Accurately acquiring catalyst size and morphology is essential for supporting catalytic reaction process design and optimal control. We report an intelligent catalyst sizing and morphological classification method based on the Mask-RCNN framework. A dataset of 9880 high-resolution images was captured by using a self-made fiber-optic endoscopic system for 13 kinds of silicoaluminophosphate-34 (SAPO-34) catalyst samples with different coke. Then there were approximately 877881 individual particles extracted from this dataset by our AI-based particle recognition algorithm. To clearly describe the morphology of irregular particles, we proposed a hybrid classification criterion that combines five different parameters, which are deformity, circularity, roundness, aspect ratio, and compactness. Therefore, catalyst morphology can be classified into two categories with four types. The first category includes regular types, such as the spherical, ellipsoidal, and rod-shaped types. And all the irregular types fall into the second category. The experimental results showed that a catalyst particle tends to be larger when its coke deposition increased. Whereas particle morphology remained primarily spherical and ellipsoidal, the ratio of each type varied slightly according to its coke. Our findings illustrate that this is a promising approach to be developing intelligent instruments for catalyst particle sizing and classification.

G protein coupled receptor kinase 2 (GRK2) is a kinase that regulates cardiac signaling activity. Inhibiting GRK2 is a promising mechanism for the treatment of heart failure (HF). Further development and optimization of inhibitors targeting GRK2 are highly meaningful. Therefore, in order to design GRK2 inhibitors with better performance, the most active molecule was selected as a reference compound from a data set containing 4-pyridylhydrazone derivatives and triazole derivatives, and its scaffold was extracted as the initial scaffold. Then, a powerful optimization-based framework for de novo drug design, guided by binding affinity, was used to generate a virtual molecular library targeting GRK2. The binding affinity of each virtual compound in this dataset was predicted by our developed deep learning model, and the designed potential compound with high binding affinity was selected for molecular docking and molecular dynamics simulation. It was found that the designed potential molecule binds to the ATP site of GRK2, which consists of key amino acids including Arg199, Gly200, Phe202, Val205, Lys220, Met274 and Asp335. The scaffold of the molecule is stabilized mainly by H-bonding and hydrophobic contacts. Concurrently, the reference compound in the dataset was also simulated by docking. It was found that this molecule also binds to the ATP site of GRK2. In addition, its scaffold is stabilized mainly by H-bonding and π-cation stacking interactions with Lys220, as well as hydrophobic contacts. The above results show that the designed potential molecule has similar binding modes to the reference compound, supporting the effectiveness of our framework for activity-focused molecular design. Finally, we summarized the interaction characteristics of general GRK2 inhibitors and gained insight into their molecule-target binding mechanisms, thereby facilitating the expansion of lead to hit compound.

The biomass and coal co-pyrolysis (BCP) technology combines the advantages of both resources, achieving efficient resource complementarity, reducing reliance on coal, and minimizing pollutant emissions. However, this process still encounters numerous challenges in attaining optimal economic and environmental performance. Therefore, an ensemble learning (EL) framework is proposed for the BCP process in this study to optimize the synergistic benefits while minimizing negative environmental impacts. Six different ensemble learning models are developed to investigate the impact of input features, such as biomass characteristics, coal characteristics, and pyrolysis conditions on the product profit and CO₂ emissions of the BCP processes. The Optuna method is further employed to automatically optimize the hyperparameters of BCP process models for enhancing their predictive accuracy and robustness. The results indicate that the categorical boosting (CAB) model of the BCP process has demonstrated exceptional performance in accurately predicting its product profit and CO₂ emission (R²>0.92) after undergoing five-fold cross-validation. To enhance the interpretability of this preferred model, the Shapley additive explanations and partial dependence plot analyses are conducted to evaluate the impact and importance of biomass characteristics, coal characteristics, and pyrolysis conditions on the product profitability and CO₂ emissions of the BCP processes. Finally, the preferred model coupled with a reference vector guided evolutionary algorithm is carried to identify the optimal conditions for maximizing the product profit of BCP process products while minimizing CO₂ emissions. It indicates the optimal BCP process can achieve high product profits (5290.85 CNY·t^-1) and low CO₂ emissions (7.45 kg·t^-1).

