Temperature monitoring and AI help reduce maintenance costs and downtime of wind turbines
A researcher at VUB has developed a system that can predict wind turbine failures caused by early-stage component faults. He specialises in condition monitoring, using data from turbine sensors combined with artificial intelligence to track the machine’s health. “If operators can foresee a part is likely to fail, they can replace it during regular maintenance, avoiding costly downtime,” says Dr Xavier Chesterman.
Early failure of turbine components leading to downtime has a direct impact on profitability. On average, an offshore wind turbine breaks down 8.3 times a year. Certain parts—depending on the turbine model—are particularly prone to defects. These are often the generator, gearbox, or subcomponents such as bearings and other moving elements.
Unplanned downtime is expensive for operators, whether offshore or on land. “Replacing those components during routine maintenance can significantly reduce both maintenance costs and downtime,” says Dr Xavier Chesterman. “Predicting and diagnosing wind turbine failures is still a challenge that hasn’t been fully resolved. Any useful method needs to detect different types of faults before they actually occur. It has to be able not just to detect the moment a component starts behaving abnormally, but also to interpret patterns in that abnormal behaviour—and catch the fault in time.”
Turbine sensors monitor a wide range of parameters, including vibrations, unusual increases in temperature, and much more. The core aim of Chesterman’s research was to develop an automated fault prediction and diagnosis system for the drivetrain of wind turbines. He worked with data that is already commonly available—namely the so-called 10-minute Supervisory Control And Data Acquisition (SCADA) data and information from the status logbook.
Chesterman focused primarily on one type of signal: temperature. His system had to be capable of predicting drivetrain faults and failures in advance by analysing temperature signals from different components. “In addition the system also had to identify the type of fault, based on patterns in the turbine’s abnormal behaviour,” he explains. “The system uses artificial intelligence (AI), specifically machine learning and data mining techniques. The sheer volume of data makes it difficult for experts to interpret these patterns on their own. Often it's a combination of different signals that points to where a fault will occur.”
The system was tested using real-world data from three operational wind farms in the North Sea and the Baltic Sea. “Validation showed that the most effective fault prediction method could accurately detect certain failures early, with a confidence level of eighty percent,” says Chesterman.
He now plans to take his research a step further during his postdoctoral work. He aims to apply his data analysis approach to other types of machinery, such as compressors and agricultural equipment.
