Robust calibration of a multi-physics model of an offshore wind turbine using in situ data

The aim of the work in the thesis is to propose a robust method for data assimilation to be used with the measurements available from EDF's farms, using fluid-structure interaction models and in situ data. The scientific problematic is part of research in vibrational dynamics with inverse problem solving and identification through measurements of the structural parameters of interest in the system in operation. Inverse problem solving is generally complex, and the for this particular context it has been therefore necessary to tackle three scientific problems : (1) how can the process of recalibration of fluid-structure interactions be made robust, taking into account the high number of parameters and the low amount of noise-affected data ? (2) how can this method be adapted to the uncertainties on the environment and the boundary conditions (soil-structure interactions, stochastic loading, time-variable parameters,…) and to the multi-physics aspects in dynamics ? (3) how can the model precision be defined with respect to the amount and quality of the available measurement data (multi-fidelity approach), using relatively simple models (beams), and how can this biased model be enhanced so that it remains sensible ?In order to study these three points the thesis implemented the concept of modified Constitutive Relation Error (mCRE) for the recalibration process. This deterministic concept naturally implies the physics of the problem. It is linked with an error function built with a primal-dual formulation which highlights the reliable pieces of information of the system (equilibrium equations, sensors location,…) and lowers the importance of the less reliable pieces of information (material behavior, boundary conditions, measured values,…). Using the mCRE function also naturally allows to have access to indications on the optimal positioning of sensors and on the error of the model.The data assimilation method development first required to analyze the available in-situ data and select the appropriate and optimized content for model-data gap expression based on the mCRE calibration process and wind turbine models requirements. Second, the mCRE was extended to FSI loads using a physic-based linearization of the continuous mCRE formulation and a relaxable formulation of the offshore wind finite element analysis equilibrium. Finally, the newly defined mCRE extension was implemented in Python into a simulator and calibrator for numerical simulation, data assimilation, and validation of the method.