Home > Research > Publications & Outputs > Machine learnt prediction method for rain erosi...

Links

Text available via DOI:

View graph of relations

Machine learnt prediction method for rain erosion damage on wind turbine blades

Research output: Contribution to journalJournal articlepeer-review

Published
Close
<mark>Journal publication date</mark>31/08/2021
<mark>Journal</mark>Wind Energy
Issue number8
Volume24
Number of pages18
Pages (from-to)917-934
Publication StatusPublished
Early online date5/01/21
<mark>Original language</mark>English

Abstract

This paper proposes a paradigm shift in the numerical simulation approach to predict rain erosion damage on wind turbine blades, given the blade geometry, its coating material, and the atmospheric conditions (wind and rain) expected at the installation site. Contrary to what has been done so far, numerical simulations (flow field and particle tracking) are used not to study a specific (wind and rain) operating condition but to build a large database of possible operating conditions of the blade section. A machine learning algorithm, trained on this database, defines a prediction module that gives the feature of the impact pattern over the 2-D section, given the wind and rain flow. The advantage of this approach is that the prediction becomes much faster than using the standard simulations; thus, the study of a large set of variable operating conditions becomes possible. The module, coupled with an erosion model, is used to compute the erosion damage of the blade working on specific installation site. In this way, the variations of the flow conditions due to dynamic effects such as variable wind, wind turbulence, and turbine control can be also considered in the erosion computation. Here, we describe the method, the database creation, and the development of the prediction tool. Then, the method is applied to predict the erosion damage on a blade section of a reference wind turbine, after one year of operation in a rainy onshore site. Results are in good agreement with on field observations, showing the potential of the approach.