Damage Detection and Localization at the Jacket Support of an Offshore Wind Turbine Using Transformer Models

Abstract

Early detection of damage in the support structure (submerged part) of an offshore wind turbine is crucial as it can help to prevent emergency shutdowns and extend the lifespan of the turbine. To this end, a promising proof-of-concept is stated, based on a transformer network, for the detection and localization of damage at the jacket-type support of an offshore wind turbine. To the best of the authors' knowledge, this is the first time transformer-based models have been used for offshore wind turbine structural health monitoring. The proposed strategy employs a transformer-based framework for learning multivariate time series representation. The framework is based on the transformer architecture, which is a neural network architecture that has been shown to be highly effective for natural language processing tasks. A down-scaled laboratory model of an offshore wind turbine that simulates the different regions of operation of the wind turbine is employed to develop and validate the proposed methodology. The vibration signals collected from 8 accelerometers are used to analyze the dynamic behavior of the structure. The results obtained show a significant improvement compared to other approaches previously proposed in the literature. In particular, the stated methodology achieves an accuracy of 99.96% with an average training time of only 6.13 minutes due to the high parallelizability of the transformer network. In fact, as it is computationally highly efficient, it has the potential to be a useful tool for implementation in real-time monitoring systems.