On the Use of Ultrafiltration or Microfiltration Polymeric Spiral-Wound Membranes for Cheesemilk Standardization: Impact on Process Efficiency

Ultrafiltration (UF) and microfiltration (MF) are widely-used technologies to standardize the protein content of cheesemilk. Our previous work demonstrated that protein retention of a 0.1-µm MF spiral-wound membrane (SWM) was lower, but close to that of a 10 kDa UF one. Considering that the permeability of MF membranes is expected to be higher than that of UF ones, it was hypothesized that the former could improve the efficiency of the cheesemaking process. Consequently, the objectives of this work were to compare 0.1-µm MF and 10 kDa UF spiral-wound membranes in terms of (1) hydraulic and separation performance, (2) energy consumption and fouling behavior, (3) cheesemaking efficiency of retentates enriched with cream, and (4) economic performance in virtual cheesemaking plants. This study confirmed the benefits of using MF spiral-wound membranes to reduce the specific energy consumption of the filtration process (lower hydraulic resistance and higher membrane permeability) and to enhance the technological performance of the cheesemaking process (higher vat yield, and protein and fat recoveries). However, considering the higher serum protein retention of the UF membrane and the low price of electricity in Canada, the UF scenario remained more profitable. It only becomes more efficient to substitute the 10 kDa UF SWM by the 0.1-μm MF when energy costs are substantially higher.


Introduction
Dairy products can be manufactured through many processing itineraries, but the most eco-efficient (i.e., having the lowest environmental impact and provides higher incomes) has to be selected to improve the competitiveness of the industry. In the cheese sector, the use of membrane filtration processes to concentrate the milk prior cheesemaking represents one way to improve its profitability and plant capacity [1,2]. Ultrafiltration (UF) is a widely used membrane technology. It allows to concentrate all the milk proteins, notably caseins (CN) and serum proteins (SP), whereas lactose and minerals are collected in the permeate [3,4]. It leads to improved coagulation properties [5][6][7] and increases cheese yield through a higher retention of milk components in the curd [3,6,8]. Microfiltration (MF) is also used for the protein standardization of cheesemilk [3]. As UF membranes, MF ones retain CN in the retentate, but generally have a low SP retention coefficient because of its higher mean pore size [9]. Due to the presence of SP in the MF permeate, it is considered as a "clean whey" since it is free of residues from cheesemaking such as caseinomacropeptite (CMP), cheese fines, colorants, lactic acid, or starter cultures found in traditional whey [10]. Furthermore, SP are found in the permeate in their native form, allowing the production of highly valuable whey protein isolates (WPI) that can be sold for their technological (foaming and gelling) and nutritional properties [11,12].
In a previous study, the performance of both 0.1-and 0.2-µm microfiltration (MF) spiral-wound membranes (SWM) was compared in the context of cheesemilk standardization [13]. Mercier-Bouchard et al. [13] confirmed the low SP retention property of a 0.2-µm MF SWM, but observed that SP are highly retained with a 0.1-µm polymeric MF SWM. A similar conclusion was reported in the literature, revealing the higher SP retention with MF SWM than with ceramic uniform transmembrane pressure (UTP), ceramic graded permeability (GP) membranes or hollow fiber ones [14,15]. In fact, SWM MF membranes, notably the 0.1-µm pore size, suffer from much more severe deposit formation so that the retained casein micelle layer exerts the retention effect.
Considering the largest pore size of MF membranes (higher permeability and hydraulic performance) and the fact that a more valuable permeate could be obtained (even if diluted), it was hypothesized that the cheesemilk standardization could be more efficient with the use of a 0.1-µm polymeric MF SWM, exhibiting a UF-like behavior, instead of a traditional 10 kDa UF SWM. The filtration performance of both membrane types was evaluated in the total recirculation and concentration modes at the pilot scale with skim milk at 50 • C. The cheesemaking efficiency of retentates collected during both filtration methods was also evaluated and compared with that of unconcentrated milk, and an economic assessment of the three scenarios tested was finally presented.

