Executive Summary : | The use of fossil fuels for energy production leads to pollution and global warming. Wind energy is a valuable renewable source, but its high operating and maintenance costs can lead to breakdowns, higher downtime, and structural, mechanical, and electrical faults. Early detection of these faults is crucial for enhanced energy production, reduced energy costs, and increased turbine plant reliability. This work proposes developing efficient techniques for early fault detection in wind turbines, which will improve turbine reliability, increase uptime, and reduce energy costs. Most literature reports assume a single fault in the turbine, but multiple faults can be present simultaneously. Correlating the spectrum with faults can be challenging due to the similar-looking changes in the spectrum. Machine learning approaches are increasingly being used for condition monitoring, but they struggle to identify faults based on discrete probability distributions. This presents a need for a machine learning-based condition monitoring system capable of detecting a maximum number of faults but a single fault at a time and more than one simultaneous fault.
The proposed experimental setup will simulate more wind turbine faults, focusing on faults associated with bearing, gearbox, generator, and blades. The wavelet decomposed signal will be fed to machine learning algorithms for fault classification. The selection and design of suitable mother wavelets for physics-based efficient signal enhancement and the selection of suitable machine learning algorithms will be attempted. |