Comparação de métodos para detecção de oscilações em unidades industriais afetadas por distúrbios

Abstract

Detecting the presence of oscillations in industrial processes allows to quickly correcting the damaging effects that they produce. Four important methods in the literature are analysed and compared in this work: Autocorrelation Function, Spectral Envelope, Independent Component Analysis and Discrete Cosine Transform. The comparison is made with signals of an industrial plant with multiple oscillation frequencies, the presence of noise and disturbances, and oscillations of different amplitudes. Considering the operations staff requirements, proposals are presented to the different methods to characterize the oscillations that most impact in the variability of the signals analyzed, guiding staff to quickly identify their sources. A methodology is proposed to select data free of outliers and severe disturbances that interfere in the results obtained by the oscillation detection algorithms. The proposed methodology is applied to data collected from an industrial plant. Finally an oscillation detection method is applied a moving window data to evaluate the repeatability of the results obtained on successive analysis.