An AI-empowered system enabling wind farm operators to lower the excessive cost of maintaining their operational assets

Despite rapid decrease of renewable energy production cost, the cost of operating and maintaining offshore wind turbines is estimated at 25% of their installation cost.

Modern offshore wind turbines operate in hazardous environments in the sea creating significant health and safety challenges for crews carrying inspection, survey or research work. Accurate diagnostics produced by our prototype system will ensure that engineering personnel are sent only when need is critical. Our primary customer segments are the Wind Farm Operators and we aim for a diversified customer business model that will consider both offshore and onshore wind farm operators.

Developed over the last 3 years by Dr Adrian Stetco from The University of Manchester’s Department of Computer Science while working in the EPSRC’s project, the system promises to unlock new efficiencies in the condition monitoring of wind turbines.

By utilising state-of-the-art deep neural network architectures our system achieves 99% accuracy in predicting operational states and demagnetisation failures in generators running in real time. The proposed model works on raw sensor signals without the need of expert preprocessing and provides actionable support for engineers through dynamic web interfaces.

Adrian says “We are really excited to see these new automated technologies outperforming their traditional counterparts by a significant margin.  Our hope is to see them ultimately deployed on every wind turbine, reducing their O&M costs and ultimately making renewable energy the most affordable kind of energy.”

2Wind Features Include: 

  • Novel approach: Our previous research (published in Renewable Energy and IEEE Conf. on Big Data) has identified a lack of penetration of deep learning AI technologies in wind turbine condition monitoring. Unlike traditional physics-based or data-driven methods, this approach does not require expensive, error-prone expert intervention for feature pre-processing. The models make use of convolutional neural networks which automatically learn the discriminating patterns in electric signals.
  • High Performance: We have achieved state-of-the-art accuracy when predicting operational state and faults in a lab generator.
  • Highly customizable: The system can be customised to the specific needs of individual customers. Our research demonstrates the applicability of this approach to generators as well as multi-chip power electronics; in the future we may customise the system to electric motors in electric cars.
  • Lower Cost and Less Time: Because of the automatic way the models are constructed from electric signals, there is no need for engineers spending time (and associated cost) working on signal feature pre-processing for each individual model.
  • Lower Risk: As a side effect of increased accuracy, we believe the system will result in engineers being sent to repair wind turbines only when absolutely necessary hence reducing safety and hazard risk.
  • Convenience: Highly dynamic web-based visualisation tools are used to inspect and send email notifications when predictions indicate damage.

For more information, please contact:

Erol-Valeriu Chioasca
+44 (0)161 606 7243 (office)*
+44 (0)7760 551 355 (mobile)

*As a result of lockdown related to Covid-19, office landline numbers are not currently being monitored. Therefore, please email or phone an alternative number, where possible.