Prediction of Liquid-Liquid Flow Patterns in a Y-Junction Circular Microchannel Using Advanced Neural Network Techniques

The flow pattern map for a liquid liquid system in a 600 mu m circular microchannel was experimentally investigated for a varying Y-junction confluence angle (10 to 180). The experimental results showing the distinguishing nature of transition boundaries were established using graphical interpretation. This paper tries to find a better objective flow pattern indicator for vast amounts of experimental data. Studies have been carried out using significant feed-forward back-propagation networks and radial-basis networks such as artificial neural network pattern recognition (ANN-PR), artificial neural network function fitting (ANN-FF), cascade-forward network (CFN), probabilistic neural network (PNN), generalized regression neural network (GRNN), and adaptive neuro-fuzzy inference system (ANFIS). From the study, we found that GRNN showed better prediction ability than the other prediction techniques. Discrete- and continuous-time state-space models for the system were also developed using the system identification technique.