DAMAGE DETECTION ON OFFSHORE WIND TURBINE JACKET FOUNDATIONS BASED ON AN AUTOENCODER

This work addresses the problem of damage detection on offshore wind turbine jacket-type foundations based on deep learning algorithms. The work utilizes data obtained from the vibration response of a lab-scale wind turbine foundation. The main contributions of this manuscript to damage detection are: (i) an autoencoder neural network trained with only healthy data drawing a normality model, and (ii) a threshold in the function of prediction errors to define the bound limits of damage. The methodology is evaluated using real vibration data from the lab-scale wind turbine foundation tagged with different noise levels and damage scenarios. The results of damage detection show a 100% accuracy, demonstrating that the proposed methodology is practical and promising to be employed in this kind of challenges.