ABSTRACT
This doctoral thesis presents a computational-experimental framework for the analysis, design, and optimization of auxetic metamaterials with enhanced energy absorption performance. The proposed methodology combines experimental characterization, nonlinear finite element analysis, additive manufacturing, structural design, and machine learning to enable the efficient development of lightweight metamaterials with tailored mechanical properties.
The thesis first investigates the mechanical behavior of conventional and auxetic cellular structures through experimental testing and validated nonlinear finite element simulations. The analyses provide insight into the influence of cellular topology, loading direction, and deformation mechanisms on the mechanical response, demonstrating the superior energy absorption capability of re-entrant auxetic structures over conventional honeycomb architectures.
Building on these findings, novel geometrical modifications of the conventional re-entrant unit cell are proposed to enhance its auxetic response and energy absorption capacity while preserving its characteristic deformation mechanisms. The numerical predictions are experimentally validated, confirming the effectiveness of the proposed designs. Compared with the base design, the modifications achieve improvements of up to 1250\% in energy absorption capacity and 860\% in specific energy absorption.
To accelerate the design process, a machine learning-based forward and inverse design framework is developed using datasets generated from validated finite element simulations. The proposed methodology enables efficient prediction of nonlinear mechanical responses and inverse identification of unit-cell geometries for both uniform and thickness-graded auxetic structures with prescribed mechanical performance. Furthermore, a physics-guided machine learning framework that combines a layer-wise response decomposition strategy with deep neural networks is developed to substantially reduce the computational cost of dataset generation while enhancing predictive performance. The proposed framework achieves prediction errors below 2.5\%, including in extrapolation scenarios, using only 10 training samples, compared with 2,145 samples required by the conventional deep neural network.
Overall, this thesis establishes a framework integrating computational mechanics, experimental validation, structural optimization, and physics-guided machine learning for the efficient design of advanced auxetic metamaterials. The proposed methodologies provide practical tools for developing lightweight structures with tailored mechanical performance and enhanced energy absorption, while providing a methodological basis for extending data-driven design approaches to other metamaterial systems.
PhD Advisors:
-
- Dr. Narges Dialami
- Dr. Miguel Cervera
CANDIDATE
Sima Farshbaf is a PhD candidate in Civil Engineering at the Universitat Politècnica de Catalunya (UPC) and CIMNE. Her research lies at the intersection of computational mechanics, metamaterials, additive manufacturing, and machine learning. She develops computational and data-driven frameworks for the design and optimization of auxetic metamaterials, combining experimental validation, finite element modelling, and artificial intelligence to enhance the mechanical performance and energy absorption capacity of lightweight structures.