Raw Material
Pasteurized skim milk was purchased from a local dairy farm and stored at 4 • C until MF and UF experiments. These were performed in triplicate, in the recirculation mode, with the same batch of milk divided into three equal volumes of 300 L. For single-stage concentration and diafiltration (DF) modes, a different batch of milk was used for each membrane tested (0.1-µm MF and 10 kDa UF).

Filtration System
A pilot system described previously by Mercier-Bouchard et al. [13] was used for all the filtration experiments (model 393, Tetra Pak Filtration Systems, Champlin, MN, USA). Only one stage was used during this work with two MF or two UF membranes installed in series in the same loop. The 0.1-µm MF SWM was made of polyvinylidene fluoride (PVDF) (membrane v0.1, element specification model 3838, Synder Filtration, Vacaville, CA, USA), and the 10 kDa UF membrane was made of polyethersulfone (PES) (model DS-UH-3838, Microdyn-Nadir, Raleigh, NC, USA). They were mounted horizontally with respective surface areas of 13.38 m 2 and 10.14 m 2 .

Total Recirculation Mode
Filtration was performed at 50 • C in the total recirculation mode in order to determine the optimal transmembrane pressure (TMP) to be used in the concentration one. Experiments with 0.1-µm MF membranes were carried out at a TMP of 89.6, 106.9 and 124.1 kPa, as described in Mercier-Bouchard et al. [13], whereas values of 310.4, 379.4 and 447.5 kPa were applied on 10 kDa UF ones. Since neither type was operated with the same TMP, the permeability was chosen to compare them, as described by Methot-Hains et al. [16].

Concentration/DF of Skim Milk
Single-stage batch concentration and discontinuous DF with 1.5 diavolume (DV) were both performed at 50 • C at the optimal TMP, as determined in the recirculation mode, until reaching a targeted mass concentration factor (MCF) of 2.5×. Two DF were carried out (DF #1 and DF #2). Again, the permeability was used to compare the two membrane types. The true protein (TP) rejection coefficient was also calculated [17].

Membrane Fouling Characterization
The resistance-in-series model was applied to evaluate membrane fouling. The membrane resistance (R m ), reversible resistance (R rev ), irreversible resistance (R irrev ), and total resistance (R tot ) were calculated [4].

Chemical Analysis
Skim milk, retentate, permeate, and whey samples were analyzed according to the methodology described by Tremblay-Marchand et al. [17]. Briefly, the contents of the TP, CN, and non-protein nitrogen (NPN) were determined by the Kjeldahl digestion (AOAC International 991.20, 998.05, and 991.21, respectively). The total solids (TS) and fat (TF) were determined by the forced-air oven drying method (AOAC International 990.20) and the Mojonnier extraction one (AOAC International 989.05), respectively.

Energy Consumption
The electric energy consumption (Wh) of the MF and UF processes were obtained following the calculation of the power requirement (P, W) to operate the filtration system (Equation (1)).
where U, I and cos(ϕ) represent the voltage, current and power factor (0.65), respectively. The current used to calculate P was that measured at the end of each concentration or diafiltration steps for the feed and recirculation pumps. The specific energy consumption (SEC, Wh per kilogram of permeate that is removed) was determined as follows (Equation (2)): where ∆t and V P represent the time needed to complete concentration and diafiltration steps, and the volume of permeate that is collected during them, respectively. As in Mercier-Bouchard et al. [13], only electricity used to pump the fluids in the filtration system was considered.

Cheesemaking
The contents of TP of retentates during MF and UF were standardized to a final concentration of 7% (w/w) by reincorporating UF permeate. Fat standardization of retentates collected during both filtration types and of unconcentrated milk, was carried out by the addition of unpasteurized cream, obtained from a local dairy farm, in order to reach a final true protein to fat (TP/TF) ratio of 0.65. This low TP/TF corresponds to a high-fat cheesemilk [18]. Following standardization, the retentates were pasteurized at 68 • C for 30 min in a double-jacketed vessel mixer (model UMC-5, Stephan Machinery™, Hameln, Germany). The pH of the cheesemilks was adjusted with glucono-δ-lactone (GDL) to 6.50, and the model curds were produced according to the method described by Lauzin et al. [19] with the following modifications. At the end of the cheesemaking process, the curd was drained in two steps: in cotton cheesecloth over 30 min, and finally by centrifugation at 10,816× g for 30 min. The curds were vacuum-packaged and stored at 4 • C until chemical analysis. All experiments were performed in triplicate.

