Damage Diagnosis in a Floating Wind Turbine Lab-Scale Model Under Varying Wind Conditions Using Vibration-Based Machine Learning Methods

The diagnosis of early-stage damages including their detection and type identification in a Floating Wind Turbine (FWT) lab-scale model under different wind speed and direction is presently investigated. The problem is of high importance as the timely diagnosis of critical damages may lead to enhanced FWT maintenance and thus operational reliability and performance. The effects of the considered damage scenarios on the FWT dynamics are minor and overlapped by the effects of the different wind conditions resulting in a highly challenging damage diagnosis problem, which is herein addressed via two vibration-based Machine Learning (ML) methods. These are based on data driven modelling of the FWT dynamics under different health states and wind conditions using conventional AutoRegressive (AR) models. The vector of the AR model parameters serves as the feature vector in a Multiple Model (MM)-based and a k-Nearest Neighbours (k-NN)-based method, which are employed for robust damage diagnosis under different wind conditions. The methods' performance assessment and comparison are achieved via hundreds of experiments with the FWT model normally rotating under healthy state, as well as under three different types of early-stage damage including blade crack, small added mass on the blade edge simulating potential ice accumulation and connection degradation at the mounting of the main tower with the FWT floater. The results indicate almost perfect damage diagnosis performance of the MM-based method achieving 99.38% correct classification outperforming its k-NN counterpart.