Errors can happen – but it is important to recognize them. This also applies to the operation of wind power plants. Major damage and failures can be disastrous for the operators of such plants. The anomalies detected by monitoring systems often only indicate that a potential error is present or may occur.
In the future, monitoring systems could additionally include estimates of the underlying root cause. The ARCANA project deals with error diagnosis using deep learning. As a result, the AI system not only issues an alarm, but also specifies the causes that led to a malfunction.

The project is interesting to:

Wind power plant operators, maintenance companies, power plant manufacturers, insurance providers, modelers


  • Further development of anomaly detection in wind power plants towards error root cause analysis based on machine learning
  • Developing an algorithm that offers solutions for practical applications, enabling errors to be remedied more specifically, and error prevention measures to be planned ahead
  • Applying the method on an open dataset and on existing data
  • Publishing the results in a paper

Cyriana Roelofs

Project Manager

Fraunhofer IEE

+49 (0) 561 7294-448


  • Autoencoder, explaining machine learning models, machine learning methods, neural networks, AI methods
  • What are autoencoders? Autoencoders are neural networks and are part of AI methods, which in turn belong to ML methods. They can be used to implement normal behavior models.

Project information

Project participants, Fraunhofer IEE

Cyriana Roelofs, Marc-Alexander Lutz, Stephan Vogt, Stefan Faulstich

Project period

October 2020 – April 2021


Autoencoder-based anomaly root cause analysis for wind turbines

Journal: Energy and AI

Share this Spotlight with your network.

Probabilistic grid condition forecasting: Simplified decision-making