Detecção de desbalanceamento de massa em turbinas eólicas utilizando arquiteturas de rede neural Inception

Abstract

Wind energy has experienced significant growth both in Brazil and worldwide, becoming one of the fastest-expanding renewable technologies. In Brazil, this sector gained momentum in 2009 with the first exclusive wind energy auction. Since then, the share of wind energy in the Brazilian electrical grid has grown and currently represents over 13% of the total. As the wind energy sector advances, various challenges arise, including the costs associated with the operation and maintenance of these turbines. One of these challenges is related to mass imbalance, which can affect the performance and lifespan of the turbines. In this context, the use of artificial intelligence algorithms is proposed for mass imbalance detection as a predictive maintenance technique. Early detection of faults in wind turbines is crucial to ensure their efficiency and safety. The application of artificial intelligence techniques, such as machine learning and data analysis, can enable faster and more accurate detection of mass imbalance in wind turbines. Furthermore, it is worth noting that the application of advanced architectures, such as InceptionTime, InceptionTimePlus, and mWDN, has shown promise in addressing issues related to fault detection in wind turbines. InceptionTime, in particular, combines concepts of convolutional neural networks (CNNs) and recurrent neural networks (RNNs), enabling efficient capture of complex patterns in time series data. With the use of these architectures, satisfactory results have been obtained in the detection of mass imbalance in wind turbines, contributing to increasing the reliability and effectiveness of predictive maintenance operations in this rapidly growing sector. The implementation of advanced artificial intelligence architectures provided satisfactory results in the classification of mass imbalance and reduced computational costs, primarily due to the optimization of model training.