Enrichment of Wind Turbine Health History for Condition-Based Maintenance

Abstract

This research develops a methodology for and shows the benefit of linking records of wind turbine maintenance. It analyses commercially sensitive real-world maintenance records with the aim of improving the productivity of offshore wind farms.
The novel achievements of this research are that it applies multi-feature record linkage techniques to maintenance data, that it applies statistical techniques for the interval estimation of a binomial proportion to record linkage techniques and that it estimates the distribution of the coverage error of statistical techniques for the interval estimation of a binomial proportion. The main contribution of this research is a process for the enrichment of offshore wind turbine health history.
The economic productivity of a wind farm depends on the price of electricity and on the suitability of the weather, both of which are beyond the control of a maintenance team, but also on the cost of operating the wind farm, on the cost of maintaining the wind turbines and on how much of the wind farm’s potential production of electricity is lost to outages. Improvements in maintenance scheduling, in condition-based maintenance, in troubleshooting and in the measurement of maintenance effectiveness all require knowledge of the health history of the plant. To this end, this thesis presents new techniques for linking together existing records of offshore wind turbine health history.
Multi-feature record linkage techniques are used to link records of maintenance data together. Both the quality of record linkage and the uncertainty of that quality are assessed. The quality of record linkage was measured by comparing the generated set of linked records to a gold standard set of linked records identified in collaboration with offshore wind turbine maintenance experts. The process for the enrichment of offshore wind turbine health history developed in this research requires a vector of weights and thresholds. The agreement and disagreement weights for each feature indicate the importance of the feature to the quality of record linkage. This research uses differential evolution to globally optimise this vector of weights and thresholds.
There is inevitably some uncertainty associated with the measurement of the quality of record linkage, and consequently with the optimum values for the weights and thresholds; this research not only measures the quality of record linkage but also identifies robust techniques for the estimation of its uncertainty.