Gearbox Fault Early Warning for Wind Turbines Based on Mutual Information Recurrent Networks

The damage and failure of the gearbox can directly or indirectly lead to the fault or shutdown of wind turbines, resulting in significant economic losses. To enable early warning of gearbox fault conditions, this paper proposes a fault early warning model for wind turbine gearboxes based on the mutual information recurrent network. Firstly, the maximal information coefficient is utilized to analyze the correlation between the operating parameters and the temperature of the gearbox output shaft. Highly correlated parameters are then extracted as input features for a recurrent neural network with memory capabilities. Following this, the residual distribution of the gearbox temperature under normal conditions is constructed, and a warning threshold is defined to realize early warning of the fault state of the wind turbine gearbox.