynamic simulation data exchange with... [more]

In the domain of process systems engineering, developing software embedded with advanced computational methods is in great demand to enhance the kinetic comprehension and facilitate industrial applications. Polymer production, characterized by complex reaction mechanisms, represents a particularly intricate process industry. In this work, a scientific software is developed for polymerization modelling and simulation with insight on microscopic resolution. From a software architecture perspective, the software is built on a self-developed process modelling platform that allows flexible user customization. A specific design for polymer species with microscopic chain structure information is conducted. From an algorithm perspective, the software offers high-performance solution strategies for polymerization process modelling by utilizing advanced computation approaches. A Ziegler-Natta copolymerization is presented to demonstrate the software’s capability in capturing the microscopic stru... [more]

A new method for efficient process development is the direct scale-up from laboratory scale to production scale using mechanistic models [1]. The integration of the Quality by Design approach into this scale-up concept may prove beneficial for a variety of product- and process-related aspects. This paper presents a workflow for the digital twin-supported direct scale-up of processes and process plants, which integrates elements of the Quality by Design methodology. To illustrate the concept, the workflow is implemented for the example of an esterification reaction in a stirred tank reactor. Finally, benefits of the implementation of Quality by Design in the direct scale-up using digital twins regarding the product quality and the process development are discussed as well as its limitations.

In applications involving fine powders and small particles, the accuracy of numerical simulations, particularly those employing the Discrete Element Method (DEM) to predict granular material behavior, can be significantly affected by uncertainties in critical parameters. These uncertainties include the coefficients of restitution for particle-particle and particle-wall collisions, viscous damping coefficients, and other related factors. In this study, we use stochastic expansions based on point-collocation non-intrusive polynomial chaos to perform a sensitivity analysis of a fluidized bed system. We treat four key parameters as random variables; each assigned a specific probability distribution over a designated range. This uncertainty is propagated through high-fidelity Large Eddy Simulation (LES)-DEM simulations to statistically quantify its impact on the results. To effectively explore the four-dimensional parameter space, we analyze a comprehensive database comprising over 1,200 si... [more]

Random signals are crucial in chemical and process engineering, where industrial plants generate big data that can be used for process understanding and decision-making. This makes it necessary to unveil the underlying probability histograms from these signals with a finite number of bins. However, the search for the optimal number of bins is still based on empirical optimisation and general rules of thumb. In this work, we introduce an alternative and general method to unveil probability histograms. Our method employs a novel variable-order QMOM, which adapts automatically based on the relevance of the information contained in the random data. The number of bins used to recover the underlying histogram is found to be proportional to the information entropy, where a search algorithm is developed that generates bins and assigns probabilities to them. The algorithm terminates when no more significant information is available for assignment to the newly created nodes, up to a user-defined... [more]

Liquid-liquid extraction stages often deviate from equilibrium due to factors like insufficient mixing, making accurate efficiency modeling essential for process simulation. This study addresses the limitations of Aspen Plus (AP), which distorts equilibrium calculations by directly multiplying efficiency with the distribution coefficient. A modified Murphree efficiency definition, more suitable for liquid-liquid systems but absent in AP's Extraction Column module, was implemented using Aspen Custom Modeler (ACM). The custom multi-stage extraction column model replaces mole fractions with mole flows to better represent mass transfer and phase interactions, enhancing simulation accuracy when imported into AP. Two test cases validated the custom model's effectiveness. Test Case I, utilizing the UNIQ-RK thermodynamic model, compared the ACM model to AP's built-in module, revealing that the ACM model provides a more realistic representation of extraction processes under varying stage effici... [more]

Rapid and robust simulation of chemical production processes is critical to address core scientific questions related to process design, optimization, and sustainability. Efficiently solving a chemical process, however, remains a challenge due to their highly coupled and nonlinear nature. Graph abstractions of the underlying physical phenomena within unit operations may help identify potential avenues to systematically reformulate the network of equations and enable more robust convergence of flowsheets. To this end, we further refined a flowsheet graph-theoretic abstraction that consists of a mesh of interconnected variable nodes and equation nodes. The new network of equations is formulated at the phenomenological level agnostic to the thermodynamic property package by extending equation formulations widely used to solve multistage equilibrium columns. Decomposition of the graph by phenomena linearizes material and energy balances across the flowsheet by decoupling phenomenological n... [more]

The definition of strategies for operation of process networks is a key research focus in process systems engineering. This challenge is commonly formulated as a numerical constraint satisfaction problem, where most practical algorithms are limited to identifying inner approximations to the feasible operational envelope. Sampling-based approaches so far have only been developed for formulations that required coordinated operation of the units within the network. We propose a decomposition approach that enables decentralized operation for acyclic muti-unit processes by sampling. Our methodology leverages problem structure to decompose unit-wise and deploys surrogate models to couple the resultant subproblems. We demonstrate it on a serial, batch chemical reactor network. In future research, we will extend this framework to consider the presence of uncertain unit parameters robustly.

