Processamento de sinais para monitoramento de vibrações em unidades geradoras hidrelétricas

Abstract

The vibration monitoring gets highlighted in diagnosing the condition of rotating electrical machines because it is effective in detecting the occurrence of several failures with relatively simple instrumentation. It is critical to ensure safe and reliable operation of hydroelectric generating units, which, being subject to failures of mechanical, electrical and hydraulic origins, offer significant risks and the possibility of large scale accidents. The vibration monitoring systems of these units typically uses proximity sensors, accelerometers and pressure sensors. Signals of such sensors are processed with techniques in the time domain and frequency domain, and features are extracted seeking a diagnosis of the condition of the unit. In this master thesis, signals collected by proximity sensors and accelerometers installed both on a generating unit of Emborcação plant are processed by filtering and decimation, and features are extracted from these signals, for timing analysis. The orbit described by the axis of the unit is analyzed over time and also in the frequency domain with the full spectrum. The power spectral density of the accelerometers signals, in other hand, is estimated by periodograms, and wavelet transform is also used in order to extract useful parameters for condition monitoring. The experimental results obtained with these techniques are shown and discussed. As it turns out, this work is a step of great importance in the development of a monitoring system; it is the study and choice of processing techniques of the system according to the faults you want to detect.