Digital Chemical Engineering最新文献

{"title":"Self-learning multichannel olfactory display","authors":"Philipp Müller ,&nbsp;Oleg Špakov ,&nbsp;Jarmo Verho ,&nbsp;Timo Salpavaara ,&nbsp;Mariaana Savia ,&nbsp;Atte Sinkkonen ,&nbsp;Venla Kamppari ,&nbsp;Jussi Rantala ,&nbsp;Pasi Kallio ,&nbsp;Veikko Surakka","doi":"10.1016/j.dche.2026.100293","DOIUrl":"10.1016/j.dche.2026.100293","url":null,"abstract":"<div><div>Recreating odors is considerably more challenging than, for example, recreating colors. Odor recreation necessitates knowledge of chemistry, sensory perception, sensors, automation, artificial intelligence, and machine learning. In general, recreating an odor requires analyzing its chemical composition by means of gas chromatography and/or mass spectrometry. However, odors can consist of hundreds of chemicals, which makes recreating them a complex, time-consuming and expensive task. Alternatively, one could try to create synthetic odors from a limited number of chemicals and compare them to the original odor. For the comparison, original and synthetic odors could be measured by an electronic nose, such as differential mobility spectrometry (DMS). In this article, a self-learning multichannel olfactory display that enables recreation of a target odor by mixing up to five chemicals was developed and tested. The display incorporates a machine learning module based on differential evolution algorithms for automatically finding a composition of a synthetic odor yielding a DMS measurement similar to the DMS measurement of the target odor. The article demonstrates via simulations and tests with real odors that the self-learning multichannel odor display can find synthetic odors resembling the target odors. Simulations proved that the selected algorithm is capable of yielding synthetic odors almost identical to the target odor. For two-component mixtures, the average deviations in flow rates between the target and trial odors were 0.8<!--> <!-->sccm after ten and 0.3<!--> <!-->sccm after 20 iterations, while for five-component mixtures the average deviations were 2.8<!--> <!-->sccm after ten and 1.4<!--> <!-->sccm after 20 iterations. Tests with isopropanol–ethanol mixtures also resulted in accurately reproduced synthetic mixtures but demonstrated challenges due to the unequal chemical imprints in DMS measurements. For ethanol, the average offset in the estimated flow rate was 2.2<!--> <!-->sccm, while it was 7.0<!--> <!-->sccm for isopropanol due to two runs with large offsets of 15 and 16<!--> <!-->sccm respectively.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100293"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147385095","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Kolmogorov–Arnold and deep learning networks for industrial explainable product quality prediction","authors":"Balázs Fricz ,&nbsp;Gergely Horváth ,&nbsp;Alex Kummer","doi":"10.1016/j.dche.2026.100289","DOIUrl":"10.1016/j.dche.2026.100289","url":null,"abstract":"<div><div>In the chemical process industry, the frequent measurement of quality variables is not always a viable path to take. It can be due to the measurement being time- or cost-consuming, leading to a sampling time of hours or even a day. As such, soft sensors were developed as a solution to this problem, providing a model trained on the available output measurements and predicting its value between samplings. This paper presents the process of soft sensor development for an industrial case study and the application of recently developed Kolmogorov–Arnold Networks (KAN) as a possible soft sensor model, where also the inherent interpretability of KAN through symbolic regression is analyzed.</div><div>The soft-sensor development and interpretability analysis is performed on a benchmark dataset (steam turbine dataset), and on a real industrial case study (aniline synthesis plant). The prediction performance of KAN is compared to different complexity ML models, and in addition, the models were compared on the basis of their interpretability and explainability using Shapley values. The results show that the inherent interpretability of KAN models using symbolic regression is too complicated on simple tasks, thus losing its ability to explain the model estimations, though its model performance is similar to the classic feedforward neural networks.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100289"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146022558","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Optimal rules base development in Type-2 fuzzy logic for stochastic chemical control system","authors":"Mohd Fauzi Zanil ,&nbsp;Zainal Ahmad ,&nbsp;Syamsul Rizal Abd Shukor ,&nbsp;Mohmmad Jakir Hossain Khan ,&nbsp;Mohd Hardyianto Vai Bahrun","doi":"10.1016/j.dche.2025.100275","DOIUrl":"10.1016/j.dche.2025.100275","url":null,"abstract":"<div><div>This study proposes the optimal development of rule bases using Type-2 fuzzy logic specifically designed for stochastic chemical control systems. The research addresses complexities and uncertainties inherent in stochastic pH neutralisation processes with Type-2 fuzzy logic as inversed hybrid model which able to provide good control action in the fuzzy rule. Comprehensive simulation and experimental-based performance evaluations, including setpoint tracking