Recurrent Neural Networks for Efficient Modeling of Nonlinear Random Vibration Analysis

We propose and demonstrate the use of Long Short-Term Memory Networks (LSTM) to predict dynamic response of nonlinear structures to random vibration. We use LSTMs as an alternative to running expensive finite element analyses to find statistically significant responses. Nonlinear structures are pervasive in the aerospace industry where energy dissipation mechanisms, such as friction or contact, result in nonlinear behavior. The proposed methodology seeks to reduce the overall cost of random analysis while providing sufficient data to quantify uncertainty.

 Speaker: David Najera, ATA Engineering
 Speaker: Adam Brink, Sandia National Labs