Modern semiconductor manufacturing requires extreme precision as yield margins narrow in the "More-than-Moore" era. While physics-based models (PBMs) provide high-fidelity insights into plasma etching, their computational intensity-often requiring hours per simulation-renders them impractical for direct iterative optimization. This work demonstrates a hybrid framework that utilizes data-driven surrogate models to enable rapid, cost-effective process optimization. A 2D axisymmetric fluid model of an inductively coupled O2 plasma (ICP) reactor was developed to generate a training dataset for two neural architectures: a Multi-Layer Perceptron (MLP) and a Kolmogorov-Arnold Network (KAN). These surrogates predict radial etching rates across a wide operating window of power, pressure, gas flow, and bias voltage. By replacing the expensive PBM with these high-speed surrogates, derivative-free optimization algorithms (Nelder-Mead and Powell) successfully identified a profit-maximizing operatin... [more]

As industries pursue more sustainable and flexible manufacturing strategies, batch processes continue to play a vital role across a wide range of applications. Batch operations offer the ability to handle diverse feedstocks and accommodate varying product specifications. These processes are broadly used in sectors such as pharmaceuticals, specialty chemicals, food production, and bioprocesses, where precise control over reaction conditions and product quality is essential. However, maintaining optimal conditions in a batch process can be challenging due to the minimal opportunities for mid-batch interference. This work focuses on a real industrial batch process that frequently sees batches with poor yields resulting in large financial losses. Despite utilizing a mid-infrared spectrometer analyzing the batch medium in real-time, the reduced product accumulation observed in faulty batches is not evident until over a third of the batch time has passed, by which point the batch is not econ... [more]

Plastic injection molding is a widely used polymer-processing method. As the requirements for processing accuracy have become increasingly stringent, defect analysis in plastic injection molding is necessary to improve the product yield. Causal discovery has recently gained attention for defect analysis in many processes. Because injection molding is a spatial process involving the distribution of physical quantities, spatial autocorrelation should be considered. Although the linear non-Gaussian acyclic model (LiNGAM) is a well-known causal discovery method, it cannot properly model spatial autocorrelation. In this study, a new causal discovery method for a spatially autocorrelated dependent variable, referred to as the Causal Structure Search for the Spatial Autoregressive Model (CASSPAR), is proposed. It models the causal relationships among the observed points without prior knowledge of the spatial structure. The proposed method represents the causal relationships among the observed... [more]

Recurrent neural nets (RNNs) are now commonly used for the surrogate modeling of process systems, leading to better control and faster real-time optimization. However, when trained with small training data sets, most experiments show that RNNs exhibit poor generalization abilities outside the range of the training data space. Nonetheless, recent advances in deep learning research have shown that certain characteristics of the loss landscape of trained models, such as the flatness around the local minimum, tend to relate to generalization ability. This paper investigates this phenomenon for the case of RNN surrogates of the well-known Three Tanks case study, which is representative of many continuous processes. We trained a total of 200 LSTMs (long short-term memory networks) differing in initialization, architecture, and training dynamics on the same data of 500 samples. The number of model parameters ranges from 238 to 11, 353. We estimated the loss curvature of each trained model usi... [more]

In this work, we propose a methodology for monitoring the performance of model predictive controllers (MPCs). A sequence of binary classification machine learning models, organized in cascade, called Cascade Machine Learning Models (CMLM), is evaluated to give a diagnosis of the control conditions. The proposed methodology was assessed using two case studies: a benchmark problem (the van de Vusse reactor under nonlinear MPC, NMPC) and a simulated industrial debutanizer column under commercial MPC. The ML models evaluated were the Random Forest and the Multilayer Perceptron. The results show that the proposed approach outperforms both a single multiclass model and traditional MPC performance monitoring methodologies, while remaining adaptable and scalable to larger applications.

