Machine learning techniques for damage detection in wind turbine blades

December 3, 2021

Damage detection plays a vital role in the operation of machinery and represents one of the greatest challenges in the mechanical, aerospace, and aeronautical industries. The wind energy industry, where wind turbines are subjected to enormous mechanical and aerodynamic loads and extreme environmental conditions during operation, is one of the industries that can benefit most from this kind of study. The installation of wind turbines requires significant investments over a long period concerning maintenance and repairs. In order to reduce these costs and guarantee the integrity and longevity of such structures, the use of a reliable structural health monitoring system coupled with a robust anomaly detection methodology is crucial to assess the viability of the investments during their lifespan. Regarding the subsequent fault detection, numerous machine learning and deep learning techniques have succeeded in a broad range of applications. This thesis aims to implement algorithms capable of consistently detecting and assessing faults in two wind turbine blades of different materials. Such techniques as multivariate Gaussian anomaly detection and anomaly detection autoencoders were successfully incorporated by recognising deviating patterns from the healthy state to its damaged state in experimentally acquired data such as modal parameters and frequency response functions. These vibration-based data were obtained through modal shaker testing, modal hammer testing and pull-and-release testing applying experimental modal analysis and operational modal analysis.

Authors in the community:

• 

  B

  Bernardo Henrique Santos Lopes

  ist181762

Supervisors of this institution:

• 

  N

  Nuno Silvestre

  ist13506

mechanical-engineering - Mechanical engineering

por - Portuguese

Instituto Superior Técnico