Cheesemaking Efficiency
The cheesemaking efficiency was evaluated with manufacturing yield (Y), moisture-adjusted yield (Y MA ), and protein and fat recovery (Y P and Y F , respectively). These variables were calculated based on the mass of the standardized cheesemilk (Vat) or one of the inputs (Inp). The Y (Equation (3)) and Y MA (Equation (4)) were determined as follows, according to Guinee et al. [1]: where m curd and m Vat represent the mass of the curd and that of the standardized cheesemilk, respectively.
where M curd and M ref represent the moisture (%) of the curd and that of a reference cheese, respectively.
The latter corresponded to the cheese made from unconcentrated milk. The protein recovery in the standardized cheesemilk (Y PVat ) (Equation (5)) was determined as follows: where m Pw and m PVat represent the protein mass in the whey and that in the standardized cheesemilk. The fat recovery in the standardized cheesemilk (Y F ) (Equation (6)) was determined as follows: where m FW and m FVat represent the fat mass in the whey and that in the standardized cheesemilk.

Economic Assessment
A process simulation was finally performed to compare the economic assessment of cheesemaking involving the three technological approaches presented previously: from standardized cheesemilk with MF or UF retentates, as well as from unconcentrated milk. It was done using the same parameters (i.e., membrane types, MCF, TP/TF ratio of the cheesemilk, number of DV or DF) as the experimental part, with the following modifications. Regarding the MF and UF scenarios, instead of concentrating the total volume of milk to a MCF of 2.5×, and to dilute it with the permeate, only a fraction was skimmed and concentrated. The retentate obtained was combined with cream and whole milk to standardize cheesemilk to a TP of 5.87% (w/w) and a TF of 9.03% (w/w). Missing data were interpolated from the ones obtained in the experimental part (i.e., membrane permeation fluxes at specific MCF). It was assumed that filtrations in virtual plants were performed in three-stage filtration systems.

Effect of TMP on Normalized Permeation Flux in the Total Recirculation Mode
Three TMP were tested in the total recirculation mode for MF and UF of skim milk at 50 • C. As shown in Figure 1, no limiting flux was reached between 89.6 kPa and 124.1 kPa for the MF type, and between 310.4 kPa and 447.5 kPa for the UF one. Consequently, the highest TMP obtained for both membranes was selected for further concentration and diafiltration experiments. At these TMP, the permeation flux normalized per unit of pressure was of 0.34 and 0.13 kg h −1 m −2 kPa −1 for the MF and the UF membranes, respectively ( Figure 1).

Effect of Concentration and Diafiltration Modes on Normalized Permeation Flux
During concentration and DF of skim milk, the difference between the permeability of the membrane types varied significantly at all the MCF tested (p < 0.05) (Figure 2). The DF performed with both membranes increased the initial permeability of each reconcentration step. For example, the value for the MF type was of 0.39 kg h −1 m −2 kPa −1 and 0.44 kg h −1 m −2 kPa −1 at the beginning of the first and second DF, respectively, while the initial one was of 0.30 kg h −1 m −2 kPa −1 (Figure 2). The permeability decreased with the MCF during the concentration and both DF steps with the MF and UF membranes, but the flux reduction rate was similar in all the processing steps ( Figure 2). It was, however, greater during MF with values between 34.4% (DF #2) and 38.2% (concentration), whereas those obtained during UF were between 13.4% (DF #2) and 31.5% (concentration) (Figure 2).

