Imbalanced Multi-class Classification of Structural Damage in a Wind Turbine Foundation

Damage diagnosis for offshore wind turbine foundations is a topic that remains open in the scientific community due to the importance of increasing safety and ensuring functionality. To deal with the challenge of online and in-service structural health monitoring (SHM) for wind turbines, approaches based on the vibration-response of the structure and captured by sensors such as the accelerometers need to be considered. This work presents a novel methodology to improve the structural damage classification of wind turbine foundations. This methodology consists of several stages. First, the data acquisition to collect and organize the information from the sensors attached to the structure, following the use of a mean-centered group scaling (MCGS) procedure to normalize the raw data and eliminate the difference between the magnitudes of the sensors. Next, a data unfolding to allow the multivariable analysis is performed. Then a linear feature extraction stage is applied to reduce the high dimensionality of the signals. Subsequently, the new feature array serves as the input to a supervised machine learning algorithm which allow to perform the classification task. A five-fold cross-validation procedure is used to obtain the goodness of classification. Several classification performance measures are calculated considering an imbalanced data set obtained with experimental data of a small-scale wind turbine foundation structure to validate the results of the proposed methodology.

Automated summary

This work presents a novel methodology to improve the structural damage classification of wind turbine foundations. The methodology includes stages such as data acquisition, data normalization, data unfolding, feature extraction, and machine learning classification. Through validation with experimental data, this methodology demonstrates good classification accuracy.