Pressure drop estimation in tube flow of non-Newtonian fluid foods by neural networks

A tube flow viscometer complete with data acquisition system was designed and developed for continuous measurement of pressure drop and flow velocity. Experiments were carried out with tomato ketchup, oyster sauce, mayonnaise, and 1% and 1.5% CMC solutions in the laminar flow region using stainless steel tubes of four diameters (0.00751-0.01636 m). The flow parameters determined with the tube viscometer after slip correction and a rotational viscometer were correlated for estimating the pressure drop indirectly. Finally, it was shown that neural networks could accurately predict the pressure drop in tube flow without making any correction for wall-slip from the input data on tube diameter, fluid density, mass flow rate, and power-law parameters determined with a rotational viscometer. Among three neural network architectures tested, the generalized regression neural networks were most easy to train and they predicted the pressure drop gradient in tube flow with greatest accuracy (4.7% average absolute error).