### 1. Introduction

^{−1}tangential velocity and 1bar TMP, the pore constriction model could describe the flux decrease during raw rice wine tangential microfiltration. As the velocity increases to 1 m.s

^{−1}, both models (cake formation and pore constriction) could predict fouling behavior with alternative dominance between the two as pressure increases from 1 to 2 bar.

### 2. Materials and Methods

^{2}. The membrane operates following an in/out configuration as mentioned in Table 2.

^{−1}and 1.05 bar, respectively. These conditions have been chosen as they ensure a maximal conversion rate and minimal permeate turbidity (0.2 NTU) [29].

^{−1}.

### 2.1. Fouling Mechanism Identification

V : permeate volume (m

^{3}), t: filtration time (s).K : Hermia model constant related to the nature of fouling (K

_{sb}: standard clogging constant, K_{cb}: complete blocking constant, K_{ib}: intermediate blocking constant, K_{cf}: cake formation constant).m: characteristic value of each model: cake filtration m = 0, intermediate blocking m = 1, pore constriction (called also standard blocking) m = 3/2, and complete blocking m = 2.

##### (2)

$$\%\hspace{0.17em}average\hspace{0.17em}error=\frac{{J}_{\text{exp}}-{J}_{\text{mod}el}}{{J}_{\text{exp}}}\times 100$$*J*

*and*

_{exp}*J*

*are respectively the experimental and the predicted permeate flux.*

_{model}### 2.2. Experimental Design

J

_{0}: Initial permeate flux of the filtration (L.h^{−1}.m^{−2})J

_{i(n+1)}: Permeate flux at the beginning of the last filtration cycle (L.h^{−1}.m^{−2})

##### (4)

$$Reversible\hspace{0.17em}fouling=Total\hspace{0.17em}fouling-Irreversible\hspace{0.17em}fouling$$##### (6)

$$\begin{array}{ll}Y=\hspace{0.17em}\hfill & {b}_{0}+{b}_{1/A}{X}_{1A}+{b}_{2/A}{X}_{2A}+{b}_{2/B}{X}_{2B}+{b}_{2/C}{X}_{2C}+\hfill \\ \hspace{0.17em}\hfill & {b}_{3/A}{X}_{3A}+{b}_{3/B}{X}_{3B}+{b}_{3/C}{X}_{3C}+{b}_{4/A}{X}_{4A}+\hfill \\ \hspace{0.17em}\hfill & {b}_{4/B}{X}_{4B}+{b}_{4/C}{X}_{4C}+{b}_{4/D}{X}_{4D}+{b}_{5/A}{X}_{5A}+\hfill \\ \hspace{0.17em}\hfill & {b}_{5/B}{X}_{5B}+{b}_{5/C}{X}_{5C}+{b}_{5/D}{X}_{5D}\hfill \end{array}$$*Y*is the theoretical response,

*b*

*are the theoretical model coefficients corresponding to the relative influence of each variable on the response.*

_{i/j}*i*(1, 2, 3, 4 and 5) is the factor number and

*j*(A, B, C and D) is the difference between a factor level and its last level (

*A*is the difference between the first and the last level,

*B*is the difference between the second and the last level,

*C*is the difference between the third and the last level,

*D*is the difference between the fourth and the last level). For example (b

_{2/A}: i = 2 = second factor BWD and j = A = difference between level 1 and the last level 4).

*X*as mentioned in table 5 is the factor.

_{i/j}coefficients are calculated by the Nemrodw software. Each coefficient represents the effect of the corresponding factor (factor number i) when it varies from one level to another (represented by j). The choice of a factor as important or not depends on its effect on the response (statistically significant or insignificant). The significance of each coefficient is evaluated by student’s test (at 95% confidence level). The corresponding P-value must be less than 0.05 to ensure significance. The calculated effects can thus, be represented on bar chart (Fig. 5). We consider that if b

_{i/j}exceeds the significance level (represented by the dashed line in Fig. 5) the effect of this parameter is statistically significant.

### 3. Results and Discussion

### 3.1. Fouling Mechanisms Identification

^{2}values for the four Hermia models in the tangential ultrafiltration of polyethylene glycol and wastewater, respectively. In the case of a good fit (high R

^{2}), Torkamanzadeh et al. [15] concluded that the decrease in flow can be described by one of the four models because it seems that all the mechanisms coexist. In our case, this coexistence of mechanisms might be due to the feed complexity and the large distribution of the particles sizes (from 1 nm to 1 μm) as well as in the polymers (from 20 to 80 kDa) or in biologically active organisms. This can cause a synergistic fouling by internal clogging, pore blocking and cake formation [1, 33]. On the other hand, it may be assumed that cake formation might dominate for longer filtration cycles (more than 60 min) after pores have been blocked.

*TD*represents the Taylor development of Hermia model.

*i*and

*j*correspond to the designation of the Hermia model approximation and

*ɛ*is the error.

*TD*

*and*

_{pc}*TD*

*is observed after 45 min.*

_{ib}### 3.2. Screening Study

#### 3.2.1. The fouling

#### 3.2.2. The net water production

^{−1}(b5/2-1). A significant decrease in net water production is observed when the backwash flow increases to 20 L.min

^{−1}(b5/3-1) and reaches its highest value when BWF reaches 22.5 L.min

^{−1}(b5/4-1) (Fig. 7). This is expected because bigger backwash flow consumes more permeate water without acting on reversible and irreversible fouling. No further decrease in the net water production is observed when the BWF increases from 22.5 to 25 L.min

^{−1}.

### 4. Conclusions

^{−1}or more also engenders a significant decrease in the net water volume.