accuracy and disturbance rejection capabilities, were conducted to rigorously compare the proposed Type-2 fuzzy logic controller with traditional PID and conventional fuzzy logic controllers. Results demonstrate that the optimized Type-2 fuzzy logic controller significantly outperforms existing methods, offering faster system responses, minimized overshoot, and improved system stability. Further, robustness tests involving stochastic perturbations, such as variable flow rates of NaOH and HCl solutions and random acid injections during operational conditions, confirm the controller’s enhanced adaptability and effectiveness. The study concludes that the developed Type-2 fuzzy logic controller provides a robust, efficient, and reliable control solution constructed through simulation and validated using real experimental data, suitable for real-time (stochastic) management of complex stochastic chemical systems.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100275"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145976987","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"DFT in catalysis: Complex equations for practical computing applications in chemistry","authors":"Artur Brotons Rufes ,&nbsp;Sergio Posada Pérez ,&nbsp;Albert Poater","doi":"10.1016/j.dche.2025.100285","DOIUrl":"10.1016/j.dche.2025.100285","url":null,"abstract":"<div><div>Density Functional Theory (DFT) has become the cornerstone of modern computational catalysis, providing a practical balance between accuracy and efficiency in describing molecular structure, bonding, and reactivity. This review presents a comprehensive overview of DFT methodology, from its quantum-mechanical foundations and basis-set construction to the hierarchy of exchange–correlation functionals defined by Jacob’s ladder. We discuss how DFT enables mechanistic elucidation of homogeneous and heterogeneous catalytic processes, highlighting benchmark studies that compare functional performance across representative reactions and transition states. Key interpretative tools, such as bond order analysis (Mayer, Wiberg, AIM/QTAIM), Natural Bond Orbital (NBO) theory, Energy Decomposition Analysis (EDA), and Non-Covalent Interaction (NCI) plots, are introduced as essential descriptors linking electronic structure to reactivity. The review also explores the integration of DFT with machine learning, microkinetic modeling, and automated reaction discovery, outlining recent advances toward predictive catalysis. Collectively, this work provides both conceptual and practical guidance for applying DFT to catalytic problems, emphasizing methodological awareness, descriptor-based interpretation, and emerging data-driven strategies for rational catalyst design. However, the main take-home message is that for DFT calculations, while in-depth methodological expertise is not essential, a clear comprehension of the theory’s practical application is crucial.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100285"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145840618","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Data driven prediction of hydrochar yields from biomass hydrothermal carbonization using extreme gradient boosting algorithm with principal component analysis","authors":"Tossapon Katongtung ,&nbsp;Nattawut Khuenkaeo ,&nbsp;Yuttana Mona ,&nbsp;Pana Suttakul ,&nbsp;James C. Moran ,&nbsp;Korrakot Y. Tippayawong ,&nbsp;Nakorn Tippayawong","doi":"10.1016/j.dche.2025.100283","DOIUrl":"10.1016/j.dche.2025.100283","url":null,"abstract":"<div><div>Dimensionality reduction plays a critical role in efficiently managing large and complex datasets in machine learning (ML) applications. This study presents an innovative integration of principal component analysis (PCA) and extreme gradient boosting (XGB) to model the hydrothermal carbonization (HTC) process. PCA effectively reduced the feature space from 18 to 9 principal components with minimal impact on model accuracy (R² decreased slightly from 0.8900 to 0.8480), significantly simplifying the model complexity. To enhance interpretability, one- and two-dimensional partial dependence plots (PDP) were employed, revealing key features and their interactions influencing HTC outcomes. This combined approach not only improves predictive performance but also provides meaningful insights into the underlying process variables, addressing common challenges of ML opacity. While the model demonstrates strong predictive capability, further experimental validation and extension to diverse biomass types are recommended to confirm practical applicability and enhance versatility. The proposed methodology offers a robust, interpretable, and computationally efficient framework for optimizing HTC and can guide future research involving high-dimensional datasets.