Off-campus SJSU users: To download campus access theses, please use the following link to log into our proxy server with your SJSU library user name and PIN.

Publication Date

Fall 2024

Degree Type

Thesis - Campus Access Only

Degree Name

Master of Science (MS)

Department

Mechanical Engineering

Advisor

Feruza Amirkulova; Ali Tohidi; Ozgur Keles

Abstract

This work presents a comprehensive framework for designing tetrahedral pentamode unit cells to construct highly anisotropic 3D lattice structures that mimic the acoustic properties of water while ensuring manufacturability. The study integrates deep learning, generative modeling, reinforcement learning (RL), and COMSOL Multiphysics simulations to automate and optimize the design process of pentamode metamaterials (PMs). The initial unit cell design is optimized using a forward design approach, adjusting parameters within a manufacturable range, and further refined with hybrid generative modeling. Building on the Conditional Wasserstein Generative Adversarial Networks (CWGAN) model from Amirkulova’s group at San Jose State University, additional features were incorporated, and a Conditional Variational Autoencoder (CVAE) was implemented to predict relationships between geometric parameters and outputs such as bulk modulus, shear modulus, density, and impedance. The optimized parameters were then used to construct models and lattice structures, which were verified in COMSOL to ensure alignment with design and manufacturing requirements. By dynamically adapting learning rates and leveraging the predictive capabilities of the CVAE, the RL framework efficiently explored the design space, achieving significant improvements in mechanical properties, including a bulk-to-shear modulus (B/G) ratio of 170. These results underscore the effectiveness of this data-driven approach in modeling manufacturable PMs structures and devices, paving the way for their rapid adoption in engineering applications and stealth technologies.

Download

Share

COinS