As a source of abundant and clean renewable energy,wind power performs a significant part in achieving carbon neutrality and promoting the transformation of the energy structure.The safe and reliable operation of wind power equipment is the prerequisite for ensuring the effectiveness of its complex electromechanical systems and creating value,and the condition monitoring is an essential instrument to assure the long-term efficient and sustainable operation of wind power equipment.This research is supported by the Lanzhou Talent Innovation and Entrepreneurship Project of Health Assessment for Mechanical Equipment Operation and Maintenance And Intelligent Decision Optimization(No.2018-RC-25),which takes 2.5MW direct-drive wind turbine as the research object,combined with the monitoring data generated by Supervisory Control And Data Acquisition(SCADA)system,clean the wind speedpower data that can characterize the operating status of wind turbine,carry out in-depth research on monitoring the health condition of wind turbine critical components and failure warning around the main bearings and spindles,furthermore evaluate the realtime operating condition of the wind turbine spindle system in a multi-state manner.The main research contents are as follows: 2(1)Investigation of wind turbine wind speed-power data cleaning.A wind speedpower data cleaning approach based on QM-DBSCAN method is proposed.Firstly,the wind speed-power data are classified into four categories by analyzing the causes and distribution characteristics of the anomalous data,which are denoted as Class A,Class B,Class C and Class D.Then,the wind speed-power data are cleaned using the quartile method,DBSCAN algorithm and QM-DBSCAN method respectively.Finally,the cleaning effects of the mentioned approaches are compared and verified.The results indicate that the QM-DBSCAN method has the optimum cleaning effect for wind speed-power anomalies,with the Spearman coefficient improving by 0.0045 and0.0047 compared to the quartile method and DBSCAN method correspondingly.(2)Research on main bearing health condition monitoring of wind turbine.A normal data-driven main bearing health condition monitoring model is developed.First,the main bearing temperature of the wind turbine is adopted as the model output parameter and and the Spearman correlation coefficient method is applied to extract the modeling input parameters.Then,the Prophet normal behavior model,XGBoost normal behavior model and the corresponding combined normal behavior model on the basis of the main bearing temperature are established.Finally,the main bearing failure warning threshold is determined by residual analysis based on the optimal model.The results show that the combined model has the highest prediction accuracy,and its evaluation indexes MAPE are 0.0447 and 0.0006 lower than that of the single model,RMSE is 0.0508 and 0.0047 lower than that of the single model,MAE is 0.0406 and0.0009 lower than that of the single model,the approach provides advance monitoring of abnormal conditions in the wind turbine main bearings and issues early warning signals.(3)Research on wind turbine spindle health condition monitoring.A data-driven approach for wind turbine spindle health condition monitoring is proposed.First of all,the spindle speed is employed as the model output parameter,and the model input parameters are extracted by combining the Spearman correlation coefficient and principal component analysis(PCA).Then the ELM prediction model,SVR prediction model,Elman prediction model and its combined prediction model for the spindle speed under normal operation are established.In the end,the sliding window principle is implemented to calculate the spindle operating condition indicators and to determine the fault warning thresholds for abnormal spindle operating conditions.The results demonstrate that the combined model has the optimal prediction effect,and the early warning time of spindle operation is 5.7 h earlier than the actual fault occurrence time,which enables the identification and early warning of spindle abnormal conditions without spindle-related abnormal data or a priori knowledge of the failure.(4)Assessment of the health condition of wind turbine spindle systems.Polymorphic assessment of the wind turbine spindle system health condition is based on a fuzzy comprehensive evaluation approach.At first,the entropy weight method is applied to calculate the weights of the main bearing temperature and spindle speed.After that,the membership degree matrix is constructed for the four operating conditions(healthy,normal,warning and fault)of the spindle system,which is combined with the weights of each metric parameter to derive the membership evaluation vector and realize the polymorphic assessment of the spindle system health condition.At last,the validity of the method is verified by example analysis of actual monitoring data from selected wind turbines.The results indicate that the wind turbine spindle system health condition assessment methodology according to fuzzy comprehensive evaluation provides an assessment of spindle system performance degradation trend,early detection of potential failures and early warning signals. |