Abstract
In this study, two-phase flow tests have been carried out in a 3×3 rod bundle channel to investigate the feasibility on flow regime classification using artificial neural network (ANN). The rod bundle channel was constructed with nine 11.5 mm-diameter rods in a rectangular array with a pitch of 15.4 mm and enclosed by a 52×52 mm rectangular channel, and the overall height was about 5 m. The test flow conditions covered the superficial gas and liquid velocities of <jg>= 0.02-10 m/s and <jf>= 0.02-2 m/s. The transient void fraction of each test was measured and converted into probability density function (PDF) and cumulative probability distribution function (CPDF) and rearranged as the input matrices for Kohonen neural network for training and classification with the training epoch of 1500 times. While assigning 3 groups, the bubbly and cap-bubbly flows can be clearly distinguished; whereas when assigning 5 groups, the bubbly and cap-bubbly flow can be further subdivided with an additional transition region, and the cap-turbulent and churn flows can be roughly separated. This study has preliminarily demonstrated the feasibility on classification of two-phase flow regimes in the 3×3 rod bundle channel using artificial neural network technique.