The convergence of artificial intelligence (AI) and chemical process systems engineering is creating unprecedented opportunities to transform current refineries from conventionally operated plants into intelligent, automated, and resilient systems. However, the practical deployment of AI in these complex industrial environments faces several critical challenges. First, most existing process datasets contain minimal labeled data, making it difficult to apply supervised learning techniques that require extensive annotations to generate meaningful insights. Furthermore, refinery data typically consists of high-dimensional, multivariate time series, which pose additional complexities in capturing temporal dynamics and system interactions. Traditional Fault Detection and Diagnosis (FDD) frameworks often struggle to address these complexities, lacking adaptability to evolving process conditions, scalability to large plant networks, and explainability in their diagnostic reasoning. To address... [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]

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]

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]

Reliable multimodal monitoring in crystallization processes remains challenging due to heterogeneous PAT signal quality, sensor drift, asynchronous sampling and nonstationary noise. This work presents a machine-learning-assisted fusion framework that integrates multimodal PAT alignment, estimation and physics-guided regularisation to generate coherent concentration and particle-size trajectories. A mechanistically informed simulation platform is developed to produce synthetic Raman, FTIR, FBRM and image-based crystal size data with realistically simulated drift, heteroscedastic noise, dropouts and distortion patterns. Sensor reliability is inferred through a Random Forest model trained on variance-normalised discrepancies and quality metrics, which allows the dynamic adjustment of channel contributions. Across modalities, the Random Forest achieves MAE values of 0.03-0.20 for probability-type indicators and shows stable explanatory power for variance-inflation factors on particle-size... [more]

Environmental sustainability is increasingly recognized as a critical consideration in pharmaceutical development, yet it is rarely incorporated at the scale of molecular-level design. This study introduces a strategy to predict cradle-to-gate indicators that can be flexibly incorporated into multiple early-stage molecular prioritization scenarios. A dataset of 150 pharmaceutical-relevant molecules was compiled, with each molecule described by structural descriptors, thermophysical properties, and ReCiPe endpoint indicators representing human health, ecosystem quality, and resource scarcity. A dual-branch multi-task model combining graph-based and descriptor-based representations was trained to predict these three endpoint indicators. Model performance was evaluated through validation metrics, local sensitivity analysis, and SHAP-based interpretability. A case study with solubility-based feasibility constraints was then used to illustrate how different sustainability weighting schemes... [more]

Microstructure segmentation is essential for quantitative materials analysis; however, supervised deep learning demands substantial annotation, whereas general-purpose foundation models such as the Segment Anything Model (SAM) offer limited domain-specific semantic control. This paper presents MatStudio, a human-in-the-loop framework for microstructure segmentation that is proposed and implemented end to end in this work. MatStudio couples an interactive workflow for batchwise micrograph annotation and model adaptation with a dual-head convolutional architecture and SAM-guided boundary refinement. The loop combines sparse supervision with SAM-assisted labeling, task-specific training, and iterative batch-level correction, typically converging within two to three cycles.. The network comprises a shared encoder initialized from a pretrained backbone and two decoders: a UNet-style segmentation head that jointly predicts class labels and pixelwise uncertainty, and a prototype branch that m... [more]

The increasing demand for energy, water, and chemical products signals the need for more sustainable and efficient process design methodologies. Traditional methods for conceptual process design constrains the exploration of novel and intensified process alternatives, as they rely on prior knowledge in defining the design space. Previous studies employing bottom-up approaches, such as phenomena building blocks (PBBs), suggest that the synthesis of complex bottom-up flowsheets remains computationally challenging and is thus limited to the synthesis of individual units of operation. This work proposes a bottom-up, data-driven framework for process synthesis and intensification based on phenomena building blocks (PBBs), in which process flowsheets are constructed from their underlying physical and chemical phenomena rather than conventional units of operation. The proposed framework introduces a phenomena-based text representation and data collection module. Furthermore, a sequence traini... [more]

Real-time monitoring of metabolite concentrations is critical for optimising bioprocess performance. While Raman spectroscopy offers a non-invasive solution, translating spectra into metabolite concentration estimates requires robust machine learning models. Foundation models such as TabPFN demonstrate exceptional predictive performance but suffer from high inference complexity when trained on large calibration datasets, hindering their use in real-time laboratory settings. This study proposes a batch Active Learning (AL) strategy to efficiently calibrate TabPFN using a minimal subset of data. We employ a weighted K-means clustering strategy that balances model uncertainty and dataset diversity to select the most informative calibration samples. We evaluated this method on a dataset of nearly 7, 000 Raman spectra covering eight substances. Our AL strategy achieved a mean R² score greater than 0.95 with approximately 1, 000 samples, significantly outperforming random sampling. Notably,... [more]

