Early Detection of Gearbox Failures in Wind Turbines Using Artificial Neural Networks and SCADA Data

This paper explores the application of artificial neural networks (ANNs) for the early detection of gearbox failures in wind turbines, using operational SCADA data for predictive analysis. Gearbox failures, which can significantly disrupt wind turbine operations, are critical to maintaining high reliability and efficiency in wind farms. The authors propose a methodology that involves the creation of a normal behavior model using ANNs, trained with healthy condition data from selected SCADA variables closely related to gearbox operations. This model aims to predict deviations in gear bearing temperatures, an early indicator of potential failures. The research uses comprehensive SCADA data spanning from January 2018 to February 2022, from a wind farm comprising multiple turbines. The study ensures the robustness of the model through a meticulous process of data cleaning, normalization, and splitting into training, validation, and testing sets. The ANN model is detailed, featuring two hidden layers and focusing on eleven key variables identified as critical for accurate fault prediction. The results indicate that the model can successfully detect anomalies in gear bearing temperatures several months before a failure occurs, thus providing valuable lead time for maintenance interventions. This early detection capability is underscored by a case study of a gearbox failure in one of the turbines, where the proposed ANN model alerted the imminent issue months in advance of the actual failure.