Haitham Yousef

DYNAMIC ULTRAFILTRATION OF PROTEINS-A NEURAL NETWORK APPROACH

Abstract A neural network approach for the prediction of the rate of ultrafiltration of proteins has been developed. The approach has been used to predict the rate of ultrafiltration of bovine serum albumin as a function of pH and ionic strength. This is a very non-linear problem that has previously been best described through sophisticated descriptions of protein–protein interactions within the layer close to the membrane surface. Networks with a single hidden layer have been used to predict the dynamic rate of filtration from very few data points. Emphasis has been placed on using a small number of training data points and small networks. Variation of the number of training points and use of different training point selection schemes have shown that it is the quality of training points rather than the quantity that leads to the best predictions. The network training process may be optimised by using physical insights to select appropriate input variables. Testing of the neural network approach showed that it could give excellent agreement with experimental results, with average errors less than 2.7%.

Prediction of the rate of crossflow membrane ultrafiltration of colloids: A neural network approach

Prediction of the dynamic crossflow ultrafiltration rate of colloids poses a complex non-linear problem as the filtration rate has a strong dependence on both the solution physicochemical conditions and the operating conditions. As a result, the development of general physics-based models has proved extremely challenging. In this paper an alternative artificial neural network approach is developed. The approach has been used to predict the time-dependent rate of ultrafiltration of silica suspensions under different conditions of pH, ionic strength and applied pressure. Neural networks with a single hidden layer were used to predict the filtrate flux-filtration time profiles from a small number of training points. Training points were chosen from both three and five sets of solution conditions to study how network predictability would be affected. Physical understanding of the process helped in choosing the right input variables, which in turn optimised the training. The neural network approach was found to be capable of modelling this complex process accurately.

Predicting salt rejections at nanofiltration membranes using artificial neural networks

Abstract An artificial neural network (ANN) has been used to predict the rejections of single salts (NaCl, Na2SO2, MgCl2 and MgSO4) and mixtures of these salts at a nanofiltration membrane. Such rejections show complex non-linear dependencies on salt concentration, mixture composition, pH and applied pressure and provide a demanding test of the application of ANN analysis to membrane processes. A single optimized network was used for all predictions, the network having the ability to switch on/off its internal parts depending on the process solution. A qualitative physical understanding of the process was used in choosing the appropriate input variables. The predictions have been compared to pilot plant rejection data obtained with a spiral-wound membrane. The overall agreement between ANN predictions and experimental data was very good for both single salts and mixtures. In practical circumstances, the ANN approach to nanofiltration has the advantage of only requiring simple and readily available inputs and a minimum understanding of the complex phenomena controlling rejection.

Effect of salts on water viscosity in narrow membrane pores

The effect of various salts on the viscosity, and by implication structure, of water in polymeric membrane pores of radius approximately 1.69 nm and low charge density has been studied. Permeation of pure water and various electrolyte solutions was analyzed using the Hagen-Poiseuille equation expressed in a ratio form to exclude membrane-specific quantities such as pore radius and length. The analysis produced viscosity ratios of electrolyte to pure water inside the membrane pores. Comparing the viscosity ratios inside the pores with their bulk counterparts showed that confinement significantly increased the sensitivity of water structure to the presence of ions. It has been found that, in relative terms in the pores, Cl- was a strong structure breaker, K+ was a moderately strong structure breaker, Na+ was a weak structure breaker, SO4(2-) was a weak structure maker, and Mg2+ was a strong structure maker. Predictive modeling of membrane separation performance would benefit from such effects being taken into account in cases where the pore ion concentrations may be high.

Dynamic crossflow ultrafiltration of colloids: a deposition probability cake filtration approach

Abstract A description of the dynamic, time-dependent ultrafiltration of charged colloids has been developed by combining an adhesion model for crossflow filtration of non-interacting particles with a Wigner–Seitz cell model for frontal filtration of charged colloidal dispersions. The adhesion model helps provide a criterion for the deposition of particles in a filter cake on the membrane surface. The Wigner Seitz cell model provides a quantitative description of the hydraulic resistance of the resulting filter cake. Two means of evaluating the deposition probability of charged colloidal particles have been considered — a cell based approach and an adjustable parameter approach, which was subsequently combined with multiple linear regression. The latter combined approach gave excellent agreement with experimental filtration flux–time data for silica colloids as a function of applied pressure, particle concentration, ionic strength and pH. Overall, this is a relatively simple yet useful means of correlating the time-dependent performance of such a membrane process over a wide range of operating conditions.

Interpretation of water isotherm hysteresis for an activated charcoal using stochastic pore networks

Water vapor adsorption equilibria on activated carbons typically exhibit hysteresis. The size and shape of the hysteresis loop which separates the adsorption and desorption branches is a strong function of the pore size and interconnectivity of the pores. Neither conventional pore filling models nor statistical thermodynamics approaches provide a means for predicting the extent of hysteresis from only adsorption measurements. This work uses the Kelvin Equation in conjunction with the structural concept of a stochastic pore network to describe measured water isotherms on BPL carbon. Using a pore segment distribution function determined from the adsorption branch, it is shown that totally random assemblies underestimate the extent of hysteresis. It is possible, however, to closely fit the measured BPL-water hysteresis loop using a patchy heterogeneity in which a proportion of the larger pores are preferentially located on the exterior, mid-range pores are concentrated in a sub-surface layer and some large pores form shielded voids behind much smaller pores.

Experimental measurement and numerical modelling of dye washout for investigation of blood residence time in ventricular assist devices

Ventricular assist devices have become the standard therapy for end-stage heart failure. However, their use is still associated with severe adverse events related to the damage done to the blood by fluid dynamic stresses. This damage relates to both the stress magnitude and the length of time the blood is exposed to that stress. We created a dye washout technique which combines experimental and numerical approaches to measure the washout times of ventricular assist devices. The technique was used to investigate washout characteristics of three commercially available and clinically used ventricular assist devices: the CentriMag, HVAD and HeartMate II. The time taken to reach 5% dye concentration at the outlet (T05) was used as an indicator of the total residence time. At a typical level of cardiac support, 5 L/min and 100 mmHg, T05 was 0.93, 0.28 and 0.16 s for CentriMag, HVAD and HeartMate II, respectively, and increased to 5.06, 1.64 and 0.96 s for reduced cardiac support of 1 L/min. Regional variations in washout characteristics are described in this article. While the volume of the flow domain plays a large role in the differences in T05 between the ventricular assist devices, after standardising for ventricular assist device volume, the secondary flow path was found to increase T05 by 35%. The results explain quantitatively, for the first time, why the CentriMag, which exerts low shear stress magnitude, has still been found to cause acquired von Willebrand Syndrome in patients.