Uncertainty-quantified parametrically upscaled continuum damage mechanics (UQ-PUCDM) model from microstructural characteristics induced uncertainties in unidirectional composites

• Published in: Computer methods in applied mechanics and engineering 2025-02, Vol.434, p.117571, Article 117571
• Main Authors: Xiao, Yanrong, Ozturk, Deniz, Ghosh, Somnath
• Format: Article
• Language: English
• Subjects: Artificial neural network; Bayesian principal component analysis; Micromechanical response database; Parametrically upscaled continuum damage model; Uncertainty quantification
• ISSN: 0045-7825
• DOI: 10.1016/j.cma.2024.117571

This paper develops an uncertainty-quantified parametrically upscaled continuum damage mechanics (UQ-PUCDM) model for efficient multiscale analysis of unidirectional composite structures. Its constitutive parameters explicitly incorporate representative aggregated microstructural parameters (RAMPs), connecting structural response to the local microstructure. Uncertainty quantification accounts for microstructure characteristic model reduction error, neural network-based model reduction error, and aleatoric uncertainty due to inherent microstructural variability through uncertainty propagation. Development of the UQ-PUCDM framework involves machine learning tools operating on datasets generated by micromechanical simulations. The model quantifies uncertainty in RAMPs due to dimensionality reduction and quantifies the uncertainty in the upscaled elastic stiffness and damage coefficients propagated through ANN. A Bayesian principal component analysis (BPCA) derives probabilistic microstructure-dependent constitutive parameters in the PUCDM model. A Taylor expansion-based uncertainty propagation method enables computationally efficient, time-integration of the stochastic material response with consideration of uncertainty in the RAMPs. Validation studies are conducted with homogenized micromechanical solutions of SERVEs, with Monte Carlo analysis, and limited experimental results with satisfactory agreement. Finally, a single-edge notched beam (SENB) simulation is conducted to explore multiscale damage evolution problems in a structure with uncertainty propagation. [Display omitted] •Uncertainty Quantified Parametrically Upscaled Continuum Damage Mechanics Model•Unidirectional Fiber Reinforced Polymer Matrix Composites•Microstructure Characteristic Model Reduction Error and Bayesian PCA•Machine Learning with Microstructural Response Database•Taylor Expansion-based Uncertainty Propagation