Is this project an undergraduate, graduate, or faculty project?

Undergraduate

individual

What campus are you from?

Daytona Beach

Authors' Class Standing

Anthony LoRe Starleaf, Senior

Lead Presenter's Name

Anthony LoRe Starleaf

Faculty Mentor Name

Siddharth Parida

Abstract

This study builds on previous work developing Finite Element-based Physics-Informed Neural Networks (FE-PINNs), which combine the Finite Element (FE) method with Physics-Informed Neural Networks (PINNs) to solve inverse problems. In earlier research, we introduced the FE-PINN framework and demonstrated its effectiveness in performing parameter regression on simple structural models. In this work, we propose a temporally discrete variant, which extends the original method by discretizing the governing equations not only in space but also in time. We compare the performance of both frameworks --Continuous FE-PINN and Discrete FE-PINN -- under varying levels of noise, initialization errors, and available system measurements. Continuous FE-PINN achieved reliable convergence with up to 30\% initialization error and across all tested noise levels. In contrast, Discrete FE-PINN, while more computationally efficient, was sensitive to noise and performed reliably only in homogeneous media. Both methods required at least two measured degrees of freedom (DOFs) for accurate parameter estimation. Continuous FE-PINN is better suited for noisy environments, while Discrete FE-PINN is advantageous in noise-free settings with limited computational resources. These results provide a foundation for applying FE-PINNs to structural analysis and lay the groundwork for future extensions to more complex systems.

Did this research project receive funding support from the Office of Undergraduate Research.

Yes, SURF

Share

COinS
 

Finite Element-Based Physics Informed Neural Networks

This study builds on previous work developing Finite Element-based Physics-Informed Neural Networks (FE-PINNs), which combine the Finite Element (FE) method with Physics-Informed Neural Networks (PINNs) to solve inverse problems. In earlier research, we introduced the FE-PINN framework and demonstrated its effectiveness in performing parameter regression on simple structural models. In this work, we propose a temporally discrete variant, which extends the original method by discretizing the governing equations not only in space but also in time. We compare the performance of both frameworks --Continuous FE-PINN and Discrete FE-PINN -- under varying levels of noise, initialization errors, and available system measurements. Continuous FE-PINN achieved reliable convergence with up to 30\% initialization error and across all tested noise levels. In contrast, Discrete FE-PINN, while more computationally efficient, was sensitive to noise and performed reliably only in homogeneous media. Both methods required at least two measured degrees of freedom (DOFs) for accurate parameter estimation. Continuous FE-PINN is better suited for noisy environments, while Discrete FE-PINN is advantageous in noise-free settings with limited computational resources. These results provide a foundation for applying FE-PINNs to structural analysis and lay the groundwork for future extensions to more complex systems.

 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.