Most Powerful Open Source ERP

WINDELIN - Predictive Maintenance and Optimisation of Wind Turbines using an open-source Big Data Machine Learning Cloud

WINDELIN is a R & D project developed under Eureka programme using machine learning and big data technologies for anomaly detection and predictive maintenance inside wind turbines.
  • Last Update:2019-04-23
  • Version:002
  • Language:

Predictive Maintenance and Optimisation of Wind Turbines using an open-source Big Data Machine Learning Cloud

Wind Turbines

Wind energy is a fast growing global market in demand of innovative solutions that can optimise operations while reducing production costs. Machine learning and big data have increased competitiveness in many industries by gathering data and using A.I. to predict certain behaviours. During the WINDELIN project, we introduced machine learning and big data to predict the structural health of wind turbines with the same goals of reducing wind farm maintenance costs and increasing competitiveness.

What did we do?

WINDELIN is a Big Data as a Service (BDaaS) solution which is offered to operators of wind farms as as an "all-in-one" subscription service. Subscriptions include sensors, smart data acquisition units (DAU), network connectivity and real time big data analytics through physical modeling and deep learning. Our plug-and-use solution can be tailored to specific needs and scenarios. Subscription revenues are shared among partners through a revenue sharing agreement.

Thanks to WINDELIN project we were able to develop a so-called "smart" sensor, improve our Big Data cloud solution and its underlying databases NEO and MariaDB and explore new AI methods in the field of anomaly and outlier detection. A large part of the project work resulted in significant contributions to generic components of our system from which other parties can also benefit from.

The project itself consisted of three important phases:

1) developing a smart sensor which can utilize GPU (Graphical Processing Unit) and can be embedded in a wind turbine to collect vibration data

2) Use collected data (tens of TBs) from smart sensors to build machine learning models using scikit-learn and other open source python libraries. Models had to be able to successfully "define" what a normal wind turbine behavior is and what is not considered normal.

3) Create a fully functional system solution (BDaaS) which can integrate all sensors, data, wind turbine management, machine learning and control into "all-in-one" subscription service

How we do it?

We used latest achievements in machine learning techniques on top of which we developed specific machine learning models for anomaly detection and predictive maintenance. Over the course of the project we have developed our own smart sensors which are using these machine learning models to detect system anomalies in real-time.

During the R&D phases within the project, it was discovered that state of the art algorithms at present could not cope with the vast amount of incoming data from wind turbines. It required our team to focus on finding ways to reduce the amount of data required to be processed and stay within the limitations of available computing power. We eventually encorporated ideas used in the Large Hadron Collider (LHC) to reduce data transferred and preprocess data directly on the sensor.

What we delivered at the end?

We are successfully using a customized WINDELIN version for a large wind turbine service operator in Germany. At this moment this solution collects and analyzes data in close to real-time for more than 450 wind turbines.

Despite initially targeting only wind turbine operators and farms, our solution was developed with high level of abstraction which allows to use it not only on horizontal but also vertical markets. Such possible markets are all industry related fields where automation of detection and control of physical processes is needed. Such prospects can be considered in the oil industry, the automotive and aviation industries as well as industrial automated manufacturing in general.