Condition Monitoring Of Wind Turbine Components Model Based On The Principal Component Analysis

In order to make the operation of wind turbine stably and reliably, and reduce the maintenance cost, monitoring the running status of wind turbine is the most effective means. By means of timely and effectively analyzing the information gathered from the state monitoring system, and using feasible monitoring analysis, condition monitoring of wind turbine can be realized effectively. The research of wind turbine’s condition monitoring is good for reducing the maintenance cost and promoting the riding quality, especially for offshore wind farm.The results of traditional condition monitoring and fault diagnosis technique usually can not be satisfactory, which is mainly due to the complicated correlation of so many variables. Principal Component Analysis (PCA) for modeling the key components of wind turbine was used in this paper, which can compress the correlated variables to several independent variables, and this has great practical value.With modeling and analysis of vibration signals, incipient failure of key components such as tower, drive train and rotor could be detected. In this paper, with good understanding of the tower vibration dynamic characteristics and its main influence factors, a tower vibration model was established. The merits and physical interpretation of PCA are also discussed and analyzed. With analyzing the SCADA data of one wind turbine during March to May in2011, PCA model for tower vibration in normal working condition was established and validated by simulation and calculating.For the reason of data uniform distribution after standardizing the data of PCA, relative changes of principal component analysis was introduced in modeling the vibration performance of wind turbine. The validating results shows that some misinformation has been eliminated.Adaptation of the model for metabolic climate influence factor is necessary, as wind speed and temperature is fluctuant. Self-adaption principal component analysis was introduced to solve the problem. By using sliding window to update the model, it is proved to be reliable.