The chemical composition of the natural raw clay was analyzed using a wavelength dispersive X-ray fluorescence apparatus (Shimadzu, XRF-1800) to evaluate the proportions of impurities in the raw material as well as the Loss of Ignition (LOI). Elemental composition of the red sawdust sample was analyzed with energy dispersive analysis of X-ray spectroscopy (EDS) coupled to the Scanning Electron Microscope. Thermogravimetric analysis (TGA) and differential scanning calorimetry (DSC) analyses for both samples were performed using a 2960 TA Instrument under argon from room temperature up to 1200˚C at a heating rate of 5˚C/min. The analysis was performed on the raw material to evaluate the thermal stability and the minimum sintering temperature required for membrane fabrication.

It has been reported in the literature that numerous factors can influence the viscosity of ceramic paste [12] [13]. These factors are the size and proportion of the raw materials (clay, porogen and binder). In order to determine the optimum size and proportion of raw materials which influence the viscosity of ceramic paste for the production of tubular support by extrusion technique, the Response Surface Methodology (RSM) was used. It is a statistical method that uses experimental data obtained from specified experimental design to model and optimize any process in which response of interest is influenced by several variables [14]. Primarily, this optimization is done by following three major steps viz., performing the statistically designed experiments, estimating the coefficients in a mathematical model and predicting the responses and examining the adequacy of the model [15]. RSM helps to enumerate the relationships between output variables called responses (Y) and input variables called factors (X_i) [16].

Y = f ( X 1 , X 2 , X 3 , ⋯ , X n ) (1)

A standard RSM design called Box-Behnken design was applied in the present work. This method is suitable for fitting a quadratic surface and it helps to optimize the effective parameters with a minimum number of experiments, as well as to analyze the interaction between the parameters [17]. Generally, the Box-Behnken for 3 factors implies three blocks, in each one, a factorial design 2² which is represented by each pair of treatment while the third factor remains fixed at 0. Thus the Box-Behnken design for 3 factors has 15 experiments (12 edges and 3 center points). The centre points are utilized to evaluate the experimental error and the reproducibility of the data.

The experimental sequence was randomized in order to minimize the effects of uncontrolled factors [18]. The outcome of each experimental run was analyzed and the response was correlated with three input factors for the viscosity of the ceramic paste through an empirical second degree polynomial equation as given by the following equation:

Y i = β 0 + ∑ β i x i + ∑ β i i x i 2 + ∑ β i j x i x j + ε (2)

where, Y_i is the predicted response (viscosity of ceramic paste), β₀ the constant coefficient, β_i the linear coefficients, β_ij the interaction coefficients and β_ii the quadratic coefficient [19]. ANOVA was used to model the system represented by independent parameters and dependent output response and to optimize the system by estimating the statistical parameters. The minimum (−1), intermediate (0) and maximum (+1) level of the amount of porogen, size of porogen and amount of binder were 20 g, 30 g and 40 g, 50 g, 100 µm and 150 µm, and 2 g, 3.5 g and 5 g respectively. The statistical experimental design as specified by the software is shown in Table 1.

In order to put in the form of an equation the phenomenon observed and to make it possible to predict the answers in the field defined for the study, it is important to validate the empirical models obtained. To do that, the execution of the model was measured by comparing the values of the answers envisaged and those observed.

­ R² determination Coefficient

The analysis of the variance makes it possible to calculate very useful statistics: R² or R square. These statistics are the ratio of the sum of squares of the calculated answers (corrected average) to the sum of squares of the measured answers (corrected average):

R 2 = corrected average sum of squares of the calculated answers corrected average sum of squares of the measured answers (3)

In addition to the linear regression coefficient (R²), other mathematical procedures and tools were used. Thus, Absolute Average Deviation (AAD) which informs about the average manipulation error, the Biais factor (Bf) were given according to following expressions:

• Absolute Average Deviation (AAD)

AAD = ∑ i = 1 p ( | Y i e x p − Y i c a l | Y i e x p ) p (4)

With: Y_iexp the experimental answer and Y_ical the calculated response starting from the model for an experiment I; p being the total number of experiments.

