EUVIMEDEUVIMED entdecken
Crossref

Prediction of the effective thermoelastic properties of resin-matrix composites with microvoids and analysis of microstructural effects using interpretable machine learning

Yunxiao Zhang, Guangxu Cheng, Xiakang Ma, Cheng Qiu, Xiao Guo

Journal of Composite Materials · 20.11.2026

DOI: 10.1177/00219983261510920 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Composite Materials, online erschienen am 20.11.2026. Verfasst von Yunxiao Zhang und 4 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Gefördert von: Zhejiang Province Key R&D Program Project Funding.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

To address the high computational cost of predicting the thermoelastic properties of resin-matrix composites with microvoids and the limited understanding of how microstructural characteristics affect their macroscopic response, this study develops an integrated framework that combines parametric representative volume element simulations, machine-learning surrogate models, and Shapley additive explanations. A dataset of 2975 high-fidelity RVE simulations has been generated. The dataset spans fiber volume fractions of 0.30∼0.60 and microvoid volume fractions of 0.5%∼5%. Extreme gradient boosting (XGBoost) and artificial neural network (ANN) models are trained to map 14 simulation parameters to 8 effective macroscopic thermoelastic properties. On the 20% test set, XGBoost achieves R 2 = 0.972∼0.994 across the 8 outputs. Its RMSE for the transverse shear modulus G 23 is 0.0780 GPa, 14.4% lower than that of the ANN. For the transverse coefficient of the thermal expansion α 2 , the RMSE is 1.11 × 10 -6 °C -1 . The longitudinal modulus is governed primarily by fiber volume fraction and modulus of longitudinal fiber, whereas transverse properties are more sensitive to microvoids. The coefficients of thermal expansion are controlled mainly by the matrix coefficient of thermal expansion and fiber volume fraction. Microvoids disrupt continuous stress transfer through the matrix and markedly degrade transverse properties. Increasing the fiber content partly mitigates this deterioration and strengthens the constraint imposed by the fibers on matrix thermal deformation. Within the parameter ranges investigated and under the linear-elastic assumption, the proposed framework provides an efficient route for rapid property prediction and microstructural optimization of resin-matrix composites containing microvoids.

Förderung

  • Zhejiang Province Key R&D Program Project Funding

Zitieren

APA

Zhang, Y., Cheng, G., Ma, X., Qiu, C., & Guo, X. (2026). Prediction of the effective thermoelastic properties of resin-matrix composites with microvoids and analysis of microstructural effects using interpretable machine learning. Journal of Composite Materials. https://doi.org/10.1177/00219983261510920

Vancouver

Zhang Y, Cheng G, Ma X, Qiu C, Guo X. Prediction of the effective thermoelastic properties of resin-matrix composites with microvoids and analysis of microstructural effects using interpretable machine learning. Journal of Composite Materials. 2026. doi:10.1177/00219983261510920

BibTeX
@article{Zhang2026,
  author = {Zhang, Yunxiao and Cheng, Guangxu and Ma, Xiakang and Qiu, Cheng and Guo, Xiao},
  title = {Prediction of the effective thermoelastic properties of resin-matrix composites with microvoids and analysis of microstructural effects using interpretable machine learning},
  journal = {Journal of Composite Materials},
  year = {2026},
  doi = {10.1177/00219983261510920}
}
Kostenlos starten →