Degree
Doctor of Philosophy (PhD)
Department
The Department of Mechanical & Industrial Engineering
Document Type
Dissertation
Abstract
Shape memory polymers (SMPs) are a class of stimuli-responsive materials capable of recovering large deformations when exposed to external triggers such as temperature changes. Their distinctive thermomechanical behavior, including the shape memory effect, nonlinear viscoelasticity, and time-dependent softening under cyclic loading, requires a robust and comprehensive constitutive framework. Traditional modeling approaches often struggle to accurately capture the coupled nonlinear, temperature-dependent, and damage-driven features observed in SMPs at large strains. After a brief review of SMP applications in Chapter 1, Chapter 2 focuses on developing a finite-deformation constitutive model rooted in rational thermodynamics. The formulation integrates nonlinear hyperelasticity, viscous dissipation through a multi-branch Maxwell network with nonlinear viscosity, and stress-softening (Mullins effect) using internal state variables. The model is calibrated and validated through extensive experimental testing across various temperatures and strain rates, demonstrating its ability to reproduce complex SMP behavior with high fidelity and computational efficiency. Although physics-based models provide interpretability and adherence to thermodynamic principles, their complexity and calibration demands often become prohibitive in multi-physics environments or large datasets. To address these limitations, Chapter 3 introduces a hybrid physics-informed machine learning (PIML) framework. This approach combines Gaussian Process Regression (GPR) for modeling equilibrium hyperelastic behavior with Recurrent Neural Networks (RNNs) for capturing history-dependent, nonlinear viscoelastic responses. Physical constraints—including objectivity, symmetry, and the Clausius–Duhem inequality—are embedded into the architecture to ensure thermodynamic consistency. The resulting surrogate model captures multiaxial, rate-dependent behavior with robustness against data sparsity and noise, offering a generalizable and physically constrained predictive tool for advanced soft-material simulations. Chapter 4 further advances this direction by developing a physics-informed Temporal Convolutional Network (TCN) capable of learning nonlinear thermo-viscoelastic behavior with Mullins-type damage. The model enforces thermodynamic consistency while delivering accurate cyclic predictions under large deformations. Finally, Chapter 5 presents a Hencky-strain-based, Holzapfel-type constitutive framework for both solid and foam SMPs. Implemented in Abaqus/Explicit and validated against experimental data and numerical benchmarks, this model effectively predicts highly nonlinear thermomechanical responses across diverse loading and thermal conditions.
Date
11-24-2025
Recommended Citation
Ostadrahimi, Alireza, "Modeling Shape Memory Polymers: From Thermodynamic Principles to Physics-Informed Data-Driven Machine Learning" (2025). LSU Doctoral Dissertations. 6964.
https://repository.lsu.edu/gradschool_dissertations/6964
Committee Chair
Guoqiang Li
Included in
Aerodynamics and Fluid Mechanics Commons, Structures and Materials Commons, Systems Engineering and Multidisciplinary Design Optimization Commons