Large-scale mixed-integer programs (MIPs) typically cannot be solved by standard solvers with reasonable computational cost. MIP-heuristics decompose large-scale monolithic mixed-integer programs into polylithic programs such that they can be solved with reasonable computational cost at the price of loosing their optimality certificate. The decomposition is steered by hyperparameters that impact the solution quality and the computational cost diametrically. The proper selection of the hyperparameter values is a black-box optimization problem which is mostly solved by grid search or random search. In previous publications the authors proposed a novel hyperparameter optimization method based on Bayesian optimization and studied a use case from the PSE domain. Computational studies showed that the BO-based algorithm is superior for objective functions with few optimal solutions.This contribution generalizes the description of the MIP-Heuristic Optimization Problem (MIP-HOP) and the comput... [more]

Machine learning (ML) has attracted growing interest in process systems engineering for its potential in process design, synthesis, and optimization. By learning complex patterns from data, ML methods complement traditional first-principles modelling and heuristic approaches, particularly for conceptual process design and the exploration of alternatives. Although current text-based representations capture unit-level connectivity, they lack a holistic view of process intent, equipment hierarchy, and contextual information to guide learning and inference. Consequently, models trained on such linear token sequences tend to reproduce syntactic structure rather than underlying process reasoning, thus limiting interpretability and user control. In this work, we introduce a contextual framework for representing process flowsheet information in ML models that embeds process engineering logic directly into the model inputs. The approach combines a structured, text-based representation of proces... [more]

Anomaly detection is essential for ensuring the safe and efficient operation of chemical plants. Although many deep-learning-based methods have been proposed in recent years, their evaluation remains largely limited to synthetic benchmarks such as the Tennessee Eastman Process (TEP) [1]. While these simulators enable controlled and reproducible comparisons, they fail to capture the noise characteristics, operational complexity, and irregular fault dynamics of real industrial plants, leaving the practical generalizability of many methods unclear. In this work, we extend our earlier ESCAPE study [2] beyond water-based systems to industrially relevant chemical processes. We analyze multivariate time-series data from two continuously operated pilot-plant scenarios at the Technical University of Munich, namely n-butanol/water heteroazeotropic distillation and poly(oxymethylene) ether purification, whose datasets were recently published at NeurIPS 2025 [3]. Using the open-source TimeSeAD lib... [more]

Batch manufacturing is central to fine chemicals, pharmaceuticals, and bioprocessing. Its operation evolves across phases and recipes, which yields high-dimensional trajectories and strong batch-to-batch variability. Meanwhile, key quality-indicative variables are often measured offline and cannot be used as online model inputs. This work presents an integrated deep learning framework that unifies soft sensing and process monitoring in a single module using only process variables as inputs. A multi-scale Temporal Convolutional Network with multiple kernel sizes extracts complementary dynamic features from sliding windows. These features are concatenated and pooled into a compact representation that feeds two task branches. A variational autoencoder branch reconstructs the input window and provides fault monitoring signals via reconstruction deviation while regularizing the latent space through KL divergence. In parallel, a prediction branch estimates the quality-indicative variable dir... [more]

In this work, we present quantum reinforcement learning algorithms for process flowsheet synthesis. Particularly, we discuss the implementation of encoding strategies to improve the algorithmic scalability. Reinforcement learning (RL)-driven flowsheet synthesis techniques provide a promising approach for conceptual process design, in addition to traditional optimization-based methods. These RL-based strategies identify the optimal flowsheet configurations from a maximum set of available processing units, without requiring to pre-postulate an interconnected superstructure. However, the resulting combinatorial design space for RL can scale extensively with the increased number of available processing units, which can render the algorithms to be computationally intensive or even intractable. To address this challenge, our prior work has introduced a quantum-enhanced approach to RL-driven process synthesis. However, this algorithm was limited in its capacity to solve larger flowsheeting pr... [more]

The accurate prediction of pure component properties has consistently been a critical issue in fields such as chemical engineering, biomedicine, and environmental science. In recent years, end-to-end deep learning methods have shown significant improvement over traditional machine learning approaches. This is due to their ability to automatically learn task-relevant representations from raw molecular data. In addition to accurate property prediction, researchers have increasingly focused on how specific fragment structures influence molecular properties. However, existing fragmentation methods based on predefined rules and group libraries struggle to capture novel molecular structures, which hampers the development of new materials and drugs. To address these challenges, this work proposes a hierarchical molecular fragmentation method. This method can automatically segment molecules into multiple fragments containing key functional groups. Then a three-branch graph attention network wa... [more]

