Research On Damage Detection Of Wind Turbine Support Structure Based On Response Analysis Under Operation Condition

The service environment of offshore wind turbines is harsh,and the supporting structure of offshore wind turbines is prone to be damaged.In order to ensure the safe of offshore wind turbines during operation,it is necessary to conduct the structural damage identification of offshore wind turbine support structures.Aiming at the problem that the vibration response signal under the structural damage state cannot be obtained when the damage identification of offshore wind power support structure is carried out,a scale model experiment of offshore wind turbine support structures using displacement excitation was designed and completed.The damage of structure is simulated by removing the flange connecting bolts of the tower section,the measured acceleration response signals contain structural modal responses and harmonic responses,which are basically consistent with the frequency components of the vibration response signal of the wind turbine support structure under actual operating conditions.Then,the measured signals were analyzed with Variational Modal Decomposition(VMD)to select modal responses containing structural resonance frequencies,and which were used to construct structural damage features.Structural damage identification is conducted with supervised learning algorithm and unsupervised learning algorithm separately.The results prove the effectiveness of the proposed method.The research in this paper is of great significance for identifying the damage of wind power support structure under operating conditions and ensuring the safe operation of offshore wind turbines.The main contents are as follows:(1)The scaled experimental model for damage identification of offshore wind turbine support structures using displacement excitation was designed and completed.Taking the NREL-5WM offshore wind turbine with a single pile foundation as the prototype of offshore wind turbine,an experimental model of the offshore wind turbine support structure with a scale ratio of 40(prototype/model)was made.The numerical model of the prototype of offshore wind turbine was established using Bladed software.The displacement response signal of the hub position for the prototype wind turbine under wind and wave excitation was calculated,and the calculated displacement response signal is input into MTS as the displacement excitation of the test model after similarity ratio transformation,resampling and amplitude modulation,and the acceleration response signal of the test model under the excitation is collected.By comparing with the signal of prototype of offshore wind turbine,it is proved that the frequency components obtained by this test method are consistent with the vibration response signals of the offshore wind turbine support structure under actual operating conditions.(2)Research on structural damage identification based on unsupervised learning algorithms.Firstly,VMD was used to decompose the acceleration response signal collected from the experiment and to extract the structural modal response components(The Intrinsic Mode Function component signal containing only the first-order structural natural frequency),which were then reconstructed.The sliding window method was used to generate the sample collection.The energy and energy ratio characteristics,sample entropy characteristics and time domain characteristics of each sample are calculated to construct the feature matrix.The damage feature matrix was constructed by standardization and PCA dimensionality reduction,and which was input into FCM clustering and K-means clustering algorithms for structural damage degree identification.The minimum accuracy of the two algorithms was 72.22% and72.78%,respectively.The damage identification accuracy of this method needs to be improved,but there is no need to manually define the label.(3)Research on structural damage identification based on supervised learning algorithms.The energy and energy ratio characteristics,sample entropy characteristics and time domain characteristics of each sample are calculated to construct the feature matrix.The damage feature matrix was constructed by standardization.Three supervised learning algorithms,RF algorithm,BP neural network algorithm,and SVM algorithm,are used to identify the degree of structural damage.The minimum accuracy of the three algorithms is 93%,89.5%,and 90.83%,respectively.The method based on supervised learning algorithms has a relatively high accuracy for damage identification,but this method requires manual label definition.