Wind turbine gearbox condition monitoring and fault diagnosis based on multi-sensor information fusion of SCADA and DSER-PSO-WRVM method

Abstract

A gearbox is one of the important parts in wind turbines (WTs). This part has a high downtime rate. In this paper, a structure is presented based on SCADA information fusion, and there is no need for adding new sensors and extra data collection equipment. The proposed method is an intelligent method based on generation and extraction of a feature from residual signal. The behavioural model of gearbox thermal signals is extracted from an Artificial Neural Network (ANN). The feature vector is the input of the feature level fusion algorithm. Relevance Vector Machine (RVM) is used for the feature level fusion. RVM is a kernel-based learning algorithm. Wavelet (W) basis is used as kernel function. To determine the optimal parameters of WRVM, Particle Swarm Optimization (PSO) is used as optimization algorithm. The initial decision-making in each PSO-WRVM block made the decision matrix based on the intensity of the fault. This matrix is the input of the Dempster–Shafer evidential reasoning (DSER) for the final decision-making. The simulation and experimental are conducted based on the data from a real 2.5 MW WT with faulty and healthy gearboxes. Results show that the proposed strategy is successful in diagnosis of gearbox serious damages.