The fractionating tower bottom in fluid catalytic cracking Unit (FCCU) is highly susceptible to coking due to the interplay of complex external operating conditions and internal physical properties. Consequently, quantitative risk assessment (QRA) and predictive maintenance (PdM) are essential to effectively manage coking risks influenced by multiple factors. However, the inherent uncertainties of the coking process, combined with the mixed-frequency nature of distributed control systems (DCS) and laboratory information management systems (LIMS) data, present significant challenges for the application of data-driven methods and their practical implementation in industrial environments. This study proposes a hierarchical framework that integrates deep learning and fuzzy logic inference, leveraging data and domain knowledge to monitor the coking condition and inform prescriptive maintenance planning. The framework proposes the multi-layer fuzzy inference system to construct the coking risk index, utilizes multi-label methods to select the optimal feature dataset across the reactor-regenerator and fractionation system using coking risk factors as label space, and designs the parallel encoder-integrated decoder architecture to address mixed-frequency data disparities and enhance adaptation capabilities through extracting the operation state and physical properties information. Additionally, triple attention mechanisms, whether in parallel or temporal modules, adaptively aggregate input information and enhance intrinsic interpretability to support the disposal decision-making. Applied in the 2.8 million tons FCCU under long-period complex operating conditions, enabling precise coking risk management at the fractionating tower bottom.

Anomaly fluctuations in operating conditions, catalyst wear, crushing, and the deterioration of feedstock properties in fluid catalytic cracking (FCC) units can disrupt the normal circulating fluidization process of the catalyst. Although several effective models have been proposed in previous research to address anomaly detection in chemical processes, most fail to adequately capture the spatial-temporal dependencies of multi-source, mixed-frequency information. In this study, an innovative multi-source mixed-frequency information fusion framework based on a spatial-temporal graph attention network (MIF-STGAT) is proposed to investigate the causes of FCC regenerator catalyst loss anomalies for guide onsite operational management, enhancing the long-term stability of FCC unit operations. First, a reconstruction-based dual-encoder-decoder framework is developed to facilitate the acquisition of mixed-frequency features and information fusion during the FCC regenerator catalyst loss process. Subsequently, a graph attention network and a multilayer long short-term memory network with a differential structure are integrated into the reconstruction-based dual-encoder-shared-decoder framework to capture the dynamic fluctuations and critical features associated with anomalies. Experimental results from the Chinese FCC industrial process demonstrate that MIF-STGAT achieves excellent accuracy and interpretability for anomaly detection.

Conceptual process design (CPD) research focuses on finding design alternatives that address various design problems. It has a long history of well-established methodologies to answer these complex questions, such as heuristics, mathematical programming, and pinch analysis. Nonetheless, progress continues from different formulations of design problems using bottom-up approaches, to the utilization of new tools such as artificial intelligence (AI). It was not until recently that AI methods were involved again in assisting the decision-making steps for chemical engineers. This has led to a gap in understanding AI's capabilities and limitations within the field of CPD research. Thus, this article aims to provide an overview of conventional methods for process synthesis, integration, and intensification approaches and survey emerging AI-assisted process design applications to bridge the gap. A review of all AI-assisted methods is highlighted, where AI is used as a key component within a design framework, to explain the utility of AI with comparative examples. The studies were categorized into supervised and reinforcement learning based on the machine learning training principles they used to enhance the understanding of requirements, benefits, and challenges that come with it. Furthermore, we provide challenges and prospects that can facilitate or hinder the progress of AI-assisted approaches in the future.

The optimization of reaction processes is crucial for the green, efficient, and sustainable development of the chemical industry. However, how to address the problems posed by multiple variables, nonlinearities, and uncertainties during optimization remains a formidable challenge. In this study, a strategy combining interpretable machine learning with metaheuristic optimization algorithms is employed to optimize the reaction process. First, experimental data from a biodiesel production process are collected to establish a database. These data are then used to construct a predictive model based on artificial neural network (ANN) models. Subsequently, interpretable machine learning techniques are applied for quantitative analysis and verification of the model. Finally, four metaheuristic optimization algorithms are coupled with the ANN model to achieve the desired optimization. The research results show that the methanol: palm fatty acid distillate (PFAD) molar ratio contributes the most to the reaction outcome, accounting for 41%. The ANN-simulated annealing (SA) hybrid method is more suitable for this optimization, and the optimal process parameters are a catalyst concentration of 3.00% (mass), a methanol: PFAD molar ratio of 8.67, and a reaction time of 30 min. This study provides deeper insights into reaction process optimization, which will facilitate future applications in various reaction optimization processes.

