Recursive Bayesian estimation of wind load on a monopile-supported offshore wind turbine using output-only measurements

Offshore wind turbine structures experience combined wind and wave loading during their lifetime, and the cyclic characteristics of these loads significantly impact the fatigue life of the support structure. Continuous monitoring of stress-related quantities, such as strain time histories at hotspot locations of the structure, can help to estimate the fatigue life. However, sparse measurements from the substructure necessitate using a digital twin as a tool for structural health monitoring by creating a virtual model. This virtual model relies on real-time input loads for virtual sensing and stress-related quantities prediction. However, the exact loading on these structural systems is often unknown or estimated with considerable uncertainty. This paper implements a recursive window-based Bayesian estimator for input load estimation and virtual sensing of acceleration and strain time history at unmeasured critical locations using output-only measurements. The estimator is numerically and experimentally validated in the context of input load estimation for an offshore wind turbine, and the results are compared with a traditional augmented Kalman filter and a linear regression approach for the quasi-static component of loading. The method is first demonstrated through a numerical study using a finite element model of a 6 MW offshore wind turbine, where input wind load time histories are accurately estimated. Then, the framework is applied to the real data measured from an offshore wind turbine in the North Sea. The study demonstrates the window-based Bayesian input estimator as a promising tool for the digital twinning of offshore wind turbines, demonstrating superior accuracy in input load estimations and full response predictions compared to the augmented Kalman filter and linear regression methods without the constraint of collocated measurements at input load locations.