Untitled

In this work, the membrane technique of nanofiltration (NF) was used to carry out partial demineralization of whey and milk ultrafiltration permeate (MUP). The experiments were performed in a completely automated NF pilot plant. An aromatic polyamide spiral wound membrane supplied by Osmonics (USA) was selected (DK2540C model). Batch and discontinuous diafiltration (DF) configurations were compared. Permeate fluxes were higher for MUP due to the lower amount of proteins, which can cause severe membrane fouling. The degree of demineralization for the divalent cations was negligible for both feed streams. Protein and lactose permeation were also observed to be very low. Ion permeation was higher for MUP probably due to the lower amount of proteins. When discontinuous DF was performed, ion removal increased in the case of whey, but it did not significantly improve in the case of MUP. The degree of ion removal was higher than 30%. Finally, the Donnan Steric Partitioning pore Model was used to predict lactose and ion rejection. A good agreement between predicted and experimental data was observed.

NF was observed to be an effective method to perform partial demineralization of whey and MUP. Permeate flux was lower when whey was nanofiltered probably due to the presence of proteins, which can form a gel layer on the membrane surface. For the same reason, the degree of mineral salt removal was observed to be higher in the case of MUP. Losses of lactose and protein in the permeate stream were very low. Divalent ion permeation was very low as well. The degree of mineral salt removal increased with VCR. By means of batch NF, the degree of overall ion removal increased up to 27% at VCR 4.0 in the case of whey; when MUP was nanofiltered at VCR 4.7, the degree of ashes removal was 36%. By means of DF, the degree of ashes reduction increased up to 37% in the case of whey and to 38% in the case of MUP. Therefore, as water consumption and lactose losses increases, the DF step is not justified in the case of MUP. The DSPM was able to predict lactose and salt rejection when model solutions were used as feed streams. The divergence between predicted and experimental rejection was lower than 10% for all the operating conditions tested.