Improving Wind Turbine Foundation Damage Detection: A Domain Adaptation Approach with Kernel Mean Matching Method

Damage detection in wind turbine's structure enables to minimize maintenance costs and prevent much larger issues. Different kinds of damages can surface and influence the wind turbine system's variables on different scales. For this study, a crack of 5 mm in two different zones of a scaled Jacket type wind turbine foundation have been performed and measures have been collected while applying a Gaussian white noise vibration. Measurements with accelerometers are taken under three different conditions: without crack, with a crack in one zone and with a crack in another zone of the foundation. Collected data were divided in two binary class datasets (one with a crack in one zone, another with a crack on another zone). Then, kernel mean matching (KMM) domain adaptation method enables to achieve damage detection, transferring the knowledge of a gradient boosting classifier trained on another dataset with different data distribution. Indeed, directly applying the gradient boosting model to the second dataset does not permit to classify correctly damaged and undamaged data. The methodology was validated by calculating an accuracy of more than 98% with radial basis function kernel.