Catalysis has made great contributions to the productivity of human society. Therefore, the pursuit of new catalysts and research on catalytic processes has never stopped. Continuous and in-depth catalysis research significantly increases the complexity of dynamic systems and multivariate optimization, thus posing higher challenges to research methodologies. Recently, the significant advancement of generative artificial intelligence (AI) provides new opportunities for catalysis research. Different from traditional discriminative AI, this state-of-the-art technique generates new samples based on existing data and accumulated knowledge, which endows it with attractive potential for catalysis research — a field featuring a vast exploration space, diverse data types and complex mapping relationships. Generative AI can greatly enhance both the efficiency and innovation capacity of catalysis research, subsequently fostering new scientific paradigms. This perspective covers the basic introduction, unique advantages of this powerful tool, and presents cases of generative AI implemented in various catalysis researches, including catalyst design and optimization, characterization technique enhancement and guidance for new research paradigms. These examples highlight its exceptional efficiency and general applicability. We further discuss the practical challenges in implementation and future development perspectives, ultimately aiming to promote better applications of generative AI in catalysis.

Accurate fault root cause diagnosis is essential for ensuring stable industrial production. Traditional methods, which typically rely on the entire time series and overlook critical local features, can lead to biased inferences about causal relationships, thus hindering the accurate identification of root cause variables. This study proposed a shapelet-based state evolution graph for fault root cause diagnosis (SEG-RCD), which enables causal inference through the analysis of the important local features. First, the regularized autoencoder and fault contribution plot are used to identify the fault onset time and candidate root cause variables, respectively. Then, the most representative shapelets were extracted to construct a state evolution graph. Finally, the propagation path was extracted based on fault unit shapelets to pinpoint the fault root cause variable. The SEG-RCD can reduce the interference of noncausal information, enhancing the accuracy and interpretability of fault root cause diagnosis. The superiority of the proposed SEG-RCD was verified through experiments on a simulated penicillin fermentation process and an actual one.

Thermal power generation systems have stringent requirements for water and steam quality, i.e., condensate water quality is one of the critical issues. In this paper, we designed a two-layer model based on an autoencoder and expert knowledge to achieve the early warning and causal analysis of condensate water quality abnormalities. An early warning model using an autoencoder model is built based on the historical data affecting the condensate water quality. Next, an analytical model of condensate water quality abnormalities was then developed by combining expert knowledge and trend test algorithms. Two different datasets were used to test the proposed model, respectively. The accuracy of the autoencoder model in the short-period test set is 88.83%, which shows that the early warning model can accurately analyze the condensate water quality data and achieve the purpose of early warning. For the long-time period test set, the model can correctly identify each abnormality and simultaneously indicates the cause of the abnormal condensate water quality. The proposed model can correctly identify abnormal working conditions and it is applicable to other thermal power plants.

The integration of artificial intelligence (AI) with high-throughput experimentation (HTE) techniques is revolutionizing catalyst design, addressing challenges in efficiency, cost, and scalability. This review explores the synergistic application of AI and HTE, highlighting their role in accelerating catalyst discovery, optimizing reaction parameters, and understanding structure-performance relationships. HTE facilitates the rapid preparation, characterization, and evaluation of diverse catalyst formulations, generating large datasets essential for AI model training. Machine learning algorithms, including regression models, neural networks, and active learning frameworks, analyze these datasets to uncover the underlying relationships between the data, predict performance, and optimize experimental workflows in real-time. Case studies across heterogeneous, homogeneous, and electrocatalysis demonstrate significant advancements, including improved reaction selectivity, enhanced material stability, and shorten discovery cycles. The integration of AI with HTE has significantly accelerated discovery cycles, enabling the optimization of catalyst formulations and reaction conditions. Despite these achievements, challenges remain, including reliance on researcher expertise, real-time adaptability, and the complexity of large-scale data analysis. Addressing these limitations through refined experimental protocols, standardized datasets, and interpretable AI models will unlock the full potential of AI-HTE integration.

