Kyusung Lee presents his BSc thesis on "Generation and Evaluation of a Neural Network-Based Sub-Model for Stress Prediction in Finite Element Method"

Abstract:
When it comes to developing a product, evaluating how long it can withstand a certain type of load is an essential aspect of the production process. The product must be very robust in order to ensure a reliable and safe experience for the end user. A common approach to analyzing the strength of a component is performing a Finite Element Method
(FEM) and identifying regions of maximal stress and reinforcing that area. FEM is a powerful tool to detect vulnerable points that could lead to potential material failure.
However, when it comes to computing solutions in sensitive regions, FEM requires intensive modeling effort and it is computationally intensive to generate a fine mesh. This poses
challenges to rapid prediction. Machine learning through the usage of neural network in a sub-model is an attractive replacement as measure to overcome this inconvenience. The
objective of this research is to assess the modeling-efficiency and predictive accuracy of machine learning method in such situations.