Soft sensors estimate process variables that are difficult or impossible to measure directly by using mathematical models and available sensor data, e.g., product concentrations. Machine learning-based approaches have become popular for soft sensing tasks. These approaches offer automatic modeling using historical process data but lack basic process information, such as the process topology. This can lead to (1) modeling of correlations instead of causation between process measurements, (2) model deterioration in deployment due to unseen process scenarios, and (3) large data requirements. To overcome these shortcomings, we propose a novel ML modeling approach incorporating the process topology into soft sensor models for improved spatio-temporal modeling. For this, we propose process topology-aware graph neural networks. We combine process topology and sensor data by representing process data in a directed graph and leverage these process graphs to train graph neural networks. Our method demonstrates enhanced model robustness, reduced data requirements, and more intuitive data representations compared to standard black-box machine learning modeling approaches. Overall, this work introduces a new paradigm for soft sensing by directly embedding process information into the data, paving the way for more efficient and reliable digital twin applications.

Large language model (LLM)-based multi-agent systems (MASs) are a recent but rapidly evolving technology with the potential to transform chemical engineering by decomposing complex workflows into teams of collaborative agents with specialized knowledge and tools. This review surveys the state-of-the-art of MASs within chemical engineering. While early studies demonstrate promising results, scientific challenges remain, including the design of tailored architectures, integration of heterogeneous data modalities, development of foundation models with domain-specific modalities, and strategies for ensuring transparency, safety, and environmental impact. As a young but fast-moving field, MASs offer exciting opportunities to rethink chemical engineering workflows.

Neural networks are widely used as surrogate models but they do not guarantee physically consistent predictions thereby preventing adoption in various applications. We propose a method that can enforce NNs to satisfy physical laws that are nonlinear in nature such as enthalpy balances. Our approach, inspired by Picard successive approximations method, aims to enforce multiplicatively separable constraints by sequentially freezing and projecting a set of the participating variables. We demonstrate our PicardKKThPINN for surrogate modeling of a catalytic packed bed reactor for methanol synthesis. Our results show that the method efficiently enforces nonlinear enthalpy and linear atomic balances at machine-level precision. Additionally, we show that enforcing conservation laws can improve accuracy in data-scarce conditions compared to vanilla multilayer perceptron.

Control structure design is an important but tedious step in P&ID development. Generative artificial intelligence (AI) promises to reduce P&ID development time by supporting engineers. Previous research on generative AI in chemical process design mainly represented processes by sequences. However, graphs offer a promising alternative because of their permutation invariance. We propose the Graph-to-SFILES model, a generative AI method to predict control structures from flowsheet topologies. The Graph-to-SFILES model takes the flowsheet topology as a graph input and returns a control-extended flowsheet as a sequence in the SFILES 2.0 notation. We compare four different graph encoder architectures, one of them being a graph neural network (GNN) proposed in this work. The Graph-to-SFILES model achieves a top-5 accuracy of 73.2% when trained on 10,000 flowsheet topologies. In addition, the proposed GNN performs best among the encoder architectures. Compared to a purely sequence-based approach, the Graph-to-SFILES model improves the top-5 accuracy for a relatively small training dataset of 1,000 flowsheets from 0.9% to 28.4%. However, the sequence-based approach performs better on a large-scale dataset of 100,000 flowsheets. These results highlight the potential of graph-based AI models to accelerate P&ID development in small-data regimes but their effectiveness on industry relevant case studies still needs to be investigated.

The chemical industry needs to undergo a significant transformation towards more sustainable and circular production systems. To guide this transformation, estimating the environmental impacts of chemical production at early product screening or development stages is highly desirable. This study leverages the molecular structure of the process products with graph neural networks (GNNs) for early-stage environmental impact approximation of chemical processes. Specifically, we use end-to-end GNN models to predict fifteen environmental impact categories, utilizing a CarbonMinds dataset of 51,905 processes producing 791 molecules produced in 91 countries, augmented with country-specific energy mix data. Our analysis begins with a comparison of Quantitative Structure-Property Relationship (QSPR) and GNN models for the climate change impact category. Specifically, we develop three different GNN models: (i) GNN with only molecular structure, (ii) GNN with molecular structure and additional geographical features, and (iii) GNN with molecular structure and additional energy mix features. The results indicate that the three GNN models show an improvement over the QSPR models. Furthermore, benchmarking our GNN models against the existing literature in the climate change impact category reveals that our models perform comparably. We then extend our approach by developing both single- and multi-task GNN models to predict all fifteen impact categories. The findings indicate that multi-task learning can improve model performance in complex environmental impact predictions compared to single-task GNNs. Therefore, we recommend using a multi-task GNN for predicting multiple impact categories, with single-task models applied to fine-tune performance on underperforming categories. Although our proposed approach shows improvements over previous models, the prediction of environmental impacts solely based on molecular information remains a rough approximation.