Data-driven approaches are extensively employed to model complex chemical engineering processes, such as hydrotreating, to address the challenges of mechanism-based methods demanding deep process understanding. However, the development of such models requires specialized expertise in data science, limiting their broader application. Large language models (LLMs), such as GPT-4, have demonstrated potential in supporting and guiding research efforts. This work presents a novel AI-assisted framework where GPT-4, through well-engineered prompts, facilitates the construction and explanation of multi-objective neural networks. These models predict hydrotreating products properties (such as distillation range), including refined diesel and refined gas oil, using feedstock properties, operating conditions, and recycle hydrogen composition. Gradient-weighted class activation mapping was employed to identify key features influencing the output variables. This work illustrates an innovative AI-guided paradigm for chemical engineering applications, and the designed prompts hold promise for adaptation to other complex processes.

Fault diagnosis in industrial process is essential for ensuring production safety and efficiency. However, existing methods exhibit limited capability in recognizing hard samples and struggle to maintain consistency in feature distributions across domains, resulting in suboptimal performance and robustness. Therefore, this paper proposes a fault diagnosis neural network for hard sample mining and domain adaptive (SmdaNet). First, the method uses deep belief networks (DBN) to build a diagnostic model. Hard samples are mined based on the loss values, dividing the data set into hard and easy samples. Second, elastic weight consolidation (EWC) is used to train the model on hard samples, effectively preventing information forgetting. Finally, the feature space domain adaptation is introduced to optimize the feature space by minimizing the Kullback–Leibler divergence of the feature distributions. Experimental results show that the proposed SmdaNet method outperforms existing approaches in terms of classification accuracy, robustness and interpretability on the penicillin simulation and Tennessee Eastman process datasets.

Accurate prediction of pure component physiochemical properties is crucial for process integration, multiscale modelling, and optimization. In this work, an enhanced framework for pure component property prediction by using explainable machine learning methods is proposed. In this framework, the molecular representation method based on the connectivity matrix effectively considers atomic bonding relationships to automatically generate features. The supervised machine learning model random forest is applied for feature ranking and pooling. The adjusted R² is introduced to penalize the inclusion of additional features, providing an assessment of the true contribution of features. The prediction results for normal boiling point (T_b), liquid molar volume (L_mv), critical temperature (T_c) and critical pressure (P_c) obtained using Artificial Neural Network and Gaussian Process Regression models confirm the accuracy of the molecular representation method. Comparison with GC based models shows that the root-mean-square error on the test set can be reduced by up to 83.8%. To enhance the interpretability of the model, a feature analysis method based on Shapley values is employed to determine the contribution of each feature to the property predictions. The results indicate that using the feature pooling method reduces the number of features from 13316 to 100 without compromising model accuracy. The feature analysis results for T_b, L_mv, T_c, and P_c confirms that different molecular properties are influenced by different structural features, aligning with mechanistic interpretations. In conclusion, the proposed framework is demonstrated to be feasible and provides a solid foundation for mixture component reconstruction and process integration modelling.

Knowledge graphs (KGs) offer a structured, machine-readable format for organizing complex information. In heterogeneous catalysis, where data on catalytic materials, reaction conditions, mechanisms, and synthesis routes are dispersed across diverse sources, KGs provide a semantic framework that supports data integration under the FAIR (Findable, Accessible, Interoperable, and Reusable) principles. This review aims to survey recent developments in catalysis KGs, describe the main techniques for graph construction, and highlight how artificial intelligence, particularly large language models (LLMs), enhances graph generation and query. We conducted a systematic analysis of the literature, focusing on ontology-guided text mining pipelines, graph population methods, and maintenance strategies. Our review identifies key trends: ontology-based approaches enable the automated extraction of domain knowledge, LLM-driven retrieval-augmented generation supports natural-language queries, and scalable graph architectures range from a few thousand to over a million triples. We discuss state-of-the-art applications, such as catalyst recommendation systems and reaction mechanism discovery tools, and examine the major challenges, including data heterogeneity, ontology alignment, and long-term graph curation. We conclude that KGs, when combined with AI methods, hold significant promise for accelerating catalyst discovery and knowledge management, but progress depends on establishing community standards for ontology development and maintenance. This review provides a roadmap for researchers seeking to leverage KGs to advance heterogeneous catalysis research.

