Prediction of the point efficiency of sieve tray using artificial neural network

Abstract EN

An application of neural network technique was introduced in modeling the point efficiency of sieve tray, based on a data bank of around 331 data points collected from the open literature.

Two models proposed, using back-propagation algorithm, the first model network consists : volumetric liquid flow rate (QL), F factor for gas (FS), liquid density (pL), gas density (pg), liquid viscosity (fiL), gas viscosity (fig), hole diameter (dH), weir height (hw), pressure (P) and surface tension between liquid phase and gas phase (a).

In the second network, there are six parameters as dimensionless group: Flow factor (F), Reynolds number for liquid (ReL), Reynolds number for gas through hole (Reg), ratio of weir height to hole diameter (hw / dH), ratio of pressure of process to atmosphere pressure (P / Pa), Weber number (We).

Statistical analysis showed that the proposed models have an average absolute relative error (AARE) of 9.3 % and standard deviation (SD) of 9.7 % for first model, AARE of 9.35% and SD of 10.5 % for second model and AARE of 9.8 % and SD of 7.5 % for the third model.