A Neural Network Approach to Analyse Cavitating Flow Regime in an Internal Orifice

Abstract

The identification of the water cavitation regime is an important issue in a wide range of machines, as hydraulic machines and internal combustion engine. In the present work several experiments on a water cavitating flow were conducted in order to investigate the influence of pressures and temperature on flow regime transition. In some cases, as the injection of hot fluid or the cryogenic cavitation, the thermal effects could be important. The cavitating flow pattern was analyzed by the images acquired by the high-speed camera and by the pressure signals. Four water cavitation regimes were individuated by the visualizations: no-cavitation, developing, super and jet cavitation. As by image analysis, also by the frequency analysis of the pressure signals, different flow behaviours were identified at the different operating conditions. A useful approach to predict and on-line monitoring the cavitating flow and to investigate the influence of the different parameters on the phenomenon is the application of Artificial Neural Network (ANN). In the present study a three-layer Elman neural network was designed, using as inputs the power spectral density distributions of dynamic differential pressure fluctuations, recorded downstream and upstream the restricted area of the orifice. Results show that the designed neural networks predict the cavitation patterns successfully comparing with the cavitation pattern by visual observation. The Artificial Neural Network underlines also the impact that each input has in the training process, so it is possible to identify the frequency ranges that more influence the different cavitation regimes and the impact of the temperature. A theoretical analysis has been also performed to justify the results of the experimental observations. In this approach the nonlinear dynamics of the bubbles growth have been used on an homogenous vapor-liquid mixture model, so to couple the effects of the internal dynamic bubble with the other flow parameters.