Structural damage detection based on multivariate probability density functions of vibration data of offshore wind foundations with comparison studies

A multivariate probability density function (PDF) is developed based on the stochastic process of vibration responses from a single sensor on a structure, and used as a feature for damage detection. The change between the PDF of the undamaged state and that of a potentially damaged state, measured by the Kullback–Leibler divergence, is defined as a damage index. This study focuses on vibration-based damage detection for a linear system under excitation and noise; both assumed to be Gaussian white noise. The effectiveness and advantages of the proposed method are demonstrated in a case study of an monopile offshore wind turbine. For comparison, an autoregressive-based method and an autocorrelation function-based method are also studied. The results show the proposed method offers significant performance improvements over the autoregressive-based method and the autocorrelation function-based method, especially when the noise level is high. Although this paper applies the idea of the multivariate PDF feature in a simplified way (i.e., linear system and Gaussian white noise), it paves the way to exploring more sophisticated statistical methods for more complex systems and excitations.