• Biais factor (Bf)

Bf = 10 B

The Biais B is given by the relation:

B = 1 n ∑ ​ log ( Y t h e o Y o b s ) (5)

Each experiment was carried out in triplicate. MINITAB version 2017 and Statgraphics plus software were used to analyze the obtained data.

The preparation of the ceramic paste was achieved by dry mixing the clay, red sawdust and binder powders in order to homogenize them and then wet mixed by adding 80 ml of water. The paste is then aged by putting it in a plastic bag and placing it in the fridge for 24 h to ameliorate its plasticity to obtain the ceramic paste as shown in Figure 1 below. Table 4 shows the different sizes and proportions of raw material mixed to obtain a homogeneous paste. The viscosity of the paste was then measured using a viscosimetre marketed by the LAMY at 25˚C. A graph of viscosity against shear rate was then plotted in order to determine the flow behavior of the ceramic paste.

The ceramic paste produced from the optimal amount and size of raw material was fed into the extrusion cylinder. Then, evacuation piston forced the paste through a die in a tabletop extruder to form a tubular shape membrane. The tubular membrane was extruded in the horizontal at room temperature. When the tube reached the length of around 19 cm, the process extruder stopped and the tube was removed. Then, the obtained tubular membrane was subjected to natural drying at room temperature for 24 h. After this, the membrane was dried at 100˚C for 24 h in a hot air oven. Subsequently, the membrane was taken to the sintering process with a heating rate of 2.5˚C/min until 500˚C for 2 h and then 5˚C/min until 950˚C and sintered at that temperature for 2 h to obtain the tubular support as shown in Figure 2 below. These restrained thermal treatment steps were followed to avoid the formation of micro cracks and bends in the membrane. Finally, the elaborated membrane was washed with water and dried at 100˚C for further characterization.

The porosity of the membrane was measured by utilizing water as a soaking agent (Archimedes method) [4]. The porosity of the membrane is measured using the below expression.

Porosity ( % ) = W 1 − W 0 W 1 × 100 (6)

whereby the Dry filter is weighed (W₀) and put into water for about 4 h in order to free the pores allowing the distilled water to be in the pores and then reweighed (W₁) and the porosity is given by the difference between the mass of the filter before and after immersion in water divided by W₁.

The mechanical strength was determined by the compression test at three-point [5]. The corrosion resistance of the membrane was evaluated by means of loss of mass after treating in aggressive environments. The acid and alkali solutions were prepared as such HNO₃ solution with pH 3 and NaOH solution with pH 9 and the membrane was soaked into the solutions for 6 days. The corrosion resistance of the recovered membrane was evaluated by weight decrement of the membrane [20].

The water permeability (L_h) of the membrane was determined by the water flux at different applied pressures. It was evaluated from the slope of the pure water flux (J_W) versus applied pressure across the membrane (ΔP) using the equation

J W = L h Δ P (7)

Hagen-Poiseuille equation was employed to estimate the average pore size of the membrane by assuming pores are cylindrical in shape [21] [22].

J w = ε r 2 Δ P 8 μ τ l (8)

where, ε is the porosity of the membranes, r is the pore radius of the membrane (m), ΔP is the applied pressures across the membrane, l is the pore length which is generally taken as thickness of the membrane, τ is the tortuosity factor (generally used as 1), and μ is the viscosity of water (0.00089 Pa∙s). By combining Equations (9) and (10), the corresponding pore radius of the membrane can be evaluated from:

r = L h 8 μ τ l ε . (9)

The laboratory scale diagram of the overall tangential microfiltration system is presented in Figure 3. The elaborated tubular ceramic membrane was incorporated into the module for tangential cross flow microfiltration. The feed solution was pumped to the module from the feed tank and the operating pressure was adjusted by hand operation using the flow control valves (security van). The permeate was collected in the permeate tank while the concentrated collected in the concentrated tank. The flow of the solution is shown by the arrows in Figure 3 below.

