Eventos Anais de eventos
DINAME 2023
XIX International Symposium on Dynamic Problems of Mechanics
A Physics-Based Machine Learning Model for Fatigue in Wind Turbines Devices
Submission Author:
Tiago Bastos Moscon Micco Puntel , RJ
Co-Authors:
Tiago Bastos Moscon Micco Puntel, Fernando Rochinha
Presenter: Fernando Rochinha
doi://10.26678/ABCM.DINAME2023.DIN2023-0052
Abstract
This paper aims, using a deep generative model, which can be viewed as a probabilistic surrogate, to build a digital representation component of a wind turbine that assesses bearing accumulated fatigue damage for real-time application, considering lubricant condition and 10-minute SCADA summary statistics: mean wind speed at the hub height and turbulence intensity. The stochastic generator creates a set of distinct wind flow samples to account for the uncertainty in the flow characterization. The aero-servo-elastic simulator provides the load time series that will be evaluated in a post-processing routine to give the accumulated damage distribution for each input. The analysis shows that The deep generative model performs well in capturing the statistical quantities, such as the mean and quantiles. However, the resulting distributions of some inputs in the domain edges and low-density regions have discrepant standard deviation values from the data distributions, indicating that the model fails to fit them adequately in these areas. Finally, this work presents application examples for the developed digital representation.
Keywords
wind turbine, bearing damage, machine learning, deep generative models, stochastic systems