This study leverages machine learning to perform high-throughput computational screening of n-hexane cracking initiators. Artificial neural networks are applied to predict the chemical performance of initiators, using simulated pyrolysis data as the training dataset. Various feature extraction methods are utilized, and five neural network architectures are developed to predict the co-cracking product distribution based on molecular structures. High-throughput screening of 12946 molecules outside the training dataset identifies the top 10 initiators for each target product—ethylene, propylene, and butadiene. The relative error between predicted and simulated values is less than 7%. Additionally, reaction pathway analysis elucidates the mechanisms by which initiators influence the distribution of cracking products. The proposed framework provides a practical and efficient approach for the rapid identification and evaluation of high-performance cracking initiators.

The potential toxicity of ionic liquids (ILs) affects their applications; how to control the toxicity is one of the key issues in their applications. To understand its toxicity structure relationship and promote its greener application, six different machine learning algorithms, including Bagging, Adaptive Boosting (AdaBoost), Gradient Boosting (GBoost), Stacking, Voting and Categorical Boosting (CatBoost), are established to model the toxicity of ILs on four distinct datasets including Leukemia rat cell line IPC-81 (IPC-81), Acetylcholinesterase (AChE), Escherichia coli (E.coli) and Vibrio fischeri. Molecular descriptors obtained from the simplified molecular input line entry system (SMILES) are used to characterize ILs. All models are assessed by the mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and correlation coefficient (R²). Additionally, an interpretation model based on SHapley Additive exPlanations (SHAP) is built to determine the positive and negative effects of each molecular feature on toxicity. With additional parameters and complexity, the Catboost model outperforms the other models, making it a more reliable model for ILs' toxicity prediction. The results of the model's interpretation indicate that the most significant positive features, SMR_VSA5, PEOE_VSA8, Kappa2, PEOE_VSA6, SMR_VSA5, PEOE_VSA6 and EState_VSA1, can increase the toxicity of ILs as their levels rise, while the most significant negative features, VSA_EState7, EState_VSA8, PEOE_VSA9 and FpDensityMorgan1, can decrease the toxicity as their levels rise. Also, an IL's toxicity will grow as its average molecular weight and number of pyridine rings increase, whereas its toxicity will decrease as its hydrogen bond acceptors increase. This finding offers a theoretical foundation for rapid screening and synthesis of environmentally-benign ILs.

The structural and operational optimization of gas-liquid stirred bioreactors presents both complexity and critical importance for enhancing mass transfer performance. This study proposes a machine learning (ML)-driven approach to identify key features and predict the volumetric mass transfer coefficient (k_La). Four ML models were adopted and compared for k_La prediction in Newtonian and non-Newtonian fluids by evaluative indices, with CatBoost and XGBoost emerging as the optimal models, respectively. Specifically, it is demonstrated that Catboost has higher prediction accuracy (AARD = 18.84%) than empirical equations by effectively incorporating multidimensional features (structural, impeller, and operational), while simultaneously extending applicability to diverse Newtonian fluids. For non-Newtonian fluids, XGBoost outperforms empirical equations by effectively incorporating fluid rheological parameters (consistency coefficient, power-law index), thereby better capturing shear-thinning behavior. Feature importance analysis further identified rotational speed (for Newtonian fluids) and liquid height (for non-Newtonian fluids) as the key features, while 2D partial dependence analysis establishes quantitative optimization ranges. This ML approach provides an efficient predictive tool for gas-liquid stirred bioreactor design and optimization.

Ionic liquids (ILs) and deep eutectic solvents (DESs) as green solvents have attracted dramatic attention recently due to their highly tunable properties. However, traditional experimental screening methods are inefficient and resource-intensive. The article provides a comprehensive overview of various ML algorithms, including artificial neural network (ANN), support vector machine (SVM), random forest (RF), and gradient boosting trees (GBT), etc., which have demonstrated exceptional performance in handling complex and high-dimensional data. Furthermore, the integration of ML with quantum chemical calculations and conductor-like screening model-real solvent (COSMO-RS) has significantly enhanced predictive accuracy, enabling the rapid screening and design of novel solvents. Besides, recent ML applications in the prediction and design of ILs and DESs focused on solubility, melting point, electrical conductivity, and other physicochemical properties become more and more. This paper emphasizes the potential of ML in solvent design, overviewing an efficient approach to accelerate the development of sustainable and high-performance materials, providing guidance for their widespread application in a variety of industrial processes.

