Assessment of mode shape-based damage detection methods under real operational conditions

The application of vibration-based damage detection methods to a wind turbine model is analyzed in this paper. Practical considerations are evaluated numerically and in experiments: This includes investigations on suitable measurement quantities, i.e. strain and accelerations, and density of measurement positions on the structure. Further it is analyzed if application of the algorithms to results from an output-only modal analysis of the wind turbine excited by wind loads is feasible. Different types of damage are investigated, including tower damage and a change of foundation stiffness. The damage localization methods considered are based on mode shape curvatures. The curvatures are obtained both by mode shape derivatives and by directly using surface strains. Two methods for the localization of damage are proposed in this paper. Basic feature of the first method is the relation between the modal strain energy stored in a section of the structure to that stored in the entire structure [1]. A second method, referred to as Gapped Smoothing Technique [2], is based on the deviation between a mode shape curvature and a smoothed polynomial of the curvature. Both methods are first applied to a numerical model of the wind turbine. The capability to localize damage will be discussed for each method in respect to the utilized density of measurement positions and the type of measurement quantity. Based on the numerical results an experimental setup is arranged, including the equipment of the tower with strain and acceleration sensors. Modal data of the wind turbine model is derived both by an experimental and by an operational modal analysis (OMA). The OMA is conducted combining the Random Decrement technique and a subsequent Frequency Domain Decomposition. The damage localization algorithms are then applied to the experimental data. It is demonstrated that the proposed methods are capable to locate damage. However, challenges remain applying the methods to experimental data. In conclusion advantages and challenges of the proposed approach are discussed