Back to search

University of British Columbia

Application of physics-informed neural network to calibration of a damage-plasticity material model for composites

Abstract

dc:description

Accurate simulation of progressive damage and failure in composite materials relies not only on robust constitutive models but also on precise characterization and calibration of these models, which for analysis of large-scale composite structures is typically represented in a smeared sense by strain-softening curves. Conventional calibration methods rely on iterative processes, necessitating repetitive numerical model runs and extensive physical testing. This study introduces a novel approach using Physics-Informed Neural Networks (PINNs) for the characterization of stress-strain response of composite materials. By integrating PINNs with a coupled damage-plasticity material model, we have developed a pragmatic framework that enhances the accuracy of damage simulations and improves the efficiency of the calibration process by eliminating iterative finite element simulations. PINNs achieve high fidelity calibration by directly satisfying physical constraints and differential equations across the domain, offering a unified framework for integrating data with rigorous mathematical models that involve fewer assumptions. The methodology uses realistic datasets from fracturing coupon-level tests, including noisy temporal full-field displacement data and global force data, without the need for local stress data. A coupled damage-plasticity material model, used as an example for the current PINN development, efficiently captures nonlinearities from material degradation and irreversible deformations. Validation of the developed PINN begins with synthetic data from finite element simulations of notched fracture tests. Challenges in PINN training based on synthetic fracture tests with severe localization are addressed by development of a pipeline of forward and inverse neural networks, enabling accurate prediction of material parameters, localized stress, and plastic strain within the damage zone. Physical experiments further validate the framework, using fracturing coupon-level tests on quasi-isotropic composite laminates. Full-field displacement data is obtained via Digital Image Correlation technique. Finite element models using PINN-characterized constitutive models are compared with experimental data, confirming the PINN's ability to accurately calibrate material parameters and the calibrated material model’s effectiveness in simulating the progressive damage behaviour of composites under both tensile and compressive loading. Additionally, a model discovery exercise is conducted to explore the plasticity and damage sequence in compressive behavior, showcasing the PINN framework's ability to discover suitable constitutive models from experimental data.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy - PhD
Level thesis:degree_level
doctoral
Discipline thesis:degree_discipline
Civil Engineering
Grantor dc:publisher
University of British Columbia
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Abouali, Sahar
Contributors dc:contributor
  • Vaziri, Reza, 1961-

Rights

dc:rights
Statement dc:rights
  • Attribution-NonCommercial-NoDerivatives 4.0 International
Language dc:language
eng

Identifiers

dc:identifier.*
Handle dc:identifier
http://hdl.handle.net/2429/89268

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Abouali, Sahar. Application of physics-informed neural network to calibration of a damage-plasticity material model for composites. doctoral thesis, University of British Columbia, 2024. http://hdl.handle.net/2429/89268