FE Surrogate Model:
How to get dynamic response (time domain) through structural parameters?
Machine Learning
SM for DR 💧
Title: A Machine Learning-Based Surrogate Finite Element Model for Estimating Dynamic Response of Mechanical Systems
- Decision trees (DTs) and deep neural networks, 直接输入结构参数，输出time-domain series，精度不会很高
- XGBoost Decision trees
- AdaBoost Decision trees
- RF: random fores
- DNNs: deep NN
LSTM
GNN-LSTM-based
使用GNN从结构 graph中提取结构信息，与振动信息fusion后，输入LSTM进行预测 位移+速度+加速度序列数据
LSTM的输入包括：
- 图嵌入网络提取的结构信息
- ground-motion motion sequence，用了地震数据作为输入, Most of the ground-motion records used in the simulations did not contain velocity pulses.
The LSTM model is thus capable of not only retaining the long-term dependencies inherent in the ground-motion data but also leveraging the structural features derived from graph embeddings.