Effect of Concentration and Diafiltration Modes on Normalized Permeation Flux
During concentration and DF of skim milk, the difference between the permeability of the membrane types varied significantly at all the MCF tested (p < 0.05) (Figure 2). The DF performed with both membranes increased the initial permeability of each reconcentration step. For example, the value for the MF type was of 0.39 kg h −1 m −2 kPa −1 and 0.44 kg h −1 m −2 kPa −1 at the beginning of the first and second DF, respectively, while the initial one was of 0.30 kg h −1 m −2 kPa −1 (Figure 2). The permeability decreased with the MCF during the concentration and both DF steps with the MF and UF membranes, but the flux reduction rate was similar in all the processing steps ( Figure 2). It was, however, greater during MF with values between 34.4% (DF #2) and 38.2% (concentration), whereas those obtained during UF were between 13.4% (DF #2) and 31.5% (concentration) (Figure 2). with both membranes increased the initial permeability of each reconcentration step. For example, the value for the MF type was of 0.39 kg h −1 m −2 kPa −1 and 0.44 kg h −1 m −2 kPa −1 at the beginning of the first and second DF, respectively, while the initial one was of 0.30 kg h −1 m −2 kPa −1 (Figure 2). The permeability decreased with the MCF during the concentration and both DF steps with the MF and UF membranes, but the flux reduction rate was similar in all the processing steps ( Figure 2). It was, however, greater during MF with values between 34.4% (DF #2) and 38.2% (concentration), whereas those obtained during UF were between 13.4% (DF #2) and 31.5% (concentration) (Figure 2).

Effect of Concentration and Diafiltration Modes on Retentate Composition
Globally, the UF membrane had a higher TP rejection coefficient (p < 0.05) ( Table 1). Consequently, the TP content of permeates collected during MF was always significantly higher than the those during UF (p < 0.05), but no significant difference was found regarding the TP of 2.5× UF and MF retentates after the second DF (p > 0.05) ( Table 1). The transmission of TP in the permeate during MF corresponded to around 11% of the TP of the skim milk. The DF effectively reduced the TS content of 2.5× retentates (MF and UF). For example, the TS content of the retentate during MF was reduced from 14.97 ± 0.65% (w/w) after the first concentration step to 11.48 ± 0.42% and 9.88 ± 0.02% (w/w) following the first and the second DF, respectively (p < 0.05) ( Table 1). After the second DF, the 2.5× retentates (UF and MF) had the same TS content (10.09 ± 1.77% and 9.88 ± 0.02% (w/w), respectively) (p > 0.05) ( Table 1). Even if the retentates were diafiltered, their contents of NPN did not differ significantly at the end of each concentration and DF steps (p > 0.05) ( Table 1). The NPN concentrations were possibly too low to observe differences.

Effect of Concentration and Diafiltration Modes on Energy Consumption
UF was more energy-demanding than MF, during the concentration or both diafiltration steps (p < 0.05) ( Table 2). Considering the whole process, UF had a SEC of 13.31 ± 0.23 Wh per kilogram of permeate that is removed, whereas that of MF was of 8.59 ± 0.50 Wh per kilogram of permeate removed ( Table 2). During MF, the concentration step was more energy-demanding than both diafiltration steps (p < 0.05), but the electricity used for these was similar (p > 0.05) ( Table 2). During UF, the electricity consumption of each step was significantly different (p > 0.05) ( Table 2).

Effect of Concentration and Diafiltration Modes on Membrane Fouling
The MF and UF membranes had a similar R rev (p > 0.05) (Table 3). However, the latter had a higher R m , R irrev and R tot than the former (p < 0.05) ( Table 3). The R tot of the UF membrane type was 2.25 times higher than that of the MF one ( Table 3). The R irrev of the UF membranes represented 42% of its R tot , whereas that of the MF membranes represented 26% of its R tot . Furthermore, the fouling type of the latter was rather reversible than irreversible. Its R rev was 2.31 times higher than its R irrev , while it was only 1.13 times higher with the UF membrane (Table 3). Table 3.
Hydraulic resistance of MF and UF membranes following concentration and diafiltration processes.

