View in EDS

Flow-based generative models for estimating exceedance distributions in unidirectional polymer matrix composites.

Saved in:
Bibliographic Details
Title: Flow-based generative models for estimating exceedance distributions in unidirectional polymer matrix composites.
Authors: Hur, Jihye (Rachel)1 (AUTHOR), Generale, Adam P.1 (AUTHOR), Ballard, M. Keith2 (AUTHOR), Varshney, Vikas2 (AUTHOR), Przybyla, Craig P.2 (AUTHOR), Kalidindi, Surya R.1,3 (AUTHOR) surya.kalidindi@me.gatech.edu
Source: Mechanics of Materials. Jul2026, Vol. 218, pN.PAG-N.PAG. 1p.
Subjects: Probabilistic generative models, Nonparametric estimation, Polymeric composites, Distribution (Probability theory), Machine learning, Microstructure
Abstract: Reliable studies of damage susceptibility in polymer matrix composites (PMCs) are precluded by various challenges associated with establishing probabilistic linkages between the complex details of the material microstructure and its high values of microscale response. Prior works have established this linkage by employing a statistical characterization of microstructure-sensitive extreme value populations of suitable microscale damage descriptors such as fatigue indicator parameters or maximum principal stress. A critical limitation of such methods is the fact that the estimated performance relies heavily on the parametric statistical model (e.g., Weibull) assumed to govern the microstructure-sensitive extreme value populations, limiting the model's predictive capacity across diverse microstructure sample populations and larger microstructural volumes. Moreover, the estimation of defect-sensitive responses conditioned directly on individual microstructures is intractable due to the astronomical number of spatial fiber arrangements realizable in the composite microstructure. This work presents a novel machine-learning framework addressing these limitations and challenges. More specifically, the framework recasts the originally intractable statistical estimation problem into one more manageable to rapidly relate fiber clustering morphologies exhibited in a given volume of the microstructure to its elastic surrogate measure of damage susceptibility, without any parametric assumptions on the governing statistical model. This is accomplished by combining the materials knowledge system framework and modern statistical estimation tools commonly referred to as conditional continuous normalizing flows. The proposed framework is demonstrated to accurately estimate and infer performance characteristics for a variety of PMC microstructures while being trained over an extremely small microstructure dataset, with enormous reductions in inference time and cost. • A probabilistic ML framework for predicting damage susceptibility in PMCs. • Non-parametric, exact density estimation via generative ML. • Fiber clustering effects linked directly to high microscale responses. • Accurate, rapid damage susceptibility estimation with a small microstructure dataset. [ABSTRACT FROM AUTHOR]
Copyright of Mechanics of Materials is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Reliable studies of damage susceptibility in polymer matrix composites (PMCs) are precluded by various challenges associated with establishing probabilistic linkages between the complex details of the material microstructure and its high values of microscale response. Prior works have established this linkage by employing a statistical characterization of microstructure-sensitive extreme value populations of suitable microscale damage descriptors such as fatigue indicator parameters or maximum principal stress. A critical limitation of such methods is the fact that the estimated performance relies heavily on the parametric statistical model (e.g., Weibull) assumed to govern the microstructure-sensitive extreme value populations, limiting the model's predictive capacity across diverse microstructure sample populations and larger microstructural volumes. Moreover, the estimation of defect-sensitive responses conditioned directly on individual microstructures is intractable due to the astronomical number of spatial fiber arrangements realizable in the composite microstructure. This work presents a novel machine-learning framework addressing these limitations and challenges. More specifically, the framework recasts the originally intractable statistical estimation problem into one more manageable to rapidly relate fiber clustering morphologies exhibited in a given volume of the microstructure to its elastic surrogate measure of damage susceptibility, without any parametric assumptions on the governing statistical model. This is accomplished by combining the materials knowledge system framework and modern statistical estimation tools commonly referred to as conditional continuous normalizing flows. The proposed framework is demonstrated to accurately estimate and infer performance characteristics for a variety of PMC microstructures while being trained over an extremely small microstructure dataset, with enormous reductions in inference time and cost. • A probabilistic ML framework for predicting damage susceptibility in PMCs. • Non-parametric, exact density estimation via generative ML. • Fiber clustering effects linked directly to high microscale responses. • Accurate, rapid damage susceptibility estimation with a small microstructure dataset. [ABSTRACT FROM AUTHOR]
ISSN:01676636
DOI:10.1016/j.mechmat.2026.105707