Análise de desbalanceamento de turbina eólica em escala utilizando rede neural de base radial como classificador

Abstract

Due to the increased concern with environmental issues, sustainable methods of energy production have gained more and more space, and with this, the improvement in the efficiency of these technologies, especially wind power - due to its production power and low production cost - is of short importance. This present dissertation performs a study of the vibration signals extracted from a scaled wind turbine. Masses of 0.5 g, 1.0 g, and 1.5 g were added to the tips of one and two blades, simulating possible problems such as erosion or ice accumulation, in addition to the normal condition, where the three blades and the system were balanced. The system ran at three different speeds: 900 rpm, 1200 rpm, and 1500 rpm. Extended fluctuation analysis (DFA) was used for the pre-processing of the original signals. Then, the vectors were classified by the Radial Base Neural Network model, a pattern recognition technique with supervised training. The classifier achieved a result of 98.83%, 98.15%, and 96.92%, for the rotations of 900 rpm, 1200 rpm, and 1500 rpm, respectively, in the recognition of the patterns under study, being able to differentiate, with indices greater than 96%, the normal operating conditions of the defined imbalance conditions. Furthermore, compared to other methods already used for the same purpose, the results were similar, being lower for 900 rpm, at 0.46%; 4% higher for 1200 rpm; and lower by 1% for 1500 rpm. Finally, the capacity of the classifier studied in the identification of unknown signals is concluded, being important in the identification of possible defects that may arise in the blades of a wind turbine, from the results obtained are very promising and can make relevant contributions in the development of a system for the detection and classification of defects in wind turbine blades.