Assessing the Utility of Early Warning Systems for Detecting Failures in Major Wind Turbine Components

This paper provides enhancements to normal behaviour models for monitoring major wind turbine components and a methodology to assess the monitoring system reliability based on SCADA data and decision analysis. Typically, these monitoring systems are based on fully data-driven regression of damage sensitive-parameters. Firstly, the problem of selecting suitable inputs for building a temperature model of operating main bearings is addressed, based on a sensitivity study. This shows that the dimensionality of the dataset can be greatly reduced while reaching sufficient prediction accuracy. Subsequently, performance quantities are derived from a statistical description of the prediction error and used as input to a decision analysis. Two distinct intervention policies, replacement and repair, are compared in terms of expected utility. The aim of this study is to provide a method to quantify the benefit of implementing the online system from an economic risk perspective. Under the realistic hypotheses made, the numerical example shows for instance that replacement is not convenient compared to repair.