dc.description.abstract | In today's dynamic energy landscape, renewable energy sources are steadily increasing their share of electricity on the grid. The world population recognises the benefits of going carbon neutral and is now willing to invest heavily in this cause. The biggest drawback with renewables like wind and solar is their intermittent nature, thus not being able to meet the demand timely. Retrofitting state-of-the-art machinery is a viable solution for satisfying the increasing need for electricity. Low-emission technology complementary to renewable energy production should be invested in and researched. The goal of this thesis is dedicated to precisely this, studying the potential of innovative solutions for future energy systems.
Research has been conducted at the Vrije Universiteit in Brussels on transforming a micro gas turbine into a micro humidified air turbine. The results have shown numerous benefits, including reduced levels of NOx and increased electrical efficiency. However, there are still areas that need improvement, and in cooperation with the University of Stavanger, a task has been set to develop data-driven models adapted for condition monitoring. These models are built using sensor measurements, which will be used to predict failures and contribute to reliable operation.
In this work, autoencoder models have successfully been trained and evaluated in detail. The task of denoising sensor measurements has produced satisfying results and has significantly enhanced the data quality. These results will be used as a preprocessing step to improve the performance of the multi-layered perceptron developed in association with this project. The second task was to develop an autoencoder model that should be able to give early alerts based on normal- and faulty operational data. A suitable baseline model has been identified for this purpose. However, a residual calculation has not been performed due to the lack of time. The development and analysis of such a model are suggested for future work. | |