Session: 01-05-01: Applications of Artificial Intelligence/Machine Learning for Aerospace Structures

Paper Number: 107057

107057 - Predicting Stress in Structures Using Convolutional Neural Networks 

Aeronautical structures are subjected to a multitude of loadings and possible damage initiators from natural incidents, e.g. bird strikes. The areas with high internal stresses greatly contribute to potential failures and can reduce the life span of these structures. High-fidelity modeling approaches such as finite element analysis, are used to study such complex aerospace structures. However, these methods are computationally expensive and time-consuming as the scale of the project increases. Therefore, reliable and computationally efficient techniques that are capable of analyzing the performance of these structures will help improve the design process while reducing the risk of potential incidents. As an attempt to reduce the time required to perform these analyses, we propose using deep neural network models to replace the common run-time analysis to produce results instantly. In this work, we implement an encoder-decoder-based convolutional neural network that can be used to predict the stress distribution in a structure based on geometry, external loadings, and boundary conditions. To implement the proposed modeling technique, we study the stress distribution in a 2D plate with different geometries and loading conditions. The data required to train this model is generated using the ABAQUS finite element software through the development of a Python script. This script is a generalized workflow that creates the 2D plates with unique conditions and runs the FEA to obtain the stress data. The developed workflow can be modified to study stress fields under various loading conditions for more complex structures based on the need of the engineer. The model is trained and tested using the data from the workflow and analyzed based on its precision. The results of our analysis show good accuracies for the stress field predictions and offer a reliable technique for the analysis of engineering structures in a fraction of the time it would take the FEA software to produce.

Presenting Author: Ryan Truhn Manhattan College

Presenting Author Biography: Ryan Truhn is a mechanical engineering graduate student at Manhattan College. He has held multiple internships working at Siemens Healthineers, Lizardos Engineering, and currently for Kerelaw Engineering as a senior product development intern. Through his previous experiences, he has developed a passion for product development, aerospace technologies, and data science. He has also previously presented his research "Analysis of a Stationary Bicycle" at the 2022 ASME IMECE conference.

Authors:

Ryan Truhn Manhattan College
Masoud Masoumi Manhattan College