Hi all! I am Gaoyuan Wu (吴高远), a Ph.D. student at Princeton University. I am currently working with Prof. Maria Garlock. My current research focuses on exploring new structural solutions (such as thin-shell structures) to coastal defense using numerical simulations. I am also interested in leveraging the high-performance computing (HPC) Python library JAX to boost structural shape optimization. Outside of the research world, I am an Residential Graduate Student (RGS) at Forbes College, Princeton University.

Before coming to Princeton, I spent 4 years in Shanghai at Tongji University, where I did my undergraduate studies in civil engineering with a focus in bridge engineering.

I am from Chengdu (成都), a city in southwest China. I love our dialects, our amazing food and of course, the PANDAS!

What’s New

• Preprint “JAX-SSO: Differentiable Finite Element Analysis Solver for Structural Optimization and Seamless Integration with Neural Networks” is available online. arXiv link: arXiv:2407.20026. July 2024
• Our new work “Investigating the Effects of Box Girder Bridge Geometry on Solitary Wave Force Using SPH Modeling” was accepted by Coastal Engineering and is available online. DOI: 10.1007/s00158-023-03601-0. PDF is available here. November, 2023
• Our new work “A framework for structural shape optimization based on automatic differentiation, the adjoint method and accelerated linear algebra” was accepted by Structural and Multidisciplinary Optimization and is available online. DOI: 10.1007/s00158-023-03601-0. PDF is available here. Code for this project is available in its repo with a few examples, check them out. June, 2023

Projects

JaxSSO: A differentiable finite element analysis (FEA) solver for structural optimization, enabled by JAX.

Features

• Automatic differentiation (AD): an easy and accurate way for gradient evaluation. The implementation of AD avoids deriving derivatives manually or trauncation errors from numerical differentiation.
• Acclerated linear algebra (XLA) and just-in-time compilation: these features in JAX boost the gradient evaluation
• Hardware acceleration: run on GPUs and TPUs for faster experience
• Support beam-column elements and MITC-4 quadrilateral shell elements
• Shape optimization, size optimization and topology optimization
• Seamless integration with machine learning (ML) libraries

Here is an implementation of JaxSSO to form-find a structure inspired by Mannheim Multihalle using simple gradient descent. (First photo credit to Daniel Lukac)

FEM solver of structural dynamics problem for 2D frame system

This is my final project for APC523 (Numerical Algorithms for Scientific Computing). It can be used for the following problems:

1. 2D-frame system under static loads.
2. 2D-frame system under dynamic loads (such as earthquake motions)

A comparison between the solver and commercial software SAP2000 is shown below, in which the nodal displacement from the solver matches well with SAP2000:

For more, please visit the GitHub repo.