Neural nets for mechanics
Project Description

We intend to use deep learning methods to build accurate and efficient mechanical models. In other words, we want to train machines to learn mechanics. Once trained, it can then provide fast predictions of mechanical responses of certain systems. The current targeted problem is the mechanical property prediction of meta materials. Meta materials are a new class of man-made materials of which the properties depend on the internal structure of their building blocks instead of their chemical constituents. Accurate and fast prediction of their properties is vital in the design of these materials.

Supervisor
YE Wenjing
Co-Supervisor
YE Wenjing
Quota
2
Course type
UROP1000
UROP1100
UROP2100
UROP3100
UROP4100
Applicant's Roles

Students are expected to work closely with PhD students on machine training (hyper parameter tuning) and generating ground-truth data using FEM commercial software.

Applicant's Learning Objectives

Students will learn (1) machine learning methods, (2) how to construct neural networks and (3) how to use FEM commercial software.

Complexity of the project
Moderate