Exploring the Utility of Machine Learning Algorithms to accelerate the Equilibrium Convergence of Finite Element Frameworks
In this work, a Machine Learning (ML) approach based on Artificial Neural Networks (ANN) is explored as a potential tool to predict the solution period for a finite element to converge to an equilibrium solution in less amount of time. For the same, elastic deformation of a 2-D triangular element and a 2-D bi-linear quadratic element were studied. Finite element deformation codes were implemented in Matlab. Deformation was applied to the elements to study the number of iterations; the element will take to attain equilibrium and for the solution to converge. The Newton-Raphson iterative method was used to study this iterative approach. Datasets were then extracted from the classical finite element codes by applying various boundary conditions to the elements and these data sets were used to train the ANN model. The iterations obtained from the Neural Network approach were then compared with those obtained from the classical finite element approach for all the elements. Results indicated that the Neural Network model is extremely robust and can provide near-accurate convergence criteria for the finite elements to converge by reducing the number of iterations to attain equilibrium. Advisor : Dr Saurabh Basu Co Advisor : Dr Christopher Carson McComb
Penn State Only
Files are only accessible to users logged-in with a Penn State Access ID.
|Work Title||Exploring the Utility of Machine Learning Algorithms to accelerate the Equilibrium Convergence of Finite Element Frameworks|
|License||In Copyright (Rights Reserved)|
|Work Type||Research Paper|
|Deposited||November 03, 2021|
This resource is currently not in any collection.