Author(s): Marko Hoč Evar PhD; Brane Širok PhD; Bogdan Blagojevič PhD; Igor Grabec PhD
Linked Author(s):
Keywords: Cavitation vortex structure; Francis turbine; experimental modeling; radial basis neural networks
Abstract: Experimental modeling of a cavitation vortex structure in a Francis turbine draft tube is presented. Pressure in the draft tube and images of vortex structure were acquired simultaneously for the experiment. Non-parametric radial basis neural networks were used for the experimental modeling. Two variables were modeled: average image intensities in the selected region, and entire images of the cavitation vortex. Both variables were modeled on the basis of experimentally provided pressure data. The learning set consisted of pressure–image pairs. The modeling consisted in providing only pressure, while average image intensities and images of the cavitation vortex were modeled. Regression coefficients r w between the modeled and measured average image intensities in an interval of 0.82–0.99 confirmed the correct choice of the experimental modeling method. The entire image modeling of the cavitation vortex also gave encouraging results with regression coefficients in an interval of 0.59–0.88. The regression coefficient of the entire image modeling was lower than the regression coefficient r w of the average image intensity modeling due to the high dimensionality of the cavitation vortex structure
DOI: https://doi.org/10.1080/00221686.2007.9521789
Year: 2007