This thesis is a contribution to the field of wind turbine maintenance management. The first chapters provide a review of wind turbine maintenance management, in particular the motivations for reliability-centred maintenance. The data requirement of this maintenance approach is considered, and the role of work order free text data as an information source are highlighted. Methods from the field of text mining are however required to extract this information in an actionable format, and an overview of the most relevant text mining approaches is given in the final chapter of the literature review.
The main output of this work is a supervised text mining algorithm for structuring maintenance data that is recorded as free text work orders. The method is applied on two datasets of SAP work orders from major onshore wind farms in Scotland. Common issues found in the raw data are highlighted and data cleaning rule sets are developed to overcome these issues. A lexicon of domain terminology is developed that can be used on these datasets as well as extender for wider use. The methodology is developed in Matlab and consists of nine modules for data cleaning, vectorisation, transformations, supervised prediction of missing values. The outputs are given both as a two-level Pareto chart and frequency tables that allows their use in maintenance decision-making. Results are analysed in terms of algorithm performance and validated against the research aims.
Improvements are also suggested to reduce supervision requirement, raise accuracy, and make the approach more universal in terms of turbine models and terminology. Finally, the economic benefits of automated work order mining, and potential ways to increase its industrial appeal are discussed.
- University Of Strathclyde
- McMillan, David, Supervisor
- Tuohy, Paul Gerard, Supervisor
|Award date||7 Nov 2017|
|Place of Publication||Glasgow|
|Publication status||Published - 25 Aug 2017|
- onshore wind
- maintenance cost