Machine Learning Techniques for Damage Detection in Wind Turbine Blades

In the wind energy industry, wind turbines are subject to enormous mechanical loads and extreme environmental conditions during operation, which is why damage detection methodologies play a vital role in their operation. In order to reduce costs and guarantee the integrity and longevity of such structures, the use of a reliable Structural Health Monitoring (SHM) methodology, capable of detecting structural defects, is crucial to plan and perform maintenance operations in a way to extend the lifetime of these structures. In this context, Machine Learning (ML) techniques have succeeded in a broad range of applications and can be leveraged to develop accurate and automated SHM procedures. In this work, two different methodologies were successfully implemented to recognize deviating patterns from the healthy state, to detect anomalies in the dynamic response of wind turbine blades. One methodology combines automated modal analysis and the classification of modal parameters with a Multivariate Gaussian anomaly detection technique. The other uses Autoencoders (AEs) to classify frequency-domain data of the structure, i.e., frequency response functions or cross-power spectral densities. The response of the structures was obtained through modal shaker, modal hammer, and pull-and-release testing.

Automated summary

In the wind energy industry, damage detection methodologies are crucial for wind turbines due to the mechanical loads and extreme environmental conditions they experience during operation. Machine Learning techniques can be used to develop accurate and automated Structural Health Monitoring methods, detecting structural defects and extending the lifetime of the structures.