The water flux of the permeate was measured at different applied pressures (0.5 - 2 bar) as a function of time by utilizing the equation below:

J w = Q A t (10)

where, Q is the permeated water in volume (m³); A is the effective membrane area (m²) and t is the measured time (s).

A clay suspension of 2 μm was prepared by soaking approximately 1 kg g of kaolinite clay sample in 3 L of distilled water for 24 hours. And let to rest for 8 hours to settle larger particles before decanting the liquid. The particle size distribution of the suspension is determined using master sizer 2000 and then brought to a turbidity of 100 NTU by dilution with distilled water using a Nephelometric Turbidimeter (HACH RATIO 2100A).

The tangential microfiltration study on the 2 μm synthetic water suspension of turbidity 100 NTU was performed at various applied pressures of 0.5 - 2 bar and the observed retention was determined by below expression:

R ( % ) = C f − C p C f × 100 (11)

where C_f is the suspension concentration in the feed stream, C_p is the suspension concentration in the permeate stream and R is the observed retention (%).

The experimental plan chosen is a Box-Behnken design with three centre points, each factor requires 3 levels, and the number of experiments is equal to 15 as shown in the experimental design given in Table 4. The residues between the computed values and experimental values, is presented in the table. The residues obtained are very weak and are taken in absolute value. In this table, one notes that the residues obtained are very weak.

Interpretation starts with the calculation of the coefficients of the model. The coefficients of the model are represented in Table 5 below. The statistical analyses showed that the values of the answers would adapt to a second degree polynomial model. It can be seen that the p-value for the linear factor amount of porogen, the p-value for the quadratic factor amount of porogen-amount of porogen and the p-value for the interaction amount of porogen-size of porogen are close to 1 and consequently their coefficient is close to 0 and thus negligible.

*: p-value close to 1.

Table 6 below gives the criterions of evaluation and validation of the model. The analyses of variance give an R² value of 97.8% which is greater than 95% meaning the model explains 97.8% the variability of the viscosity of the ceramic paste. The Biais factor is equal to the unit and the AAM (Absolute Average Deviation) equal to the zero thus model is regarded as valid. From the analyses the model obtained was:

Y paste = 24.800 − 0.062 X 1 + 2.975 X 2 + 1.687 X 3 − 6.150 X 1 2 − 4.525 X 2 2     − 0.050 X 3 2 − 1.225 X 1 ∗ X 2 + 2.100 X 1 ∗ X 3 − 0.925 X 2 ∗ X 3

where X₁ = Amount of porogen, X₂ = Size of porogen, X₃ = Amount of binder, Y = Viscosity of ceramic paste.

From the equation of model, the parameters of validation were determined and consigned in Table 6.

• Analysis of variance

An analysis of the variance, presented in Table 7, made it possible to test the validity of the model. The significance statistical tests (test F) showed that the variation due to the regression is significant and that related to the lack of adjustment is non-significant. These results show that the recommended model is valid.

• Contribution of the various factors

The diagram of contribution informs about the significance of the operational parameters. The negative or positive impact of the variables (amount of porogen, size of porogen and amount of binder) on the output (viscosity of the paste). Figure 7 below shows Pareto Diagram of the viscosity of ceramic paste. From the diagram, we observe that the size of porogen has the highest positive effect on the viscosity of the paste followed by the amount of binder. Meanwhile, the interaction amount of porogen * amount of binder has the only positive effect on the viscosity of the paste. For the quadratic factors, the amount of porogen has the highest negative influence on the viscosity of the paste. Implying that as the amount of porogen is doubled, it influences negatively the viscosity of the paste.

Finally, Figures 8-10 present graphical studies in two forms: response surfaces (two dimensional 2D) and iso-response curves (three-dimensional 3D). A major advantage of the design of this experiment is the ability to generate multiple graphic representations of the data. The shaded potion on the iso-response curves represents the optimal zone for the viscosity of the ceramic paste for each interaction.

