Turbine failures causing downtime are costly in terms of both production loss and cost of repairs. SCADA (Supervisory Control and Data Acquisition) systems are used to control the turbine, log the history of the turbine states, failures and alarms. The master state and turbine state records (10 minute interval data) can provide useful insights of turbine behavior.

High-frequency sensors can be used for condition monitoring of critical components, but are usually not standard sensor equipment in turbines. These sensors can also be used for diagnostics.

Data mining and machine learning techniques were used on Fjeldskår Wind Farm SCADA data to investigate turbine and vibration problems. The purpose of the analysis was to (1) to familiarize with data mining techniques and assess its usefulness on wind SCADA data and (2) help diagnose the vibration problems on Fjeldskår turbine number 4. The decision tree algorithm CHAID was considered the most useful to classify situations of vibration failures. A low-cost tool for measuring vibrations is also presented. Combined with technical insights and experience, the data mining techniques can serve as a tool for identifying root causes of turbine failures.

Other project participants:

Teknova AS