This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Application of Artificial Intelligence to Solve an Elasto-Plastic Impact Problem

Journal Article
2021-01-0249
ISSN: 2641-9637, e-ISSN: 2641-9645
Published April 06, 2021 by SAE International in United States
Application of Artificial Intelligence to Solve an Elasto-Plastic Impact Problem
Sector:
Citation: Chavan, S. and Hor, H., "Application of Artificial Intelligence to Solve an Elasto-Plastic Impact Problem," SAE Int. J. Adv. & Curr. Prac. in Mobility 3(6):3001-3012, 2021, https://doi.org/10.4271/2021-01-0249.
Language: English

Abstract:

Artificial intelligence (AI) is dramatically changing multiple industries. AI’s potential to transform Computer-Aided Engineering (CAE) cannot be overlooked. Conventionally, Finite Element Analysis (FEA) is the simulation of any given physical phenomenon to obtain an approximate solution to a group of problems governed by Partial Differential Equations (PDE). Implementation of AI methods in this area combines human intelligence with numerical solutions to make them more efficient. This paper attempts to develop a Deep Neural Network (DNN) model to solve an elasto-plastic impact problem of a symmetric short crush tube made of three materials impacted by a moving wall. A structured learning database was established to train and validate the model using finite element simulations. Tube size, gauge and elasto-plastic material properties were used as input attributes or features. The maximum axial displacement of the tube is the target label to predict. The dataset was analyzed to understand relations among the design features such as section size, gauge, and elasto-plastic material properties. The DNN model architecture and the hyper-parameters tuned to improve the model fit. The trained model was then evaluated with an unseen dataset to assess its prediction accuracy in real-world settings. The output of the DNN model was compared with the finite element simulation results. The established model shows a reasonable co-relation. Thus, this paper offers an alternative or a supplemental tool to conventional FEA methods to accelerate and augment design decision making. This paper offers an alternative or a supplemental tool to conventional FEA methods to accelerate and augment design decision making, and briefs an overall process to set up the Machine Learning (ML) model using TensorFlow and Python libraries. Finally, recommendations are made to propose potential research areas.