Assessment of sugars separation from a model carbohydrates solution by nanofiltration using a design of experiments (DoE) methodology

• DoE methodology was successfully used to study inulin separation from small molecules.
• Performance with different operational variables was investigated.
• Experimental results show that separation is complex but feasible within the domain.
• Important relationships between parameters and performance was established.
• Polarization model lies with experimental results obtained.

Nanofiltration (NF) of a model solution of different molecular weight carbohydrates (inulin, sucrose, glucose and fructose) has been studied with a pilot cross-flow unit, using a membrane with molecular weight cut off (MWCO) of 600 Da. The aim of this work was to screen operational parameters and to determine the most feasible domain in order to separate inulin-type fructans from model solutions containing low molecular weight (LMW) sugars. Firstly, parameters such as transmembrane pressure (TMP), feed concentration (C0) and retentate flow (QR) were screened using a Koshal design for single-component solutions. The results indicated that the most influential parameters were TMP and C0. Then, a Doehlert design was used to improve single-inulin nanofiltration. Indeed, surface response analysis (p < 0.05) revealed inulin rejection values over 90%, for high TMP (1400 kPa) and low C0 (5 g/100 mL). In order to test the robustness of these results in presence of sugar, a full factorial design was used to evaluate the separation of a binary solution containing sucrose and inulin. In this case, the permeate flux (Jv) and inulin rejection (Rinulin) results were in agreement with those obtained with single-component solutions. Furthermore, additional experiments on a complex quaternary mixture (sucrose–glucose–fructose–inulin), confirmed that the presence of mono and di-saccharides do not modify significantly Rinulin. Actually, it is possible to obtain even slightly better results for the separation process (Rinulin = 93%). All the results were statistically validated using analysis of variance (ANOVA).