Reinforcement learning has shown some success in automating process design by integrating data-driven models that interact with process simulators to learn to build process flowsheets iteratively. However, one major challenge in the learning process is that the reinforcement learning agent demands numerous process simulations in rigorous process simulators, thereby requiring long simulation times and expensive computational power. We propose employing transfer learning to enhance the reinforcement learning process in process design. This study examines two transfer learning strategies: (i) transferring knowledge from shortcut process simulators to rigorous simulators, and (ii) transferring knowledge from process variational autoencoders (VAEs). Our findings reveal that appropriate transfer learning can significantly improve both learning efficiency and convergence scores. However, transfer learning can also negatively impact the learning process when there are substantial discrepancies in decision range and reward function. This suggests that pre-trained process data should match the complexity of the fine-tuning task.

ChatGPT is a powerful language model from OpenAI that is arguably able to comprehend and generate text. ChatGPT is expected to greatly impact society, research, and education. An essential step to understand ChatGPT’s expected impact is to study its domain-specific answering capabilities. Here, we perform a systematic empirical assessment of its abilities to answer questions across the natural science and engineering domains. We collected 594 questions on natural science and engineering topics from 198 faculty members across five faculties at Delft University of Technology. After collecting the answers from ChatGPT, the participants assessed the quality of the answers using a systematic scheme. Our results show that the answers from ChatGPT are, on average, perceived as “mostly correct”. Two major trends are that the rating of the ChatGPT answers significantly decreases (i) as the educational level of the question increases and (ii) as we evaluate skills beyond scientific knowledge, e.g., critical attitude.

The estimation of polymer properties is of crucial importance in many domains such as energy, healthcare, and packaging. Recently, graph neural networks (GNNs) have shown promising results for the prediction of polymer properties based on supervised learning. However, the training of GNNs in a supervised learning task demands a huge amount of polymer property data that is time-consuming and computationally/experimentally expensive to obtain. Self-supervised learning offers great potential to reduce this data demand through pre-training the GNNs on polymer structure data only. These pre-trained GNNs can then be fine-tuned on the supervised property prediction task using a much smaller labeled dataset. We propose to leverage self-supervised learning techniques in GNNs for the prediction of polymer properties. We employ a recent polymer graph representation that includes essential features of polymers, such as monomer combinations, stochastic chain architecture, and monomer stoichiometry, and process the polymer graphs through a tailored GNN architecture. We investigate three self-supervised learning setups: (i) node- and edge-level pre-training, (ii) graph-level pre-training, and (iii) ensembled node-, edge- & graph-level pre-training. We additionally explore three different transfer strategies of fully connected layers with the GNN architecture. Our results indicate that the ensemble node-, edge- & graph-level self-supervised learning with all layers transferred depicts the best performance across dataset size. In scarce data scenarios, it decreases the root mean square errors by 28.39% and 19.09% for the prediction of electron affinity and ionization potential compared to supervised learning without the pre-training task.

The process engineering domain widely uses Process Flow Diagrams (PFDs) and Process and Instrumentation Diagrams (P&IDs) to represent process flows and equipment configurations. However, the P&IDs and PFDs, hereafter called flowsheets, can contain errors causing safety hazards, inefficient operation, and unnecessary expenses. Correcting and verifying flowsheets is a tedious, manual process. We propose a novel generative AI methodology for automatically identifying errors in flowsheets and suggesting corrections to the user, i.e., autocorrecting flowsheets. Inspired by the breakthrough of Large Language Models (LLMs) for grammatical autocorrection of human language, we investigate LLMs for the autocorrection of flowsheets. The input to the model is a potentially erroneous flowsheet and the output of the model are suggestions for a corrected flowsheet. We train our autocorrection model on a synthetic dataset in a supervised manner. The model achieves a top-1 accuracy of 80% and a top-5 accuracy of 84% on an independent test dataset of synthetically generated flowsheets. The results suggest that the model can learn to autocorrect the synthetic flowsheets. We envision that flowsheet autocorrection will become a useful tool for chemical engineers.

