Prediction of Wind turbine Gearbox Condition Based on Hybrid Prognostic Techniques with Robust Multivariate Statistics and Artificial Neural Networks

摘要

This paper presents a novel methodology for encapsulating the health of wind turbine gearbox by employing state of the art robust multivariate statistical techniques on low frequency SCADA data. A physics of failure model is employed in conjunction with a robust derivative of the Mahalanobis distance in order to ascertain current gearbox behaviour. The technique is normalised for external conditions, turbine loading and previous behaviour; allowing the technique to be readily applied by industry to various gearboxes. The model is shown to accurately encapsulate current behavioural conditions. Following this, prediction of future gearbox condition is performed through the use of artificial neural networks. An effective prognostic horizon of up to 14 days is found whilst employing suspended data, with prediction certainty increasing as the horizon is reduced. This allows for the effective scheduling of offshore maintenance actions to ensure high availability whilst simultaneously reducing maintenance costs.