How Artificial Intelligence Helped Reduce the Impact of Overheating Failures in Wind Turbines by 33%

A recurring failure that revealed a bigger problem

In wind turbines, overheating alarms are more common than one might think, and pose a direct threat to the availability of wind turbines. Despite the cooling systems present in the main components of the nacelle, such as the gearbox, generator, and converter, failures still occur, causing recurring shutdowns.

This is what happened at the Mangue Seco wind farm (comprising 39 Enercon E82 2.0 MW wind turbines), located in Guamaré (RN), where shutdowns caused by the “Failure Excitation: Overtemperature Heatsink” alarm were among the main causes of energy loss. The alarm indicated a failure in the excitation rheostat ventilation system.

The solution to this situation was to set up a predictive maintenance plan based on artificial intelligence and root cause analysis, which generated significant results:

• 33% reduction in the energy impact of these failures
• 60% less downtime due to overheating

In-depth diagnosis: where was the root of the problem?

The study, conducted by technical teams from Delfos, V2i Energia S.A. (the park operator), and Wobben Windpower (responsible for maintenance), identified the excitation cabinet's cross-flow fan as the main culprit. When this fan failed, the heat sink temperature rose, activating the alarm and stopping the machine.

The root cause analysis was performed with the support of the Ishikawa Diagram, which brought together experts to map technical, operational, and environmental factors. The conclusion was clear: it was necessary to anticipate the problem before overheating occurred.

AI in action: true predictive maintenance

Delfos has developed a tool based on artificial neural networks, capable of predicting the temperature of the heatsink based on data such as:

• Ambient temperature
• Nacelle temperature
• Wind turbine output
  

Whenever the actual temperature deviates significantly from the predicted temperature, a predictive alert is issued. This made it possible to identify which machines were at risk of failure before the shutdown occurred.

These alerts began to guide a new preventive maintenance plan, prioritizing machines at greater risk and allowing repairs to take place at strategic times, such as periods of low wind.

The results: energy savings and increased availability

The practical application of this system was carried out during the second half of 2023 and compared with the same period in the previous year.

The results were clear:

In addition, there was an 11% reduction in the duration and 28% reduction in the energy impact of maintenance, even though part of this gain cannot be directly attributed to the new strategy.

Why does this matter to the industry?

This initiative shows how intelligent monitoring of operational data, when combined with machine learning models, can transform the way wind farms are operated.

More than just treating symptoms, the strategy addresses the root cause of failures, reducing costs and improving the efficiency of the entire fleet.

Where to find the complete study

The complete study, “Identification of cooling failures in wind turbines using machine learning” can be downloaded by filling out the form below:

‍

Authors

• Nathianne Andrade (Delfos Energy)
• Letícia Xavier (Delfos Energy)
• Giovanni Aguiar (Delfos Energy)
• Brendo Usandizaga (Delfos Energy)
• Rennan Oliveira (V2i Energia)
  
  

Did you like this content?

Follow Delfos on LinkedIn to keep up with upcoming publications from the Delfos Research project and stay on top of innovations in predictive maintenance and renewable energy.

‍