-
Physics-Informed Neural Networks with Adaptive Localized Artificial Viscosity
Authors:
E. J. R. Coutinho,
M. Dall'Aqua,
L. McClenny,
M. Zhong,
U. Braga-Neto,
E. Gildin
Abstract:
Physics-informed Neural Network (PINN) is a promising tool that has been applied in a variety of physical phenomena described by partial differential equations (PDE). However, it has been observed that PINNs are difficult to train in certain "stiff" problems, which include various nonlinear hyperbolic PDEs that display shocks in their solutions. Recent studies added a diffusion term to the PDE, an…
▽ More
Physics-informed Neural Network (PINN) is a promising tool that has been applied in a variety of physical phenomena described by partial differential equations (PDE). However, it has been observed that PINNs are difficult to train in certain "stiff" problems, which include various nonlinear hyperbolic PDEs that display shocks in their solutions. Recent studies added a diffusion term to the PDE, and an artificial viscosity (AV) value was manually tuned to allow PINNs to solve these problems. In this paper, we propose three approaches to address this problem, none of which rely on an a priori definition of the artificial viscosity value. The first method learns a global AV value, whereas the other two learn localized AV values around the shocks, by means of a parametrized AV map or a residual-based AV map. We applied the proposed methods to the inviscid Burgers equation and the Buckley-Leverett equation, the latter being a classical problem in Petroleum Engineering. The results show that the proposed methods are able to learn both a small AV value and the accurate shock location and improve the approximation error over a nonadaptive global AV alternative method.
△ Less
Submitted 15 March, 2022;
originally announced March 2022.
-
Experimentally Validated Multiphysics Modeling of Fracture Induced by Thermal Shocks in Sintered UO2 Pellets
Authors:
Levi D. McClenny,
Moiz I. Butt,
M. Gomaa Abdoelatef,
Michal J. Pate,
Kay L. Yee,
Harikrishnan Rajendran,
Delia Perez-Nunez,
Wen Jiang,
Luis H. Ortega,
Sean M. McDeavitt,
Karim Ahmed
Abstract:
Commercial nuclear power plants extensively rely on fission energy from uranium dioxide (UO2) fuel pellets that provide thermal energy; consequently, generating carbon-free power in current generation reactors. UO2 fuel incurs damage and fractures during operation due to large thermal gradients that develop across the fuel pellet during normal operation. The underlying mechanisms by which these pr…
▽ More
Commercial nuclear power plants extensively rely on fission energy from uranium dioxide (UO2) fuel pellets that provide thermal energy; consequently, generating carbon-free power in current generation reactors. UO2 fuel incurs damage and fractures during operation due to large thermal gradients that develop across the fuel pellet during normal operation. The underlying mechanisms by which these processes take place are still poorly understood. This work is a part of our combined experimental and computational effort for quantifying the UO2 fuel fracture behavior induced by thermal shock. In this work, we describe an experimental study performed to understand the fuel fracturing behavior of sintered powder UO2 pellets when exposed to thermal shock conditions, as well as a multiphysics phase-field fracture model which accurately predicts the experimental results. Parametric studies and sensitivity analysis are used to assess uncertainty. Experimental data was collected from multiple experiments by exposing UO2 pellets to high-temperature conditions (900-1200C), which are subsequently quenched in sub-zero water. We exhibit that the fracture results gathered in the experimental setting can be consistently recreated by this work phase-field fracture model, demonstrating a reliable ability to our model in simulating the thermal shock gradients and subsequent fracture mechanics in the primary fuel source for Light-Water Reactors (LWRs). This model advanced the fundamental understanding of thermal shock and property correlations to advance utilization of UO2 as a fuel for nuclear reactors.
△ Less
Submitted 5 December, 2021;
originally announced December 2021.
-
TensorDiffEq: Scalable Multi-GPU Forward and Inverse Solvers for Physics Informed Neural Networks
Authors:
Levi D. McClenny,
Mulugeta A. Haile,
Ulisses M. Braga-Neto
Abstract:
Physics-Informed Neural Networks promise to revolutionize science and engineering practice, by introducing domain-aware deep machine learning models into scientific computation. Several software suites have emerged to make the implementation and usage of these architectures available to the research and industry communities. Here we introduce\linebreak TensorDiffEq, built on Tensorflow 2.x, which…
▽ More
Physics-Informed Neural Networks promise to revolutionize science and engineering practice, by introducing domain-aware deep machine learning models into scientific computation. Several software suites have emerged to make the implementation and usage of these architectures available to the research and industry communities. Here we introduce\linebreak TensorDiffEq, built on Tensorflow 2.x, which presents an intuitive Keras-like interface for problem domain definition, model definition, and solution of forward and inverse problems using physics-aware deep learning methods. TensorDiffEq takes full advantage of Tensorflow 2.x infrastructure for deployment on multiple GPUs, allowing the implementation of large high-dimensional and complex models. Simultaneously, TensorDiffEq supports the Keras API for custom neural network architecture definitions. In the case of smaller or simpler models, the package allows for rapid deployment on smaller-scale CPU platforms with negligible changes to the implementation scripts. We demonstrate the basic usage and capabilities of TensorDiffEq in solving forward, inverse, and data assimilation problems of varying sizes and levels of complexity. The source code is available at https://github.com/tensordiffeq.
△ Less
Submitted 29 March, 2021;
originally announced March 2021.
