Wind turbine fault detection by normal behaviour modelling [Poster]

2016-09-22T11:06:45Z (GMT) by Jannis Weinert Simon Watson
Up-scaling and significant technology improvements have reduced wind energy costs in the last decades. Operational costs, where the fuel is effectively ‘free’, are dominated by maintenance actions. Unscheduled maintenance particularly offshore results in high costs as accessibility is restricted by weather and availability of vessels. Advanced maintenance strategies based on actual condition rather than using corrective or preventive maintenance can reduce these costs. Evaluation of operational data recorded by the Supervisory Control And Data Acquisition (SCADA) system of a wind turbine shows promise for the purposes of condition monitoring as the cost of additional sensors is avoided. Increased temperatures in bearings or the gearbox can indicate imminent failure. Thresholds of absolute values are generally implemented in control systems to avoid overheating. But wear-related changes in the temperature trends are often hidden by normal operational fluctuations in temperature due to the variable speed nature of modern large-scale wind turbines. Normal Behaviour Modelling is a way to reveal hidden trends in temperature signals. This type of model can be used to estimate temperature using additional information from sensors in other components and is trained under normal conditions. After training, the residual of modelled minus measured temperature acts as a potential indicator of failure. Different approaches for Normal Behaviour Modelling are investigated. Artificial Neural Networks can be used as a powerful tool for detecting non-linear relationships. First tests using Artificial Neural Networks and real SCADA data show good results for predicting normal operation.