Elasticity Imaging Using Physics-Informed Neural Networks: Spatial Discovery of Elastic Modulus and Poisson's Ratio

Data and codes from our 2022 paper: Elasticity Imaging Using Physics-Informed Neural Networks: Spatial Discovery of Elastic Modulus and Poisson's Ratio

In this study, we leveraged the power of neural networks as universal function approximators to estimate the spatial distribution of elastic modulus and Poisson's ratio using available data from elasticity imaging. Our work is the first implementation of PINN that successfully discovers the spatial distribution of both materials parameters that fully define linear isotropic elasticity. This repository currently contains data and sample code from loading of a domain with embedded stiff inclusion and the PINN implementation.

Quick Setup

install SCiANN on a PC or virtual machine using:

pip install sciann

Running the PINN code

Clone this repository and navigate to the Embedded_Inclusion_PlaneStress directory. Running the parameter discovery code could be done using the following function for a batch size of 1024, 200,000 training epochs:

python PE_PINN.py -bs 1024 -e 200000

The code takes in 1D txt files containing x and y collocation points and strain terms as input and returns distributions of Young's modulus, Poisson's ratio, and stress at those coordinates.

Plotting

One can use the included helper function to plot unstructured data over a structured grid given desired number of points on the x axis

python Plot_PE_PINN_Results.py -nxp 50

Sample input and output files for a 200,000-epoch run on coarse resolution are included as sample data. Have a question about implementing the code? contact us at klaksari@arizona.edu, akamali@arizona.edu