Abstract
Applying AI-assisted design on simulation (AI-DoS) requires sufficient training database that is generated by a validated, parametric nonlinear finite element model. This study focuses to investigate a training database design method to achieve the balance between the AI model training accuracy and the computational efforts of the FE model execution. Considering the importance of training data, we will use the K-Medoids to select data points for training set. Due to small data, ensemble learning is introduced to enhance the prediction performance of the ANN model.