Neural network based damage detection on wind turbines

Abstract

In a world where the need for energy is intensifying and the call to address environmental challenges is increasingly pressing, this master thesis strategically merges neural networks and wind energy, with a focus on turbine damage analysis. From initial data analysis to the development of advanced solutions, the research covers a full spectrum. It starts with data collection and analysis, followed by a thorough comparison of various neural network architectures to optimise fault detection. An essential component encompasses the meticulous training and evaluation of a single neural network, supported by an in-depth study of post-processing techniques that significantly improve the accuracy of the analysis. In addition, preprocessing is explored as a key tool to enhance the accuracy of the system. The thesis also highlights an innovative breakthrough: the development of a cascade network composed of six neural networks working in parallel, which significantly extends the detection and diagnosis capability. Finally, the economic and environmental impacts of the project are assessed, providing a comprehensive and pragmatic view that addresses both the need for sustainable energy and environmental preservation.