Vacuum (pressure) swing adsorption (V(P)SA) has received considerable attention in the past decades. Existing studies typically estimate vacuum pump energy consumption using an approximate constant energy efficiency or an empirical energy efficiency correlation, leading to inaccurate representation of realistic vacuum pump performance. In this paper an enhanced computational approach is proposed for simulation and optimisation of V(P)SA through simultaneous integration of realistic vacuum pump data and adsorption bed fluidisation limits. The computational results show that the developed prediction models accurately represent the actual performance curves of the vacuum pump. Incorporation of the vacuum pump prediction models and fluidisation constraints in V(P)SA optimisation leads to significantly different optimal solutions compared to when these factors are not considered.

By capturing carbon dioxide from biomass flue gases, energy processes with negative carbon footprint are achieved. Among carbon capture methods, the fluidized temperature swing adsorption (TSA) column is a promising low-pressure alternative, but it has been developed on small scales. This work aims to model, simulate and optimize a fluidized TSA multi-stage equilibrium system to obtain a cost estimate and a conceptual design for future process scale up. A mathematical model described adsorption in multiple stages, each with a heat exchanger, coupled to the desorption operation. The model was based on elementary macroscopic molar and energy balances, coupled to pressure drops in a fluidized bed designed to operate close to the minimum fluidization velocity, and coupled to thermodynamics of adsorption equilibrium of a mixture of carbon dioxide and nitrogen in solid sorbents (the Toth equilibrium isotherm was used). The complete fluidized TSA process has been optimized to minimize costs,... [more]

The optimization of the power conversion system, responsible for thermal-to-electrical energy conversion, for a pulsed fusion power plant is presented. A spherical tokamak is modelled as three heat sources, all pulsed, with different stream temperatures and available amounts of heat. A thermal energy storage system is considered in the design to compensate for the lack of thermal power during a dwell. Thermal storage enables continued power generation during a dwell and can avoid thermal transients in sensitive components like turbomachines. Multiple lower grade heat sources are integrated into the process through parallel preheating trains. The evaluation of a dynamic model of the power conversion system is used to define an objective function with multiple criteria. A bi-objective optimization problem is defined to investigate the trade-off between the size of the thermal energy storage system and the variability in turbine power output during a dwell. The set of non-dominated design... [more]

Solar Thermal Power Plants (STPPs) use solar energy for large-scale electricity production but face significant operational challenges. These include variations in solar radiation, cloud cover, electricity demand fluctuations, and the need for frequent shutdowns if energy storage is inadequate. Deciding an optimal STPP operating conditions is challenging due to these factors. While revenue maximization has been used as an objective in existing literature, current models are often static and fail to capture the dynamic nature of STPPs. In contrast, this work proposes a dynamic model-based revenue optimization approach that accounts for plant dynamics and operational constraints, such as solar radiation variability and changing electricity demand. The objective function is designed to maximize revenue while considering power generation and fluctuating electricity prices. A simulation model of 1 MWe hybrid solar thermal power plant in Gurgaon, India, featuring two solar fields—Parabolic T... [more]

Distillation is not only a widely-used but also an energy-intensive separation process, in which internals such as structured packings play an important role. Increasing mass transfer efficiency by designing improved structured packings in order to provide a large interfacial area while enabling low pressure drop is one promising approach to quickly reduce the energy requirements of vacuum distillation where low pressure drop is important for separation efficiency and thermal stability of the processed media. The current work presents an innovative method to optimize structured packings by means of constrained shape optimization on the basis of computational fluid dynamics simulations to minimize the pressure drop while maintaining a constant specific surface area. To solve the fluid dynamic optimization problem, a gradient-based local optimization algorithm in a continuous adjoint formulation is utilized. The shape optimization is applied for a commonly used Rombobak packing, and test... [more]