Study of optimal design of lattice structures using Deep Learning methods
Tutor/a - Director/a
Estudiant
Carrasco Bañales, Joel
Tipus de document
Projecte Final de Màster Oficial
Data
2024
rights
Accés obert
Editorial
Universitat Politècnica de Catalunya
UPCommons
Resum
This study explores the capabilities of neural networks to accurately predict the material properties of microstructures with complex-shaped void inclusions, while significantly reducing the computational time required compared to traditional computational homogenization methods. The core of this project involves generating datasets that include parameters defining superellipse and superformula-shaped inclusions within unit microcells, alongside the corresponding homogenized tensor components derived through computational homogenization. These datasets serve as the foundation for training a neural network to predict material properties based on the geometric parameters of the microstructural voids. This innovative approach not only integrates geometry, computational homogenization, and neural networks but also represents a substantial advancement in topology optimization. It provides a significant step towards extending microstructural parametric optimizers beyond the limits of finite element methods. The outcomes of this research offer a robust framework for understanding the influence of superellipse and superformula parameters, the mechanics of computational homogenization, the inner workings of neural networks, and the impact on material science and design.
