A researcher at the Free University of Brussels (VUB) has developed a system that makes predictable the failure of wind turbines due to early component failure. He has specialised in condition monitoring. This uses the data generated by sensors on the wind turbine and artificial intelligence to monitor the state of the machine. "If the operators can foresee that a certain part will fail, they can replace it during normal maintenance so that the turbine does not have to shut down," says Dr Xavier Chesterman, who conducted his research on the complex problem in the Artificial Intelligence Lab research group.
Early failures of turbine components that result in wind turbine downtime impact profitability. On average, an offshore wind turbine fails 8.3 times per year (Carroll et al., 2016). Some components, depending on the type of wind turbine, are prone to failure. These are often the generator, gearbox or subcomponents of those parts, such as bearings and other moving elements.
High maintenance costs
The downtime costs operators, both offshore and onshore, a lot of money. "Replacing those components during normal maintenance, can significantly reduce maintenance costs and downtime," says Chesterman. "Predicting and diagnosing wind turbine failures is currently a problem that has not yet been conclusively solved."
Methodology
A useful methodology should be able to detect different failure types of wind turbines before they effectively occur. The methodology should not only be able to detect the moment when a component starts behaving strangely, but also to interpret patterns in the anomalous behaviour and get ahead of the failure."
Sensorics
The sensors measure a wide range of things on wind turbines, such as vibrations, abnormal temperature rises and more. The main objective of the research was to develop an automatic fault prediction and fault diagnosis system for wind turbine drivelines. For this purpose, data available by default were used, namely the so-called 10-minute Supervisory Control And Data Acquisition (SCADA) and data from the status log.
Temperature signals
Chesterman focused mainly on one type of signals, temperatures. His system had to be able to predict wind turbine driveline failures and errors in advance through the analysis of temperature signals from different components. "Furthermore, the system also had to be able to determine the fault type based on the patterns in the abnormal behaviour of the wind turbine," Chesterman said.
Machine learning and data mining
"The system uses artificial intelligence (AI), more specifically machine learning and data mining. Indeed, the volume of data makes analysing and interpreting patterns more difficult for experts. Sometimes it is a combination of different signals that indicates where the failure will occur."
Wind turbines in three wind farms
The developed system was tested in practice on data from wind turbines situated in three operational wind farms in the North Sea and Baltic Sea. In this study, several different techniques were tested "The validation showed that the most performant fault prediction methodology can detect certain faults accurately and early, with a certainty of 80 per cent.
Follow-up research
Chesterman now wants to go one step further in a follow-up study. He also wants to unleash his data analysis on other types of machinery, such as compressors and agricultural machinery.
Thesis title: Condition monitoring and fault diagnosis for wind turbines under the condition of low-frequency failure and slowly evolving component damage.
Image by Katerina Gicheva via Pixabay
Read also: Hollandse Kust Noord wind farm fitted with solar farm anchor system
Predicting and diagnosing wind turbine failures is currently an unsolved problemXavier Chesterman, who researched and wrote a dissertation on the complex problem
