Posters

Abstract

The current push towards renewable energy production has led to a significant increase of the investments in the wind turbine sector. The profitability of the investments is determined to a large extend by the operation and maintenance costs of the wind turbines. An important driver of these costs are premature failures due to excessive wear. Examples of components that suffer from this are generator and gearbox bearings. Being able to detect these premature failures should make preventive maintenance more efficient and result in lower operational and maintenance costs. The goal of this research paper is to predict generator bearing failures. To this end data from a large operational offshore wind farm is used. Multiple data sources, i.e. Supervisory Control and Data Acquisition (SCADA) data and status log data, are combined or fused. The novelty of the methodology presented here is the use of a combination of data enrichment through rule-based context annotation, and data fusion based on Symbolic Aggregation approXimation (SAX) and pattern mining techniques (i.e. association rule mining, contrast mining and treatment learning). The pattern mining algorithms identify patterns that might be indicative for a future generator bearing failure. These patterns are subsequently used to create an interpretable prediction model for generator bearing failures. The methodology is validated using bearing replacement information made available by the wind turbine operator. Although in this paper the methodology is used only for the prediction of generator bearing failures, it can also be used for the prediction of failures of other types of components.