Applications of artificial neural network (ANN) method for performance prediction of the effect of a vertical 90° bend on an air–silicone oil flow

Abstract

Knowledge of how the presence of a bend can change the flow patterns of a gas–liquid mixture is important for the design of multiphase flow systems, particularly to prevent burn-out and erosion–corrosion. Burn-out and erosion–corrosion both have serious implications for heat and mass transfer. The objective of this work therefore is to train an artificial neural network (ANN), a powerful interpolation technique, to predict the effect of a vertical 90° bend on an air–silicone oil mixture over a wide range of flow rates. Experimental data for training, validation, testing and final prediction were obtained using advanced instrumentation, wire mesh sensor (WMS) and high speed camera. The performance of the models were evaluated using the mean square error (MSE), average absolute relative error (MAE), Chi square test (X2) and cross correlation coefficients (R). The performance discriminator X2 for prediction of average void fraction is 2.57e-5 and that for probability density function (PDF) of void fraction MAE is 0.0028 for best performing models. The well trained ANN is then used to predict the effects of the two input parameters individually. The predicted results show that for the before the bend scenario, the most effective input parameter that reflects a change in flow pattern is the gas superficial velocity. On the other hand, the most unfavorable output parameter to measure after the bend is the average void fraction based on the fact that the flow near the bend is a developing one.