System delivers early prediction of wind turbine failure

A researcher from VUB has developed a system that can predict wind turbine failures caused by early component malfunctions. He specializes in condition monitoring, a technique that uses data from turbine sensors and artificial intelligence to track the machine’s condition. “If operators can anticipate that a specific component is about to fail, they can replace it during regular maintenance, preventing turbine downtime,” says Dr. Xavier Chesterman, who completed his Ph.D. on this complex issue.

Early component failures leading to turbine shutdowns have a significant impact on profitability. On average, an offshore wind turbine experiences 8.3 failures per year. Some components, depending on the turbine type, are particularly vulnerable to defects—typically the generator, gearbox, or subcomponents such as bearings and other moving parts.

Downtime is costly for operators, both offshore and on land. “Replacing these components during routine maintenance can significantly reduce maintenance costs and downtime,” Chesterman explains.

“Predicting and diagnosing wind turbine failures is still an unresolved challenge. A useful methodology should be able to detect different types of failures before they actually occur. It should not only recognize when a component starts behaving abnormally but also interpret patterns in this abnormal behavior to stay ahead of the failure.”

Sensors collect a vast range of data from turbines, including vibrations, abnormal temperature increases, and more. The main goal of this research was to develop an automated fault prediction and diagnosis system for the wind turbine drivetrain. This system used standard data sources, specifically the so-called 10-minute Supervisory Control And Data Acquisition (SCADA) data and status logbook entries.

Chesterman focused mainly on one type of signal: temperature. His system was designed to predict failures and malfunctions in the wind turbine drivetrain by analyzing temperature signals from various components.

“Additionally, the system had to determine the type of fault based on patterns in the turbine’s abnormal behavior,” says Chesterman.

“The system uses artificial intelligence (AI), specifically machine learning and data mining. The vast amount of data makes it difficult for experts to analyze and interpret patterns manually. Sometimes, a combination of different signals is needed to pinpoint where a failure will occur.”

The developed system was tested in real-world conditions using data from three operational wind farms in the North Sea and the Baltic Sea. “Validation showed that the most effective fault prediction methodology could accurately detect certain failures early, with an 80% confidence level.”

For his postdoctoral research, Chesterman aims to take his data analysis a step further. He wants to apply it to other types of machines, such as compressors and agricultural machinery.