Determining the trend behavior of the wind turbine powertrain using mechanical vibration and seasonal wind data

Abstract

Knowing the behavior of a wind turbine promotes better monitoring of its operation and maintenance. Wind data analysis is usually used to optimize power generation, nonetheless it is known that the greater the wind intensity, the greater the wind turbine mechanical vibration. ISO 10816 and 16079 offer guidelines for mechanical vibration instrumentation, emphasize the powertrain as it holds significant importance in the entire equipment. The powertrain has ten vibration sensors, and each one has nine features. We show a strong correlation between the features, the sensors, and between the sensors and the wind data. It can simplify the monitoring system, building a single key performance indicator (KPI) per powertrain using principal component analysis (PCA). We use this KPI for two further actions. First, we calculate the wind from this KPI by regression. If the regressed wind corresponds to the measured one, we can classify the vibration behavior as healthy or abnormally. A reduction in accuracy would show that something has changed and needs evaluation. Second, considering the wind forecast for a time window, we establish whether powertrain vibration will be compatible with healthy operation. Long Short-Term Memory neural network (LSTM) and extreme Gradient Boosting (XGBoost) methods do our regression and prediction tasks. Our results show adequate accuracy for identifying trends and therefore composing alarms in a SCADA system, for example. Our framework contributes to a better monitoring system, as it guides decision-making regarding the operation and maintenance of the wind turbine considering the seasonal behavior of the wind.