Identifying Constitutive Parameters for Complex Hyperelastic Solids using Physics-Informed Neural Networks(v1.0.0)
The present repository offers foundational code for deepmodulus, a codebase centered on the use of Physics-Informed Neural Networks (PINNs) for material parameter identification. Its primary objective is to facilitate the extraction of elastic parameters governing hyperelastic materials by leveraging loading tests and digital image correlation.
The existing code serves as a valuable tool for extracting the elastic modulus of hyperelastic materials through the analysis of both the loading curve and digital image correlation.
-
Figure-1
In this section, I will provide a brief example demonstrating the execution of the Deepmodulus code.
The code is presently developed using Python 3.7.0 as its foundation, with the primary requirement being Deepxde 1.7.0, which can be found on the official website. Detailed installation instructions can be accessed on the website. Furthermore, the "TF-2.5.yaml" file provides a comprehensive list of all the packages employed by the author. To expedite calculations, it is recommended to install a GPU environment for TensorFlow.
Step 1: Install the Deepxde 1.7.0.
Step 2: Go to the repository of the deepxde package. Replace the “boundary_conditions.py” with the new “boundary_conditions.py” provided. In the new file, we replace the "periodic boundary condition" to the "integration boundary condition" and the "Robin boundary condition" to the "operator boundary condition for the given pointset.
The existing code is designed to work with a rectangular sample containing multiple spherical inhomogeneities. For a comprehensive understanding of its functionalities and capabilities, you can refer to the paper authored by Siyuan Song and Hanxun Jin in 2023, Siyuan Song, Hanxun Jin (2023). Importantly, it is worth noting that the current code is easily adaptable to accommodate complex structures featuring arbitrary inhomogeneities.
The provided example is based on section 3.3 in the referenced source. To execute the code, you'll need to download three components: the "Data" folder, the "Template" folder, and the "Run.py" file. Place all of these items into the same directory. Next, open your Python terminal and input "python Run.py" to initiate the computation. Within the "Run.py" file, you have the flexibility to adjust several parameters, including the speckle data's noise level, the number of epochs, the quantity of speckles, and the number of tests you wish to conduct. It's worth noting that the training process operates at an approximate speed of 50 epochs per second on a Alienware m15 R7 personal laptop. While training is in progress, you can monitor the "variable_history" file, which tracks the evolution of material parameters throughout the training epochs.
In addition, a "Odb_Post_Processing_Deepxde.py" file is provided for you to export the FEM data when you are working on some other application.
Siyuan Song, Hanxun Jin developed this code in 2023. The website is currently maintained and updated by Siyuan Song.
Please cite: Siyuan Song, Hanxun Jin. "Identifying Constitutive Parameters for Complex Hyperelastic Solids using Physics-Informed Neural Networks." arXiv preprint arXiv:2308.15640 (2023)
For any questions, please contact songsiyuanxjtu@gmail.com