Operational state assessment of wind turbine gearbox based on long short-term memory networks and fuzzy synthesis

Abstract

To reduce the operation & maintenance cost of the wind turbine and improve its reliability, we propose a novel combined method for real-time operational state prediction, based on the long short-term memory and fuzzy synthetic assessment. After analyzing and filtering the monitoring data of a 2-MW wind turbine gearbox (WTG), we propose a deep learning-based multi-index operational state assessment framework. Following this, the prediction dimensions of each assessment index are established based on the correlation analysis. Meanwhile, we have obtained each index's weight and membership degree after analyzing the prediction error based on Long Short-Term Memory (LSTM). Case studies are performed using three-month Supervisory Control and Data Acquisition (SCADA) data of a 2-MW WTG with fault information. The results demonstrate that the difference between normal and fault state is more prominent when the prediction dimensions with lower correlation are selected. The degree of fault reflected by different assessment indexes is distinguished even under the same state. Then, through reviewing the alarm history of the condition monitoring system, we find that the proposed method can be used to detect the potential failures of the WTG.