Artigo de peri??dico
Two-phase flow void fraction estimation based on bubble image segmentation using Randomized Hough Transform with Neural Network (RHTN)
Registro en:
0149-1970
118
10.1016/j.pnucene.2019.103133
0000-0002-5355-0925
0000-0002-6689-3011
0000-0003-2445-1298
80.88
62.25
Autor
SERRA, PEDRO L.S.
MASOTTI, PAULO H.F.
ROCHA, MARCELO S.
ANDRADE, DELVONEI A. de
TORRES, WALMIR M.
MESQUITA, ROBERTO N. de
Resumen
The International Atomic Energy Agency (IAEA) has been encouraging the use of passive cooling systems in new
designs of nuclear power plants. Next nuclear reactor generations are intended to have simpler and robust safety
resources. Natural Circulation based systems hold an undoubtedly prominent position among these. The study of
limiting conditions of these systems has led to instability behavior analysis where many different two-phase flow
patterns are present. Void fraction is a key parameter in thermal transfer analysis of these flow instability
conditions. This work presents a new method to estimate void fraction from images captured of an experimental
two-phase flow circuit. The method integrates a set of Artificial Neural Networks with a modified Randomized
Hough Transform to make multiple scans over acquired images, using crescent-sized masks. This method was
called Randomized Hough Transform with Neural Network (RHTN). Each different mask size is chosen according
with bubble sizes, which are the main ???objects of interest??? in this image analysis. Images are segmented using
fuzzy inference with different parameters adjusted based on acquisition focus. Void fraction calculation considers
the volume of the imaged geometrical section of flow inside cylindrical glass tubes considering the acquisition
depth-of-field used. The bubble volume is estimated based on geometrical parameters inferred for each
detected bubble. The image database is obtained from experiments performed on a vertical two-phase flow
circuit made of cylindrical glass where flow-patterns visualization is possible. The results have shown that the
estimation method had good agreement with increasing void fraction experimental values. RHTN has been very
efficient as bubble detector with very low ???false-positive??? cases (< 0.004%) due robustness obtained through
integration between Artificial Neural Networks with Randomized Hough Transforms. Financiadora de Estudos e Projetos (FINEP) FINEP: REDETEC-CNEN 01.10.0248.0/2010