Early Fault Detection of Wind Turbine Gearbox Based on Adam-trained LSTM

The gearbox is an important device in the wind power generation process. An early fault detection of the gearbox can identify the current status and discover the abnormal operation in advance, thereby providing enough time to take risk mitigation measures. This paper proposes an early fault detection strategy of the gearbox oil temperature in the wind turbine gearbox based on the adaptive momentum (Adam) optimized long short-term memory (LSTM) network. First, feature variables closely related to the gearbox oil temperature are selected to construct the model inputs; Then the data under normal operating conditions are used to develop the LSTM model trained with the Adam algorithm. Residual errors are processed by a sliding window and T-distribution to determine the alarm threshold and achieve the fault detection. The proposed method is verified with field data of a wind farm in Inner Mongolia, and the results show that the method can detect oil high temperature fault nearly one hour in advance, which can provide sufficient time to make a maintenance.