Integrating Bayesian Optimization with Volume of Fluid (VOF) simulations, this work aims to optimize the operational conditions and geometric parameters of T-junction microchannels for target droplet sizes. Bayesian Optimization utilizes Gaussian Process (GP) as its core model and employs an adaptive search strategy to efficiently explore and identify optimal combinations of operational parameters within a limited parameter space, thereby enabling rapid optimization of the required parameters to achieve the target droplet size. Traditional methods typically rely on manually selecting a series of operational parameters and conducting multiple simulations to gradually approach the target droplet size. This process is time-consuming and prone to getting trapped in local optima. In contrast, Bayesian Optimization adaptively adjusts its search strategy, significantly reducing computational costs and effectively exploring global optima, thus greatly improving optimization efficiency. Additionally, the study investigates the impact of rectangular rib structures within the T-junction microchannel on droplet generation, revealing how the channel geometry influences droplet formation and size. After determining the target droplet size, we further applied Bayesian Optimization to refine the rib geometry. The integration of Bayesian Optimization with computational fluid dynamics (CFD) offers a promising tool and provides new insights into the optimal design of microfluidic devices.

Accurate prediction of solubility data in the Sodium Chloride-Sodium Sulfate-Water system is essential. It provides theoretical support for salt lake resource development and wastewater treatment technologies. This study proposes an innovative solubility prediction approach. It addresses the limitations of traditional thermodynamic models. This is particularly important when experimental data from various sources contain inconsistencies. Our approach combines the Weighted Local Outlier Factor technique for anomaly detection with a Deep Ensemble Neural Network architecture. This methodology effectively removes local outliers while preserving data distribution integrity, and integrates multiple neural network sub-models to comprehensively capture system features while minimizing individual model biases. Experimental validation demonstrates exceptional prediction performance across temperatures from -20 °C to 150 °C, achieving a coefficient of determination of 0.989 after Bayesian hyperparameter optimization. This data-driven approach provides more accurate and universally applicable solubility predictions than conventional thermodynamic models, offering theoretical guidance for industrial applications in salt lake resource utilization, separation process optimization, and environmental salt management systems.

Direct coal liquefaction products offer a considerable quantity of cycloalkanes, which are the valuable candidates for making the high energy density fuels. The creation of such fuels depends on designing molecular structures and calculating their properties, which can be expedited with computer-aided techniques. In this study, a dataset containing 367 fuel molecules was constructed based on the analysis of direct coal liquefied oil. Three convolutional neural network property prediction models have been created based on molecular structure-physical and chemical property data from the library. All the models have good fitting ability with R² values above 0.97. Then, a variational autoencoder generation model has been established using the molecular structures from the library, focusing on the structure of saturated cycloalkanes. The structure-property prediction model was then applied to the newly generated molecules, assessing their density, volumetric calorific value, and melting point. As a result, 70000 novel molecular structures were generated, and 25 molecular structures meeting the criteria for high energy density fuels were identified. The established variational autoencoder model in this study effectively assimilates the structural information from the sample set and autonomously generates novel high energy density fuels, which is difficult to achieve in traditional experimental methods.

Chemical synthesis is essential in industries such as petrochemicals, fine chemicals, and pharmaceuticals, driving economic and social development. The increasing demand for new molecules and materials calls for novel chemical reactions; however, manual experimental screening is time-consuming. Artificial intelligence (AI) offers a promising solution by leveraging large-scale experimental data to model chemical reactions, although challenges such as the lack of standardization and predictability in chemical synthesis hinder AI applications. Additionally, the multi-scale nature of chemical reactions, along with complex multiphase processes, further complicates the task. Recent advances in microchemical systems, particularly continuous flow methods using microreactors, provide precise control over reaction conditions, enhancing reproducibility and enabling high-throughput experimentation. These systems minimize transport-related inconsistencies and facilitate scalable industrial applications. This review systematically explores recent developments in intelligent synthesis based on microchemical systems, focusing on reaction system design, synthesis robots, closed-loop optimization, and high-throughput experimentation, while identifying key areas for future research.