The transformation toward renewable energy and feedstock supply in the chemical industry requires new conceptual process design approaches. Recently, deep reinforcement learning (RL), a subclass of machine learning, has shown the potential to solve complex decision-making problems and aid sustainable process design. However, its suitability in static process design still needs to be examined. We discuss the advantages and disadvantages of RL for process design. Then, we survey state-of-the-art research through three major elements: (1) information representation, (2) agent architecture, and (3) environment and reward. Moreover, we discuss perspectives on underlying challenges and promising future works to unfold the full potential of RL for process design in chemical engineering.

ReLU neural networks have been modelled as constraints in mixed integer linear programming (MILP), enabling surrogate-based optimisation in various domains and efficient solution of machine learning certification problems. However, previous works are mostly limited to MLPs. Graph neural networks (GNNs) can learn from non-euclidean data structures such as molecular structures efficiently and are thus highly relevant to computer-aided molecular design (CAMD). We propose a bilinear formulation for ReLU Graph Convolutional Neural Networks and a MILP formulation for ReLU GraphSAGE models. These formulations enable solving optimisation problems with trained GNNs embedded to global optimality. We apply our optimisation approach to an illustrative CAMD case study where the formulations of the trained GNNs are used to design molecules with optimal boiling points.

Microgels are cross-linked, colloidal polymer networks with great potential for stimuli-response release in drug-delivery applications, as their small size allows them to pass human cell boundaries. For applications with specified requirements regarding size, producing tailored microgels in a continuous flow reactor is advantageous because the microgel properties can be controlled tightly. However, no fully-specified mechanistic models are available for continuous microgel synthesis, as the physical properties of the included components are only studied partly. To address this gap and accelerate tailor-made microgel development, we propose a data-driven optimization in a hardware-in-the-loop approach to efficiently synthesize microgels with defined sizes. We optimize the synthesis regarding conflicting objectives (maximum production efficiency, minimum energy consumption, and the desired microgel radius) by applying Bayesian optimization via the solver “Thompson sampling efficient multi-objective optimization” (TS-EMO). We validate the optimization using the deterministic global solver “McCormick-based Algorithm for mixed-integer Nonlinear Global Optimization” (MAiNGO) and verify three computed Pareto optimal solutions via experiments. The proposed framework can be applied to other desired microgel properties and reactor setups and has the potential of efficient development by minimizing number of experiments and modeling effort needed.

The term hybrid modeling refers to the combination of parametric models (typically derived from knowledge about the system) and nonparametric models (typically deduced from data). Despite more than 20 years of research, over 150 scientific publications (Agharafeie et al., 2023), and some recent industrial applications on this topic, the capabilities of hybrid models often seem underrated, misunderstood, and disregarded by other disciplines as “simply combining some models” or maybe it has gone unnoticed at all. In fact, hybrid modeling could become an enabling technology in various areas of research and industry, such as systems and synthetic biology, personalized medicine, material design, or the process industries. Thus, a systematic investigation of the hybrid model properties is warranted to scoop the full potential of machine learning, reduce experimental effort, and increase the domain in which models can predict reliably.

Machine Learning (ML) techniques are encountered nowadays across disciplines, from social sciences, through natural sciences to engineering. However, teaching ML is a daunting task. Aside from the methodological complexity of ML algorithms, both with respect to theory and implementation, the interdisciplinary and empirical nature of the field need to be taken into consideration. This paper introduces the MachineLearnAthon format, an innovative didactic concept designed to be inclusive for students of different disciplines with heterogeneous levels of mathematics, programming, and domain expertise. The format is grounded in a systematic literature review and the didactic principles action orientation, constructivism, and problem orientation. At the heart of the concept lie ML challenges, which make use of industrial data sets to solve real-world problems. Micro-lectures enable students to learn about ML concepts and algorithms, and associated risks. They cover the entire ML pipeline, promoting data literacy and practical skills, from data preparation, through deployment, to evaluation.