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100283"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791149","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Transfer learning of data-driven crystallisation processes via constrained Neural Ordinary Differential Equations","authors":"Daniele Pessina ,&nbsp;Tony Tian ,&nbsp;Oliver Watson ,&nbsp;Jerry Y. Heng ,&nbsp;Maria M. Papathanasiou","doi":"10.1016/j.dche.2026.100292","DOIUrl":"10.1016/j.dche.2026.100292","url":null,"abstract":"<div><div>Modelling complex crystallisation processes remains challenging due to limited experimental datasets, high measurement noise, and the need for generalisability across varying operating conditions. Neural Ordinary Differential Equations (NODEs) and transfer learning (TL) offer promising tools to overcome these limitations by providing data-efficient, flexible, and transferable modelling frameworks. This work investigates the use of NODEs to model protein crystallisation dynamics under data-scarce conditions. A NODE trained on a data-rich source system successfully captures solute consumption and particle size dynamics, but when applied to data-sparse target systems, scratch-trained NODEs exhibit limited generalisation and unphysical behaviours. To address this, several TL strategies are evaluated, including layer freezing, parameter deviation penalisation, and system-embedding within the neural architecture. Results show that layer freezing and deviation penalty consistently improve knowledge transfer, while system-embedding offers robustness in noisy or undersampled datasets. In addition, physics-informed NODEs, constrained to enforce monotonic concentration decay and crystal growth, demonstrate greater stability under high noise and sparse measurement regimes, ensuring physically consistent predictions. Overall, the combination of constrained NODEs with appropriate TL strategies provides a robust framework for accurate, transferable modelling of crystallisation systems in low-data regimes.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100292"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147385204","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"All you need is noise — from feature selection to explainable industrial AI","authors":"Mattia Vallerio ,&nbsp;Antonio del Rio Chanona ,&nbsp;Francisco J. Navarro-Brull","doi":"10.1016/j.dche.2026.100290","DOIUrl":"10.1016/j.dche.2026.100290","url":null,"abstract":"<div><div>Modern chemical plants record thousands of sensor tags, yet only a small fraction meaningfully influence yield, quality, or throughput. Identifying those key drivers is often more difficult than building the predictive model itself. In this work, we show that appending one or more <em>Synthetic Noise Features</em> (SNFs), non-informative random variables known <em>a priori</em>, provide a simple reference for judging variable relevance. We show the impact of this model agnostic step across three workflows. In supervised learning, noise features establish an automatic cutoff for the feature importance, guide model regularization and signal when the dataset itself lacks predictive information. In unsupervised learning, they provide an unbiased threshold preventing spurious anomalies and latent dimensions. Finally, we demonstrate the applicability of this approach to small datasets typical of experimental work and Design of Experiments (DoE), including <em>Definitive Screening</em>, <em>Response Surface</em>, and <em>space-filling</em> designs, as well as <em>active learning</em> using <em>Bayesian optimization</em>. By turning <em>nothing but noise</em> into a quantitative benchmark, SNFs offer an immediately deployable safeguard against overfitting and misplaced experimental effort in data-driven chemical engineering.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100290"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147385201","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Robust reinforcement learning for nonlinear process control with stability guarantees","authors":"Xiaodong Cui ,&nbsp;Arthur Khodaverdian ,&nbsp;Panagiotis D. Christofides","doi":"10.1016/j.dche.2026.100291","DOIUrl":"10.1016/j.dche.2026.100291","url":null,"abstract":"<div><div>Reinforcement learning (RL) offers a promising route to fast, nonlinear feedback control for complex process systems; however, its deployment is hindered by the lack of formal stability guarantees and sensitivity to model-plant mismatch under constraints. This paper proposes a stability and robustness-oriented RL-based control framework for nonlinear constrained processes by explicitly integrating Lyapunov-based decision rules into the RL closed loop and importing an offset-free approach from model predictive control (MPC). The RL policy is treated as a performance-seeking candidate controller that is supervised by a Lyapunov-certified fallback controller: at each sampling instant, the learned candidate input is evaluated against a Lyapunov condition, and any violating proposal is rejected in favor of a stabilizing backup input, yielding Lyapunov-certified practical stability under sample-and-hold implementation. To mitigate steady-state offsets and enhance robustness to disturbances and mismatches, the state available to the learning agent is augmented with online-estimated uncertainty/disturbance variables in the spirit of offset-free MPC, enabling the policy/value function to condition its decisions on the magnitude of uncertainty rather than overfitting nominal dynamics. The proposed architecture is demonstrated using two representative RL methods—an HJB-based value-critic approach and a TD3-based actor-critic approach—both deployed under the same Lyapunov-supervisory switching logic. Simulation studies on nonlinear chemical process control problems show that the proposed RL-based control framework preserves the low online computational cost while enforcing Lyapunov stability and improving robustness under disturbances, thereby advancing RL toward reliable process control deployment.