Detecção de cavitação em válvulas de controle via análise estatística de sinais acústicos

Abstract

Cavitation is a physical phenomenon that causes serious damage to control valve’s internals. Its monitoring and diagnosis are critical to prevent premature degradation of the equipment where it occurs. In this work, acoustic emission and pressure signals have been evaluated to detect incipient cavitation in a pneumatic control valve operating in a water flow control loop. It has been observed, through of the collected data, that standard deviation and skewness of acoustic emission signal are more sensitive to the presence of cavitation, therefore they were used to detect it. With this information, Shewhart charts and Support Vector Machine methods are used to detect cavitation, and the results of the two methods are compared using four performance metrics. The analysis of these results provides important information for choosing the cavitation detection method depending on the complexity, data availability and presence of false diagnosis. Finally, It is evaluated the effect of valve opening on standard deviation of acoustic emission signal using only normality data. To that end, a method was proposed generating a linear model allows detecting cavitation only using normality data. The approach is relevant since, in industrial process, the cavitation data are hard to get.