Condition Monitoring of Wind Turbine Gearbox Using Multidimensional Hybrid Outlier Detection

Abstract

Gearbox is a crucial but vulnerable component in the drive train of wind turbine. With purpose with condition monitoring of this component, we propose a multidimensional hybrid outlier detection model based on feature extraction and improved Stacked Denoising Auto-encoder (SDAE). First, a multi-dimensional feature extraction model is constructed via time series analysis and time-frequency-domain features extraction. Second, an improved SDAE based framework for condition monitoring is designed through normal behavior modeling. In case study, the originally proposed method is verified by the measured data from 37 wind turbines in two wind farms from two different provinces in China. Furthermore, case analysis, statistical results and comparative experiment are illustrated in detail, which demonstrates that the proposed method can provide early warning of gearbox faults. In industrial applications, early warning can avoid prolonged downtime and increase the power generation time.