Construction of a stress distribution surrogate model using multiple regression analysis and deep learning for predicting fatigue intensity.

Abstract

In recent years, the automotive development has been required to respond to changing market needs and stricter regulations. It is believed that further efficiency can be achieved by utilizing the accumulated CAE data from developments based on the design concept of "Toyota New Global Architecture" that considers overall optimization.

In this paper, we report on the construction of a surrogate model that can predict stress with high accuracy by incorporating shape features such as surface concavity and convexity into the training data, drawing on the methods for constructing thermal boundary surrogate models using "statistical methods" and "deep learning".