Using Neural Networks to Obtain the Kernel for Peridynamic Thermal Diffusion Models  
Longzhen Wang a b, Jiangming Zhao a b, Florin Bobaru a  
a University of Nebraska-Lincoln, Lincoln, NE 68588, United States  
\* Email of Corresponding Author: fbobaru2@unl.edu

In peridynamics, the ‘kernel’ in the peridynamic equation determines the nonlocal interaction between points inside the horizon region. Selecting the “right” kernel is important because its influence on results and converge behavior under specific discretization \[1\]\[2\]. In previous studies \[3-6\], the kernels are obtained by simple fitting/matching the heat-flux under constant thermal/concentration gradient conditions (for thermal/mass diffusion, for example). The particular “shape” for the kernel functions can be assumed to be, for example, constant, linearly decreasing towards the edge of the horizon region, Gaussian, etc. It has been noticed that the discretized version of the peridynamic diffusion equation \[4\] shares a similar structure to the mathematical backbone of convolutional neural networks (CNNs). In this presentation, we use a convolutional neural network to train the PD kernel from data (temperature/concentration). For a 1D transient heat transfer problem, we trained a one-layer CNNs using TensorFlow based on the initial temperature and the temperature at 0.1s. Using the resulting kernel, we accurately predicted the temperature at the next step (t = 0.2 s). The error compared with the analytical solution for the classical problem was around 10\^(-12) in the L2-norm. We also discuss future steps in utilizing neural networks in peridynamic modeling of other engineering problems.

References  
\[1\]    Z. Chen and F. Bobaru, “Selecting the kernel in a peridynamic formulation: A study for transient heat diffusion,” Comput. Phys. Commun., vol. 197, pp. 51–60, 2015.  
\[2\]    Z. Chen, J. Woody Ju, G. Su, X. Huang, S. Li, and L. Zhai, “Influence of micro-modulus functions on peridynamics simulation of crack propagation and branching in brittle materials,” Eng. Fract. Mech., vol. 216, no. May, p. 106498, 2019.  
\[3\]    F. Bobaru and M. Duangpanya, “A peridynamic formulation for transient heat conduction in bodies with evolving discontinuities,” J. Comput. Phys., vol. 231, no. 7, pp. 2764–2785, 2012.  
\[4\]    F. Bobaru and M. Duangpanya, “The peridynamic formulation for transient heat conduction,” Int. J. Heat Mass Transf., vol. 53, no. 19–20, pp. 4047–4059, 2010.  
\[5\]    Z. Chen and F. Bobaru, “Peridynamic modeling of pitting corrosion damage,” J. Mech. Phys. Solids, vol. 78, pp. 352–381, 2015.  
\[6\]    J. Zhao, Z. Chen, J. Mehrmashhadi, and F. Bobaru, “Construction of a peridynamic model for transient advection-diffusion problems,” Int. J. Heat Mass Transf., vol. 126, pp. 1253–1266, 2018.


Using Neural Networks to Obtain the Kernel for Peridynamic Thermal Diffusion Models

Division / Sponsors: 
Aerospace, Bioengineering, Applied Mechanics, Design Engineering Division, Advanced Energy Systems, Manufacturing Engineering Division, Noise Control and Acoustics, Nondestructive Evaluation, Diagnosis, and Prognosis (NDPD), Nuclear Engineering Division, Process Industries Division, Fluids Engineering, Heat Transfer, Materials, Microelectromechanical Systems, Safety Engineering and Risk Analysis, Computers & Information in Engineering


Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2020/volume/84478

IMECE2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.