Study On The Damage Identification Of Offshore Wind Turbine Support Structure Based On Unsupervised Learning

As the infrastructure of offshore wind turbine,offshore wind power support structure has poor service environment,complex load source and is prone to damage.In order to ensure the safe operation of offshore wind turbines and avoid accidents,it is necessary to conduct health monitoring and damage identification of their supporting structures.At present,the researches on vibration-based structural damage identification mostly use the supervised learning algorithm.However,the supervised learning algorithm has the disadvantages of requiring the data before and after structural damage,huge workload of label labeling,and its efficiency depends on the accuracy of the numerical model.Therefore,the author uses unsupervised learning algorithm to study the damage identification of offshore wind power support structure,designs and completes the scale model test of single-pile offshore wind power support structure,and obtains the acceleration response signals of health and damage state under the excitation of force hammer,white noise and aerodynamic load.These signals are used to verify the proposed damage identification method,and the results prove the effectiveness of this method.The main contents are as follows:(1)The research background and significance of the subject are briefly described,and the current research status of damage identification methods for offshore wind power support structures and structural damage identification based on unsupervised learning algorithm at home and abroad are emphatically introduced.(2)The basic theory and parameter values of PCA,VMD,HT,One-Class SVM and K-Means clustering algorithms are introduced.(3)The damage identification scale model test of single pile offshore wind power support structure is designed and completed.Taking the 5 MW benchmark turbine proposed by NREL in the United States as the prototype with a single pile foundation,a scaled test model with a geometric scale of 1 : 40 was design ed based on the elastic-gravity similarity theory.The model tests of healthy conditions and four different damage conditions under the excitation of force hammer,white noise and aerodynamic load were carried out to obtain the acceleration response signal of the test model.The white noise and aerodynamic load were applied to the top of the test model through the MTS system in the form of displacement load.(4)Research on damage alarming of offshore wind power support structure based on One-Class SVM.The time-domain statistical characteristics are extracted from the measured acceleration response signal,and the correlation analysis and principal component analysis are carried out to construct the feature matrix for damage alarming.The feature matrix is input into One-Class SVM for damage alarming of offshore wind power support structure.The minimum accuracy of model test data is 93.333 %,which proves the effectiveness of the above method.(5)Research on damage degree identification of offshore wind power support structure based on K-Means clustering algorithm.The VMD is used to extract the modal response component of the structure from the measured acceleration response signal,and the Hilbert time-frequency spectrum is obtained by Hilbert transform.The damage feature vector is constructed by the energy of the structural modal response and the Hilbert time-frequency spectrum energy of the modal response.The feature matrix for damage degree identification is constructed by the sensitivity analysis of the damage feature,and the K-Means clustering algorithm is used to identify the structural damage degree.The accuracy of model test data verification of wind turbine support structure under aerodynamic load is 100%,which proves the effectiveness of the proposed structural damage identification method based on K-Means clustering algorithm.