A Surrogate‐Assisted Multi‐Scale Framework for Characterizing the Stochastic Flexural Strength of Plain Woven Composites
- 3 August 2026
- journal article
- research article
- Published by Wiley in Polymer Composites
- Vol. 47 (19) , 17250-17262
Abstract
Plain woven composites in load‐bearing structures often exhibit scatter in flexural strength, due to multi‐scale variability in fiber distribution, yarn architecture, and the resulting spatial heterogeneity of material moduli and strength parameters. This study develops a surrogate‐assisted framework to characterize the propagation of micro‐/meso‐scale structural variability to macro‐scale stochastic flexural strength. Micro‐scale RVEs with randomly distributed fibers and meso‐scale woven units with varying yarn geometry are constructed to quantify their influence on effective moduli and strength parameters. An MLP‐based surrogate framework with output‐specific regression models is established to map stochastic geometric inputs to effective material properties, in which correlation‐constrained training is further introduced to improve fitting performance while retaining the inter‐property correlation trends observed in the multi‐scale simulations. The predicted stochastic material property field is then transferred to the structural finite element model through interpolation and shape function for flexural strength analysis. The proposed framework is assessed by three‐point bending tests for cross‐ply, quasi‐isotropic, and helicoidal laminates. Comparison between the MLP models with and without correlation constraint shows that the constraint mainly improves the fitting accuracy of several output parameters, with the largest relative improvement about 12% for representative outputs. The predicted mean flexural strengths show relative errors of 11.6%, 1.9%, and 5.7% for the three layups, respectively. The predicted coefficients of variation are consistently higher than the experimental values, especially for cross‐ply laminates. This indicates that the framework provides a preliminary estimation of the layup‐dependent trend of strength scatter, while giving conservative estimates of dispersion.Keywords
Funding Information
- Fundamental Research Funds for the Central Universities (No. FRF‐BD‐20‐08A)
- Fundamental Research Funds for the Central Universities (No. FRF‐TP‐22‐029A1)
This publication has 32 references indexed in Scilit:
- Coupled Dimensional Energy Balance and Machine Learning Validation for Ballistic Response Prediction of Fiber CompositesFibers and Polymers, 2025
- A machine learning based prediction model for the impact mechanical response of composite laminates considering microstructure sensitive transverse propertiesPolymer Composites, 2024
- Experimental and multi-scale modeling research on the dynamic tensile properties of open-hole CFRP laminates subjected to strain rates of low to mediumIranian Polymer Journal, 2024
- Multi‐scale elastic properties of 2.5D woven composites with void defectsPolymer Composites, 2024
- Analysis of impact response and damage evolution in multi‐scale of novel 3D carbon fiber‐reinforced polyetheretherketone compositesPolymer Composites, 2024
- An integrated microstructure reconstruction and meshing framework for finite element modeling of woven fiber-compositesComputer Methods in Applied Mechanics and Engineering, 2024
- Investigation of the bending and quasi‐static indentation properties of three‐dimensional (3D) textile‐reinforced compositesPolymer Composites, 2023
- Study on impact resistance and parameter optimization of patch‐repaired plain woven composite based on multi‐scale analysisPolymer Composites, 2023
- Effects of micro voids on the early stage of transverse crack formation in unidirectional compositesComposites Part A: Applied Science and Manufacturing, 2023
- Multi‐walled carbon nanotube grafted 3D spacer multi‐scale composites for electromagnetic interference shieldingPolymer Composites, 2022