Prediction of Liquid Holdup for a Multi-Plane System Using Artificial Neural Network
2021 Petroleum Engineering Masters Theses
Liquid holdup was predicted for slug flow using the neural network toolbox available in MATLAB. Experimental data contained Liquid holdup measured at an interval of 5 seconds over a period of 12,000 time points for both planes of a horizontal pipe setting. 13 different gas superficial velocities were varied (within a range of 0.05-4.73 m/s) while keeping the liquid superficial velocity constant for three different cases. The neural network’s efficiency was analyzed by comparing experimental and simulated values of average liquid holdup. The neural network proved to be highly efficient as close similarities were observed for all values of average liquid holdup for both cases. The neural network’s efficiency was further ascertained by the analysis of the ‘’Error Autocorrelation chart ‘’ which showed the error at all other lags (asides from the zero lag) lying within the confidence limit for all cases considered. Computational macros were used to determine the structural velocity and frequency for both experimental and simulated run(s). A cross plot of simulated and experimental values for these parameters gave a very good fitting; further emphasizing the efficiency of the neural network as a tool for this research. The effect of gas and liquid superficial velocities on liquid holdup was also investigated. It was observed that the liquid holdup decreased with increasing gas superficial velocity and increased with increasing liquid superficial velocity. This could be attributed to the increase in the gas phase residence time with an increase in gas superficial velocity, thereby reducing the fraction of fluid available for the liquid phase and thus, the liquid holdup.