Session: CIE-01-02 AMS General

Paper Number: 115077

115077 - Efficient Mapping Between Void Shapes and Stress Fields Using Deep Convolutional Neural Networks With Sparse Data 

Establishing fast and accurate structure-to-property relationships is an important component in the design and discovery of materials. 
Physics-based simulation models like finite element method (FEM) are often used to predict deformation, stress and strain fields as a function of material microstructure in material and structural systems. 
Such models may be computationally expensive and time intensive if the underlying physics of the system is complex. 
This limits their application to solve inverse design problems and identify structures that maximizes performance. 
In such scenarios, surrogate models are employed to make the forward mapping efficient but the high dimensionality of the input microstructure and the output field of interest may render them inefficient, especially when dealing with sparse data. 
Deep convolutional neural network (CNN) based surrogate models have been typically found to be very useful in handling such high-dimensional problems. 
In this paper, the system under study is a single ellipsoidal void structure under uniaxial tensile load represented by a linear elastic FEM model. 
We consider two deep CNN architectures, a modified Convolutional Autoencoder (CAE) framework with a fully connected bottleneck and a UNet Convolutional Neural Network (CNN), and compare their accuracy in predicting the Von Mises stress field for any given input void shape in the FEM model. 
A sensitivity analysis is also performed using the two methods where the variation in the prediction accuracy on unseen test data is studied with the increasing number of training samples from 20 to 100.

Presenting Author: Anindya Bhaduri GE Research

Presenting Author Biography: Anindya Bhaduri is a Research Engineer in the Probabilistic Design Group at GE Research.
He works on developing surrogate modeling approaches using deep neural networks, stochastic collocation methods, etc. coupled with adaptive as well as space-filled design of experiments for predictive modeling, clustering analysis, and image reconstruction in fields like solid mechanics, epidemiology, microstructure reconstruction, and molecular dynamics.
Before GE Research, Anindya was an Assistant Research Scientist in the Department of Civil and Systems Engineering at Johns Hopkins University where he worked on implementing deep learning techniques in the field of computational solid mechanics.
Anindya earned his Ph.D. from Johns Hopkins University working on developing adaptive surrogate modeling algorithms for efficient uncertainty propagation.

Authors:

Anindya Bhaduri GE Research
Nesar Ramachandra Argonne National Laboratory
Sandipp Krishnan Ravi GE Research
Lele Luan GE Research
Piyush Pandita GE Research
Prasanna Balaprakash Oak Ridge National Laboratory
Mihai Anitescu Argonne National Laboratory
Changjie Sun GE Research
Liping Wang GE Research