Supersaturation control in membrane-assisted antisolvent crystallization of amino acids

Pharmaceutical and chemical industries provide us with high-value products through well-controlled and robust processes with limited tolerance to change. Process intensification emerged as a new paradigm in the wake of carbon neutrality urgency to shift different production processes to environmentally friendly ones while maintaining their economic viability. Crystallization is a purification step used to recover compounds from solution in their purest form, crystals. Stirred-tank crystallizers have been the favorable strategy to crystallization for decades as they are easy to use, but do not enable enough options to control crystallization, particularly antisolvent crystallization. The latter consists of mixing the solution where the compound of interest is dissolved (the crystallizing solution) with another fluid (the antisolvent) responsible for reducing the solubility of the target compound resulting in the appearance of crystals. Membrane technology has the capacity to embody different pathways to process control both via the optimization of operating conditions and tuning membrane properties to adjust the antisolvent contact with the crystallizing solution. This thesis sheds light on the antisolvent mass transfer through this technology and tackles correlations between the operating conditions, membrane properties and crystal characteristics such as the crystal size and shape. Membrane-assisted antisolvent crystallization (MAAC) was first used to crystallize the amino acid L-serine. Two commercial membranes made of polyvinylidene fluoride (PVDF) and polypropylene (PP) with water contact angles of 130° and 150° respectively, allowed a controlled antisolvent crystallization. These membranes controlled the transmembrane mass transfer of the antisolvent (ethanol) under different feed and antisolvent velocities at ambient conditions. In all cases, a narrow crystal size distribution (CSD) of L-serine was obtained reflected in a coefficient of variation (CV) of 31-37%, compared with batch antisolvent crystallization or drop-by-drop crystallization where the CV was 63 and 54% respectively. Thanks to the measurement of L-serine and ethanol concentration along the operation time, the mass transfer coefficient of MAAC was evaluated. Increasing the antisolvent or the crystallizing solution velocity showed that a too high value of one or the other could result in reverse permeation or system blockage (inside the membrane contactor, module, or tubing). This study explained the transmembrane mass transfer in MAAC and the resulting crystal properties. Still, optimization of the operating conditions was necessary. The impact of solution velocity, antisolvent composition, temperature and gravity, was investigated using glycine-water-ethanol as a model crystallization system, and polypropylene flat sheet membranes. Results proved that in any condition, membranes were consistent in providing a narrow CSD with a coefficient of variation in the range of 0.5-0.6 as opposed to 0.7 obtained by batch and drop-by-drop crystallization. The prism-like shape of glycine crystals was maintained as well, but slightly altered when operating at a temperature of 35 °C with the appearance of smoother crystal edges. Finally, the mean crystal size was within 23 to 40 µm and did not necessarily follow a clear correlation with the solution velocities or antisolvent composition but increased with the application of higher temperature or gravity resistance. Besides, the monoclinic form of α-glycine was perfectly maintained in all conditions. The results at each condition correlated directly with the antisolvent transmembrane flux that ranged between 0.2 and 1.0 ×10-3 kg/m2.s. Once the effect of operating parameters was clarified, the impact of the membrane properties on the dosage of the antisolvent, and the resulting crystal properties was next investigated. Little was known on the relationship between membrane characteristics, antisolvent mass transfer and crystal properties. Flat sheet PVDF membranes were prepared using non-solvent induced phase separation (NIPS), to illustrate the impact of membrane characteristics on the resulting CSD using MAAC. The crystallization of glycine was taken as a case study. The antisolvent transmembrane flux increased when the membrane thickness decreased, or when the hydrophobicity or the porosity increased. After MAAC operation, membranes had no significant change in surface functional groups, hydrophobicity nor structure. The best-performing membrane was found to have 119° hydrophobicity, 89.4% porosity and 140 µm thickness, reaching high reproducibility of CSD corresponding to stable transmembrane fluxes throughout 180 minutes of operation. To unveil the effect of membrane surface morphology, surface micro-patterned membranes were also developed using phase separation micro-molding (PSµM). The impact of these membranes on antisolvent crystallization were evaluated in terms of antisolvent mass transfer and crystal properties, taking glycine-water-ethanol as the crystallization model. The resulting crystals from MAAC using surface-patterned membranes were up to 271.56 % larger than crystal sizes from plain membranes. Finally, to understand fundamentally the interactions between solvent, antisolvent, solute and membrane, molecular dynamics (MD) simulations were conducted. Focused beam reflectance method was also used to track crystal growth empirically. Models for MD simulations were developed to understand the transitioning from an undersaturated system (S=0.74) to supersaturated systems (S=1.35 and 2.39). All models were run under an ambient temperature of 300 K for 500ns on Gromacs software. A difference in solvent to antisolvent ratio induced difference in the molecular arrangement of glycine molecules such that cyclic dimers appeared in S= 0.74 and 1.35, while it was not the case in S=2.39. Empirically, PVDF membranes of 0.2 µm pore size were used to crystallize glycine via MAAC. The mean crystal sizes were 86, 130 and 136 µm from MAAC, drop-by-drop and batch respectively, such that MAAC allowed for the narrowest crystal size distribution and yield. For a fairer comparison, a controlled stirred-tank crystallization has taken place in semi-batch mode, which demonstrated that varying the antisolvent dosage is not sufficient in this case to obtain a narrow CSD. Overall, this thesis investigated the key factors responsible for antisolvent crystallization control through porous polymeric membranes using in-line and off-line characterization tools to fathom the correlation between the antisolvent mass transfer, operating conditions, membrane properties and crystal characteristics. This work paves the way for the successful and sustainable implementation of MAAC in current and future crystallization challenges.