Cheesemaking Efficiency of Unconcentrated Cheesemilk, and Standardized with MF and UF Retentates
The higher moisture-adjusted yield (Y MAVat ) was obtained with the MF retentate (29.67 ± 0.51%, p < 0.05), but the use of UF also permitted to increase Y MAVat compared to unconcentrated cheesemilk (p < 0.05) ( Table 4). The MF retentate also allowed higher cheesemilk protein (Y PVat ) and fat (Y FVat ) recoveries in cheese (90.73 ± 0.13% and 96.91 ± 0.26%, respectively), which also significantly increased the FDM ratio (p < 0.05) ( Table 4). The use of the same TP/TF ratio in the cheesemilk in the three scenarios allowed a similar draining behavior, revealed by a similar MNFS ratio (67.69 to 67.85) between the three cheeses (p > 0.05) ( Table 4).

Economic Assessment of MF and UF Approaches through a Process Simulation
The results obtained in the experimental part were used to perform a process simulation in virtual plants receiving 1,000,000 kg of whole raw milk daily (Tables 5 and 6). Similar patterns between the Y MA observed in the experimental part and the predicted Y Vat (calculated with cheeses having the same moisture content) presented in Table 5 were observed. It is, however, important to mention that the Y Vat predicted in the process simulation was lower (i.e., by 1.16% for the UF approach); the difference between Y Vat of MF and UF scenarios was different as well.  dairy ingredients) to cheeses. However, the cheeses that were made during the present studies had a higher moisture content (lower volume of whey expelled) and a lower TP/TF ratio that was high, which possibly affected both protein and fat recoveries. Overall, both scenarios involving cheesemilk standardization with dairy retentates had a lower operating cost per kilogram of cheese and would increase the margin of the cheesemaking process in comparison with the unconcentrated scenario (Table 6), as also predicted by Papadatos et al. [30]. However, three elements may have biased results: (1) the cheesemilks had a high fat content, which generated important cream expenditures; (2) they also had a different CN content, due to the use of a TP/TF ratio; and (3) the cheesemaking performance of the MF scenario was evaluated directly on the retentate instead of the milk standardized with the latter, as generally performed in the industry [28]. The first two elements appeared more important, as the UF scenario remained the more efficient even with a possible overestimation of the MF cheesemilk performance. Indeed, the 2.5× UF retentate had a higher TP content than the MF one due to the transmission of SP in the MF permeate. Consequently, a higher volume of cream input (6485 kg) was needed in the UF scenario. It increased the operating cost of the UF process by 26,996 Can$ ( Table 6). In fact, no bias would occur with the use of a CN/TF ratio, since MF and UF have a similar CN content. This statement was confirmed by an uncertainty analysis performed with a CN/TF instead of a TP/TF ratio of 0.65 (Table 7). With the CN-based standardization target, only slight differences were observed between MF and UF scenarios in terms of inputs or filtration expenditures. The UF scenario was however the more efficient with an operating cost (per kilogram of cheese) 0.5% lower than the MF one, and with a higher Y Inp (Table 7), again due to its higher SP retention. In countries where the energy price is higher, the conclusion could be different. In fact, with the Canadian milk price, the MF scenario would become the more profitable if the one for electricity was 6.4 times higher.

Conclusions
Contrarily to the previous hypothesis proposed by Mercier-Bouchard et al. [13], even if the MF scenario was the less energy-demanding, notably because of the higher permeability of MF membranes, and had better cheesemaking performance of the MF retentate (calculated from the cheesemilk data collected), the UF one remained more efficient. In fact, with the low cost of electricity in the Canadian economic context, the gain associated with the lower use of energy with the 0.1-µm MF SWM was not sufficient to compensate the loss of SP in the MF permeate, as low as it was. By focusing only on the milk standardization and cheesemaking steps, the 0.1-µm MF SWM could not substitute the 10 kDa UF one without a substantial increase in energy costs (i.e., with energy costs at least 6.4 times higher). A further study considering the processing costs of cheese whey, MF and UF permeates as well as diafiltrates will be necessary to draw definitive conclusions on the economic assessments.
It should help to determine if processing diluted MF permeates generated from 0.1-µm MF SWM could contribute to increase the profitability of the MF process over the UF one.