Graduation Semester and Year
Fall 2025
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Mechanical Engineering
Department
Mechanical and Aerospace Engineering
First Advisor
Rassel Raihan
Abstract
Polymer matrix composites (PMCs) offer outstanding strength-to-weight ratios but remain vulnerable to environmental degradation, particularly moisture absorption. Absorbed water alters polymer chain chemistry through plasticization and molecular bonding, resulting in reduced mechanical and dielectric integrity. To account for the limited understanding of the underlying absorption mechanisms, engineers adapt to conservative design methodology, assuming a composite structure is already saturated with moisture, while it is not. This dissertation presents a unified experimental, computational, and physics-informed machine learning framework for modeling and predicting moisture-induced degradation in glass fiber–reinforced polymer (GFRP) composites. In this study, Broadband dielectric spectroscopy (BbDS) was first used to identify polarization mechanisms and dielectric permittivity associated with moisture absorption. The dielectric permittivity and relaxation strength increased with exposure time and stabilized at saturation, correlating strongly with moisture concentration and reductions in tensile and flexural strength. This established dielectric sensing as a viable non-destructive tool for assessing moisture-related damage. Data-driven Machine learning models were then developed to estimate the composite’s moisture content from its dielectric response. Supervised classification and regression models, including support vector machines and multilayer perceptron networks, accurately predicted the material’s saturation state with coefficients of determination above 0.95, revealing the dielectric features most sensitive to hygrothermal aging. Then, a multiscale–multiphysics finite-element framework was developed to mechanistically couple moisture transport and dielectric property evolution. The non-Fickian hindered diffusion model (HDM) distinguished free and bound water diffusion, incorporated interphase heterogeneity, and linked the resulting moisture fields to Maxwell’s electromagnetic equations. The coupled HDM–Maxwell model captured experimental trends, showing that a ~2.5 wt% moisture uptake produced a ~75 % increase in dielectric permittivity. Finally, a physics-informed neural network (PINN) was developed to solve the coupled HDM–Maxwell system. Using Modified Fourier architectures, signed-distance weighting, and adaptive residual balancing, the network achieved over six orders of magnitude loss reduction and good validation accuracy. Thus, the framework establishes a mechanistic understanding of moisture diffusion, molecular binding, and dielectric behavior, providing a scalable foundation for digital-twin-based monitoring and lifetime prediction of advanced polymer composites.
Keywords
Polymer Composites, Non-destructive Testing, Moisture Absorption, Non-Fickian Absorption, Structural Health Monitoring, Dielectric Analysis, Computational Analysis, Physics Informed Neural Networks, Machine Learning, Glass Fiber reinforced Polymer Composites, Durability Analysis
Disciplines
Other Mechanical Engineering | Polymer and Organic Materials | Structural Materials | Structures and Materials
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Das, Partha Pratim, "A MULTIMODAL PREDICTION FRAMEWORK FOR MOISTURE AGING ASSESSMENT IN POLYMER MATRIX COMPOSITES" (2025). Mechanical and Aerospace Engineering Dissertations. 438.
https://mavmatrix.uta.edu/mechaerospace_dissertations/438
Included in
Other Mechanical Engineering Commons, Polymer and Organic Materials Commons, Structural Materials Commons, Structures and Materials Commons