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100291"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147385203","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Smart control of heavy metal adsorption onto LDC wastes for Industry 4.0 applications","authors":"Mohammad Gheibi ,&nbsp;Seyyed Roohollah Masoomi ,&nbsp;Mohammad Eftekhari ,&nbsp;Martin Palušák ,&nbsp;Daniele Silvestri ,&nbsp;Miroslav Černík ,&nbsp;Stanisław Wacławek","doi":"10.1016/j.dche.2025.100286","DOIUrl":"10.1016/j.dche.2025.100286","url":null,"abstract":"<div><div>Due to the complexity and the need for automation in adsorption systems, this study develops a decision-making model using 20 Machine Learning Algorithms (MLAs), Response Surface Methodology (RSM), and lab tests. Inputs include pH, metal type, concentration, adsorbent mass, and time; output is removal percentage for process control. The best-performing models (Tree Random Forest: TRF, lazy Instance-Based K: IBK, and Function Multilayer Perceptron: FMLP) achieve high accuracy for prediction of Removal Percentage (RP) of heavy metals with &gt;0.92 correlation coefficient and &gt;0.8 recall/precision indicators. As a novelty of this study, a decision-making model for the heavy metal adsorption process onto waste materials is developed for the first time using the Knowledge Flow (KF) platform in WEKA software, incorporating real-time data to improve process control and operational efficiency. The results demonstrated that the cation type and pH, with a P-value &lt; 0.001, are the most significant factors affecting the RP. To achieve maximum RP, the pH should be set to 5, and the adsorbent amount should be in a range of 8.67-10 g L^-¹, regardless of the initial concentration and type of ions (Pb²⁺, Mn²⁺, and Co²⁺). Then, evaluating MLAs showed that TRF, with a correlation coefficient exceeding 0.96, performs best for predicting RP, potentially reaching 0.99 at a split percentage around 80%. TRF uses real-time experimental data in water treatment systems to anticipate RP up to 98.75% and to activate alarms for RP below 80% using the KF principle.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100286"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977798","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Advanced machine learning techniques for predicting solution gas-oil ratios in petroleum reservoirs: A comprehensive study and new empirical correlation","authors":"Mohammad Sepahvand ,&nbsp;Majid Mohammadi ,&nbsp;Ali Madani ,&nbsp;Mahin Schaffie","doi":"10.1016/j.dche.2026.100288","DOIUrl":"10.1016/j.dche.2026.100288","url":null,"abstract":"<div><div>The precise assessment of crude oil properties is crucial for various tasks in reservoir evaluation, including estimating reserves, enhancing oil recovery, predicting oil reservoir performance, designing pipelines and production equipment, and simulating reservoirs. Therefore, the reservoir fluid properties, especially the solution gas-oil ratio (<em>GOR</em> or <em>R_s</em>), which depends on bubble point pressure (<em>P_b</em>), reservoir temperature (<em>T_r</em>), gas specific gravity (<span><math><msub><mi>γ</mi><mi>g</mi></msub></math></span>) and API are of great importance. On the other side, due to the time-consuming and expensive nature of pressure-volume-temperature (PVT) laboratory models, accurate experimental correlations are significant in petroleum engineering. Therefore, a cheap, accurate, fast model is needed to estimate these parameters. In this study, to develop a comprehensive and reliable model, 1309 data points were collected from crude oil samples in different geographical locations. The development of a new correlation was presented with high accuracy. Also, intelligent models based on neural networks such as Multilayer perceptron neural networks (MLP), Radial basis function network (RBF), Extreme gradient boosting (XGBoost), Adaptive boosting (AdaBoost), and Gradient boosting with categorical features support (CatBoost), were used to predict <em>R_s</em>. To assess the performance of the models, statistical and graphical error analyses, sensitivity analyses, group errors, model trends, and a leverage approach were employed. The CatBoost model delivered the highest predictive accuracy, with an <em>AAPRE</em> of 8.13%, surpassing competitors like XGBoost (10.51%) and AdaBoost (24.78%). In addition, the new correlation developed achieved an <em>AAPRE</em> of 16.40%, offering a marked advancement over conventional empirical correlations that typically range from 17.24% to 64.88% in error. These findings demonstrate the superior precision and versatility of methods, underscoring its innovative value for estimating <em>R_s</em> across varied reservoir environments. Factorial sensitivity analysis showed that <em>P_b</em> had the greatest impact on <em>R_s</em> with 25.4%.</div></div>","PeriodicalId":72815,"journal":{"name":"Digital Chemical Engineering","volume":"18 ","pages":"Article 100288"},"PeriodicalIF":4.1,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077746","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}