Multiscale modeling of catalytical chemical reactors typically results in solving a system of partial differential equations (PDEs) or ordinary differential equations (ODEs). Despite significant progress, the numerical solution of such PDE or ODE systems is still a computational bottleneck. In the past, deep learning techniques have gained attention for developing surrogate models in chemical engineering. Also, hybrid models and physics-informed neural networks (PINNs) have been developed to integrate physical knowledge and data-driven approaches. However, it is often unclear how such modeling approaches compare for specific case studies. In this study, we investigate and compare state-of-the-art surrogate and hybrid models for the spatial evolution of the state variables in a packetbed reactor for methanol production. Firstly, we develop a tailored hybrid model based on PINNs, thereby seamlessly integrating physical knowledge and data. Secondly, we investigate a recently-developed time-series transformer model to learn the spatial evolution of the state variables. As a benchmark model, we train a traditional multilayer perceptron (MLP) and compare the models to a standard numerical integration technique. We achieve orders of magnitude in speedup using MLPs and PINNs when compared to classical ODE solvers, while maintaining high levels of accuracy in modeling the underlying system.

This “white paper” is a concise perspective of the potential of machine learning in the process systems engineering (PSE) domain, based on a session during FIPSE 5, held in Crete, Greece, June 27–29, 2022. The session included two invited talks and three short contributed presentations followed by extensive discussions. This paper does not intend to provide a comprehensive review on the subject or a detailed exposition of the discussions; instead its aim is to distill the main points of the discussions and talks, and in doing so, highlight open problems and directions for future research. The general conclusion from the session was that machine learning can have a transformational impact on the PSE domain enabling new discoveries and innovations, but research is needed to develop domain-specific techniques for problems in molecular/material design, data analytics, optimization, and control.

Developing Piping and Instrumentation Diagrams (P&IDs) is a crucial step during process development. We propose a data-driven method for the prediction of control structures. Our methodology is inspired by end-to-end transformer-based human language translation models. We cast the control structure prediction as a translation task where Process Flow Diagrams (PFDs) without control structures are translated to PFDs with control structures. We represent the topology of PFDs as strings using the SFILES 2.0 notation. We pretrain our model using generated PFDs to learn the grammatical structure. Thereafter, the model is fine-tuned leveraging transfer learning on real PFDs. The model achieved a top-5 accuracy of 74.8% on 10,000 generated PFDs and 89.2% on 100,000 generated PFDs. These promising results show great potential for AI-assisted process engineering. The tests on a dataset of 312 real PFDs indicate the need for a larger PFD dataset for industry applications and hybrid artificial intelligence solutions.

Computational Fluid Dynamics (CFD) is a powerful tool which can help with the geometry optimization of continuous milli-scale reactors, which often are highly complex devices. Attempting to perform this optimization by manually modifying and testing geometry configurations can however be tedious and computationally inefficient. Addressing this problem, we present a framework in which the CFD software COMSOL Multiphysics is coupled with the multi-objective Bayesian Optimization algorithm TSEMO (Thompson sampling efficient multiobjective optimization), implemented in MATLAB. The mixing element geometry of a Miprowa Lab millireactor is parameterized, and the framework automatically executes CFD simulations to minimize areas of stagnating flow and maximize the mixing performance. The framework is able to find Pareto-optimal reactor variations, and can easily be adapted for other devices and objectives.

Ionic liquids (ILs) are important solvents for sustainable processes and predicting activity coefficients (ACs) of solutes in ILs is needed. Recently, matrix completion methods (MCMs), transformers, and graph neural networks (GNNs) have shown high accuracy in predicting ACs of binary mixtures, superior to well-established models, e.g., COSMO-RS and UNIFAC. GNNs are particularly promising here as they learn a molecular graph-to-property relationship without pretraining, typically required for transformers, and are, unlike MCMs, applicable to molecules not included in training. For ILs, however, GNN applications are currently missing. Herein, we present a GNN to predict temperature-dependent infinite dilution ACs of solutes in ILs. We train the GNN on a database including more than 40,000 AC values and compare it to a state-of-the-art MCM. The GNN and MCM achieve similar high prediction performance, with the GNN additionally enabling high-quality predictions for ACs of solutions that contain ILs and solutes not considered during training.