Zhang Liqun 1*, Liu Jun 2

1 Xi’an Jiaotong University, School of Materials Science and Engineering / School of Chemical
Engineering, No. 28 Xianning West Road, Xi’an, Shaanxi 710049, China
2 Beijing University of Chemical Technology, State Key Laboratory of Organic–Inorganic Composites, No.
15 North Third Ring East Road, Chaoyang District, Beijing 100029, China

EXTENDED ABSTRACT: With the continuous advancement of artificial intelligence and multiscale modeling, the prediction and structural optimization of rubber material properties have entered a new stage of development. In this study, six cross-scale modeling frameworks were established for solution-polymerized styrene–butadiene rubber (SSBR) and natural rubber (NR). By integrating data augmentation and machine learning (ML) techniques, systematic investigations were conducted on key properties such as strain-induced crystallization (SIC), tensile stress, and the glass transition temperature (Tg).For the SSBR system, work 1[1] employed all-atom molecular dynamics (MD) simulation data and used the XGBoost model to predict Tg. The combination of ordinary Kriging (OK) and nearest-neighbor interpolation effectively mitigated small-sample limitations, confirming the feasibility of the proposed approach. Work 2[2] further constructed a quantitative relationship model between Tg and structural parameters by introducing generative adversarial networks (GANs) for data augmentation and TPOT for automated model optimization, improving the coefficient of determination (R²) from 0.745 to 0.985. This demonstrated the potential of generative models and automated machine learning in polymer property prediction. Work 3[3] innovatively applied natural language processing (NLP) to polymer sequence representation, combining WGAN-GP-based data augmentation with a CNN–LSTM architecture to achieve high-accuracy Tg prediction (R² = 0.94), highlighting the frontier value of NLP and data augmentation in materials representation. Work 4[4] integrated transfer learning, LSTM–MLP hybrid networks, and XGBoost to bridge the strain-rate gap between MD simulations and experiments, enhancing the predictive accuracy and generalization of stress–strain curves.For the NR system, work 5[5] developed an SIC crystallinity prediction algorithm based on the combined use of XGBoost and GAN data augmentation, significantly improving simulation reliability. Feature importance analysis revealed the regulatory effects of phospholipids and proteins on crystallization behavior. Work 6[6] addressed data scarcity in tensile stress prediction by proposing a three-stage framework consisting of variational autoencoder (VAE)-based data expansion, OK-based virtual sample labeling, and gradient boosting regression modeling, which effectively improved prediction accuracy and model stability.Through the synergistic application of multiscale modeling and intelligent algorithms, this research systematically addressed the key challenges of data scarcity, high computational cost, and cross-scale correlation in rubber property prediction. The proposed frameworks provide new theoretical and technical foundations for the efficient design and performance optimization of NR and SSBR materials.
KEYWORDS: Elastomer; Computational Simulation; Machine Learning
REFERENCES:
[1] Zhan, S.; Huang, W.; Dong, C.; Chen, Q.; Zhao, H.; Duan, P.; Hu, A.; Li, Q.; Li, Y.; Liu, J.; Zhang, L, Mater. Today Commun, 40, (2024) 110181.
[2] Liu, Z.; Huo, Y.; Chen, Q.; Zhan, S.; Li, Q.; Zhao, Q.; Cui, L.; Liu, J, MGE Adv, 2(4), (2024) e78.
[3] Li, Q.; Zhan, S.; Liu, Z.; Dong, C.; Zhao, H.; Yue T.; Zhao, Q.; Zhang, L.; Li, Y.; Liu, J, J. Chem. Inf. Model. 65(14), (2025) 7478–7492.
[4] Zhan, S.; Li, Z.; Zhao, H.; Liu, Z.; Li, Q.; Ji, S.; Zhang, W.; Zhao, Q.; Zhang, L.; Liu, J, Macromol. Rapid Commun, (2025) e00386.
[5] Chen, Q.; Liu, Z.; Huang, Y.; Hu, A.; Huang, W.; Zhang, L.; Cui, L.; Liu, J, Langmuir, 39(48), (2023) 17088–17099.
[6] Hu, A.; Liu, Z.; Chen, Q.; Zhan, S.; Li, Q.; Cui, L.; Liu, J, J. Mater. Inform, 4(3), (2024).[3] A. D. Fuchs, J. A. F. Lehmeyer, H. Junkes, H. B. Weber, and M. Krieger, Journal of Open Source Software 9 (2024), DOI:10.21105/joss.06371