This code applied a simple Conditional Variational Autoencoder (CVAE) with feed-forward layers on the problem of estimation of fatigue from coarse (10-minute) Supervisory Control and Data Acquisition (SCADA) system data.
In the following animation the effect of changing the conditioning variables on the estimated cross-section fatigue values is shown.
For more information on the simulation data please refer to our paper.
The dataset is fatigue computations for 1999 different wind conditions, performed with OpenFAST and BECAS.
The dependencies are tensorflow
(tested with version 2.4.0) and tensorflow_probability
(tested with version 0.12.1).
You can run the code in a google colab notebook: