To achieve efficient shape design of mechanical parts, we developed a technology that instantly predicts performance changes due to shape modifications using surrogate models. We created 30 data sets on performance changes associated with shape parameter modifications through CAE analysis. Based on the created data, we constructed a surrogate model using a neural network model with high prediction accuracy, achieving a relative error of less than 1%. The surrogate model can predict the performance of unknown shapes in less than one second, enabling the exploration of shapes that meet specifications in 1/10,000th of the time compared to CAE analysis.
Furutani et al. (Wed,) studied this question.