Accurate life cycle assessment (LCA) depends on robust characterization factors (CFs), which quantify impacts such as ecotoxicity through the integration of fate (FF), exposure (XF), and effect (EF) factors. While databases such as USEtox and Ecoinvent provide essential CFs, significant data gaps remain, particularly in ecotoxicity endpoints like hazardous concentration 50% (HC_50), which directly inform effect factor calculations. Existing machine learning models can predict such values, but they often lack interpretability, which limits trust and transparency in environmental modeling. To address this, a machine learning framework is applied that utilizes symbolic regression (SR) and genetic programming (GP) to predict missing HC_50 values from physicochemical descriptors. A dataset with 14 descriptors was used to train SR models capable of generating interpretable mathematical expressions that link chemical properties to HC_50 values. SR models were benchmarked against prominent bla... [more]

Addressing the urgent transition to low-carbon energy systems requires microgrids capable of locally coordinating electricity generation, storage, and flexible consumption. Their efficient integration calls for control strategies that are scalable, privacy-preserving, and robust to uncertainty. To address such a challenging control problem, this work proposes a decentralised Multi-Agent Reinforcement Learning (MARL) approach based on the Cross-Entropy Method (CEM) for the coordination of prosumers, equipped with renewable generation and vehicle-to-grid capabilities. To improve sample efficiency and robustness, the policy is warm-started using Behaviour Cloning (BC) from a classical Proportional-Integral-Derivative (PID) controller, resulting in a hybrid BC-CEM framework. The proposed method is evaluated in a realistic microgrid simulation with stochastic demand and real weather and generation profiles. Results show that BC-CEM accelerates convergence and achieves lower energy costs com... [more]

Digitalization, artificial intelligence, and autonomous experimentation are reshaping chemical process development by enabling data-driven system identification and model-based optimization. Despite these advances, mechanistic models remain a cornerstone for predicting chemical reaction behavior and supporting optimization. However, purely mechanistic models often exhibit limited predictive accuracy when key phenomena affecting kinetics, mass and energy transfer are not fully captured. To address limitations on kinetic modelling, a hybrid modelling framework is proposed in this work that integrates a lumped power-law kinetic model with a Gaussian Process (GP) residual model to predict the reaction rate across the experimental design space while quantifying the uncertainty of the predicted rate. The hybrid model is then coupled with multi-objective Bayesian optimization (MOBO) by employing a weighted-sum approach and an upper confidence bound acquisition function to guide experimental d... [more]

Biological systems are governed by coupled interactions between intracellular metabolism and bioreactor operation that span multiple time scales. Constraint-based metabolic models are widely used to describe intracellular metabolism, but repeatedly solving the optimization problem at each time step in dynamic models introduces numerical challenges related to infeasibility and computational efficiency. This work presents a multi-scale modeling framework that integrates genome-scale, constraint-based metabolic models with dynamic bioreactor simulations. Intracellular metabolism is described using positive flux variables in a parsimonious flux balance analysis, and the resulting embedded optimization problem is replaced by a neural network surrogate. The surrogate provides a smooth approximation of the embedded optimization mapping and eliminates repeated linear program solves during simulation. The approach is demonstrated for fed-batch fermentation of Escherichia coli, in which the surr... [more]

The identification of predictive mathematical model structures (i.e. set of model equations) is essential for the development of digital twin models of chemical reactor systems. Recent work demonstrated the use of artificial neural networks (ANNs) for kinetic model recognition in a conceptual batch reaction experimental system. In practical chemical processes, however, system behaviour is governed not only by reaction kinetics but also by reactor hydrodynamics and system thermodynamics. While a very recent study incorporated hydrodynamic effects, this work integrates the three aspects: reaction kinetics, reactor hydrodynamics, and system thermodynamics, to develop a general reactor modelling recognition framework. The framework, which comprises three modules: 1) model generator module; 2) data generation module; and 3) ANN classifier module, was applied to a case study of benzoic acid esterification in a Taylor vortex flow reactor system. Analysing the framework's sensitivity, results... [more]