Informatics Modeling of High Tg Polymers: Assessing the Role of Processing versus Chemistry

Mechanical and thermal properties prediction and optimization of multiscale high performance hybrid biopolymer nanocomposites

Bayesian optimization-driven exploration of process–structure–property relationship for polyethylene composite films

The mechanical properties of polymers depend on their processing conditions. However, process optimization often suffers from complex parameter space and incomplete understanding of the structure-property relationship. This study fabricates and optimizes polyethylene (PE)-thermally reduced graphene oxide (TrGO) composite films and investigates how drawing temperature and drawing ratio affect their tensile strength (TS) and elongation-at-break (EAB). A multi-objective Bayesian optimization (MOBO) framework was introduced to explore the fabrication parameter space to identify the Pareto front between TS and EAB. Structural analyses reveal that drawing increases the molecular orientation, crystallinity and crystallite size in the films, while a high drawing temperature promotes preferential crystallization along the (200) plane. The combined effects of drawing ratio and drawing temperature collectively contribute to the enhanced TS of the films. This study establishes a clear process–structure–property relationship in PE composite films and demonstrates an efficient strategy for polymer-based composite materials design.

Boosting the Prediction Accuracy of Glass Transition Temperature in Polyimides: A Hybrid Machine Learning Approach Integrating Morgan Fingerprints and Molecular Descriptors

The glass transition temperature (Tg) of polyimides is a critical parameter determining their processability and application performance. Traditional experimental methods for measuring Tg are time‐consuming and costly, while existing machine learning prediction models predominantly rely on manually defined molecular descriptors, which often fail to fully capture detailed molecular structural information, limiting their prediction accuracy and generalization capability. To address this, this study proposes a hybrid feature engineering strategy combining Morgan fingerprints and molecular descriptors to comprehensively represent the chemical structure of polyimides. Based on a dataset of 1257 polyimide samples from a public database, we systematically compared six feature selection methods and employed multiple mainstream machine learning algorithms for modeling. The results show that the CATB model performed best, achieving a coefficient of determination (R2) of 0.882 and a mean absolute error (MAE) of 17.34 °C on an independent test set, with fivefold cross‐validation further confirming the model's robustness. SHAP interpretability analysis revealed the significant influence of key features such as the number of rotatable bonds, ether bonds, and ether‐linked oxyethylene units on Tg, providing clear guidance for molecular design. External validation demonstrated the model's strong generalization ability. This study not only achieves high‐precision and robust Tg prediction but also highlights the importance of hybrid feature strategies in polymer property modeling, offering a data‐driven foundation for the rational design of polyimides.

Application of Surrogate Models for Approximating Molecular Weight Characteristics of Polymers Under Conditions of Catalytic System Heterogeneity

During the synthesis of polyisoprene on a catalyst, several types of active centers [1] with different kinetic activities are present. This leads to the formation of complex molecular weight distributions and significantly complicates the tasks of process modeling, optimization, and control. Classical kinetic [2] and statistical [3] approaches to modeling make it possible to describe the process dynamics in detail; however, their application in optimal control problems is associated with difficulties caused by the need to perform multiple computational experiments [4]. In this regard, the development of surrogate models capable of reproducing the dependence of key product characteristics on technological and kinetic parameters at substantially lower computational costs is of particular relevance. In the present work, an approach to the construction of surrogate models for approximating the molecular weight characteristics of polymers formed under conditions of catalytic system heterogeneity is considered. The initial information is based on the results of a series of simulation calculations performed using detailed polymerization process models that account for the presence of several types of active centers and differences in kinetic parameters. The generated datasets include information on the initial composition of the reaction mixture, the operating conditions of the continuous process, as well as detailed output molecular weight characteristics of the polymer product. On the basis of these data, surrogate models are trained using regression-based machine learning methods aimed at approximating complex nonlinear relationships. The performed analysis showed that the use of surrogate models provides satisfactory accuracy in reproducing molecular weight characteristics over a wide range of technological parameters and makes it possible to significantly reduce computational time compared to direct simulation modeling. This creates prerequisites for the application of the developed approach in problems of operational analysis, optimization of synthesis conditions, and the development of intelligent decision support systems for the control of polymer production processes.

PolymInsight : An Open Process Simulation Software for Polymerization With Microstructural Quality Indices

Modeling and optimization of polymerization processes incorporating microstructural quality indices are crucial for determining optimal design and operational strategies in polymer production. However, existing polymerization models that account for these indices are often characterized by large‐scale, nonlinear, and coupled equations, presenting significant computational challenges. Moreover, limited research has been conducted on developing algorithms for accurately predicting polymer microscopic properties, and only a few software tools are capable of simulating and optimizing such complex processes. To address these challenges, this study introduces “ PolymInsight ,” an innovative software tool developed in an open‐source Python environment, designed specifically for modeling polymerization processes with microstructural quality indices. PolymInsight supports both dynamic modeling of batch reactors and steady‐state modeling of continuous stirred tank reactors across a variety of polymerization reactions. The software's logical architecture includes a graphical user interface, a robust data structure, a general modeling framework, model reduction techniques, and an adaptive solution strategy. Case studies demonstrate the effectiveness, accuracy, and generalizability of PolymInsight , highlighting its potential as a powerful tool for research and industrial applications in polymer science.

Machine Learning Based Prediction of Mechanical Properties of Unsaturated Polyester Resin Nanocomposites

The use of polymer nanocomposites has become key material in membrane applications, especially unsaturated polyester resin (USPER) based nanocomposites because of their cost effectiveness and tune-able characteristics. Experimental characterization of mechanical properties over a range of reinforcement concentrations is both time-consuming and resource intensive. This paper introduces a detailed machine learning (ML) model in predicting the mechanical properties of egg shell biomaterial-reinforced and nano sand filler-reinforced USPER nanocomposites. At five weight loadings that were expressed as a percentage, four important mechanical properties were studied. Various regression models were tested in a systematic manner to determine the best predictive models of each property. The developed models demonstrated strong fitting performance within the experimentally investigated range, with R2 values above 0.99 for all properties. Ridge Regression showed consistent predictive capability across all investigated properties while providing the advantages of implementation simplicity and regularization. The developed models enabled exploratory trend-based predictions beyond the experimentally investigated reinforcement range. The proposed ML-assisted methodology may help reduce excessive experimental trials and support preliminary material design optimization for membrane-related applications, while additional experimental validation is required for higher reinforcement concentrations.