Electronic Theses and Dissertations

Date of Award

5-1-2026

Document Type

Thesis

Degree Name

M.S. in Engineering Science

First Advisor

Dr. Farhad Farzbod

Second Advisor

Dr. Yiwie Han

Third Advisor

Dr. Ohood Alsmairat

School

University of Mississippi

Relational Format

dissertation/thesis

Abstract

Periodic structures are central to the design of phononic metamaterials, where unit cell configurations determine vibration attenuation and frequency-selective behavior. While many analyses assume interactions are limited to near neighbors, systems governed by long-range forces such as electrostatic coupling exhibit non-local interactions that modify the resulting dispersion manifold. Although Bloch-based formulations accurately resolve these curves, key features like bandgap count and branch-wise energy distribution lack closed-form expressions, often necessitating computationally intensive spectral scanning for parametric studies. To address this, we present a data driven machine learning framework to estimate these characteristics directly from structural geometry. By introducing the per-branch Area Under the Curve as a predictive target alongside bandgap count, the model captures the integrated spectral weight and group velocity characteristics of the system. A hybrid modeling approach, utilizing Random Forest and Gradient Boosting architectures, was developed to map the relationship between geometric parameters and these spectral features. The resulting surrogate achieves 92.15% accuracy in predicting bandgap count and an average R2 of 0.9719 for branch-wise AUC. These results indicate that the framework can serve as an efficient framework for evaluating non-local metamaterial designs, providing a computationally accessible alternative to traditional wavenumber sweeping in long-range coupled system.

Available for download on Wednesday, August 02, 2028

Share

COinS