For the interaction size of porogen * amount of porogen as shown in Figure 8, for a size of porogen in the range 90 µm to 140 µm and for an amount of porogen in the range 25 g to 35 g, we have an optimum (maximum) viscosity of the pate of 24 Pa∙s. Below and above this range of values the viscosity reduces.

For the interaction amount of binder * amount of porogen as shown in Figure 9, for an amount of binder in the range 4.3 g to 5 g and for an amount of porogen in the range 28 g to 35 g, we have an optimum (maximum) viscosity of 26 Pa∙s. Below and above this range of values the viscosity reduces.

Table 8 below gives the optimum amount and size of raw material used for the production of the tubular support by the extrusion technique. We have an optimum size of raw material of 100 µm for an amount of clay of 66 g, amount of porogen of 30 g and amount of binder of 4 g. The optimum viscosity of the ceramic paste obtained is 26.7 Pa∙s which is close to the viscosity of the clay paste of 28.5 Pa∙s, thus good for shaping by the extrusion technique.

Figure 11 below shows the flow curve for the ceramic paste. It is a graph of viscosity of the ceramic paste against shear rate. We observe on the graph that the more shear rate is applied, the thinner the material becomes. As soon as the shear stress is removed, the viscosity of the fluid will regain its original state. The ceramic paste showed a pseudoplastic behavior.

Table 9 summarizes the properties of the prepared tubular ceramic membrane. Figure 12 shows the tubular support before and after sintering. After the sintering of the tubes, certain observations were made on their geometry and their appearance. With regard to the geometry, it was observed that the membranes preserved their geometrical configuration such as outer and inner diameters and length of the tube were 4 cm, 2 cm, and 19 cm, respectively. On appearance, the heat treatment brought about a change of the color of the tube. This change of colour is due to the presence of Fe (III) in the reaction medium. During sintering, under the influence of the temperature, there is a reaction of oxidation of the iron (II) to iron (III) whose presence in the medium is identified by the reddish colouring, which intensifies according to the content of iron (III). Similar results were obtained by Belibi et al., 2015. It’s concluded that the rate of Fe₂O₃ in the clay material and the atmosphere of the furnace is responsible for the colouring of ceramics [23].

The shrinkage in mass of the membrane after drying at 105˚C was calculated to be 7.63% and this is due to the departure of interstitial water contained the filters. Meanwhile, after sintering at 950˚C, the shrinkage in mass was calculated to be 31.45%. This reduction is due to the departure of linked water and the water of constitution. This value also corresponds to the loss in mass of the porogen (red sawdust) giving a porous support after sintering.

The open accessible porosity of the fired membrane was found around 43.5%, which is in accordance with values reported in the literature for efficient microfiltration membranes. Porosity is greater than 35% which is accepted for ceramic filtration [30].

The average pore size of the membrane was calculated to be 2.4 µm (2.4 × 10⁻⁶ m). The formulated support is a macroporous membrane adapted for microfiltration. The membrane can be directly utilized for making of multilayer membranes with exclusion of intermediate layers [31].

The average mechanical stability of the membrane was calculated to be 9.2 MPa. This result is in harmony with the results of porosity that showed that the porosity was almost the same for the different samples. In fact, the mechanical strength of ceramic drops when porosity increases [32] [33].

The corrosion resistance test was performed with respect to weight decrement of the membrane after keeping in harsh environments (acid and alkali). The weight decrement of the membrane is calculated to be 1.9% in acidic condition and 1.4% in alkali condition. In acid and alkali conditions, the membrane displays an excellent result in resisting corrosion. The loss of weight is < 5% indicating the support is stable chemically. The obtained results are more commensurable in comparison with the cordierite membrane fabricated by Dong et al. [34].

Figure 13(a) illustrates the water flux of the membrane as a function of time for different applied pressures. A steady flux is been attained with time. The variations of applied pressure on water flux are also presented in Figure 13(b). It can be noticed that the water flux increases linearly with an increase of applied pressures (0.5 - 2 bar). This stipulates that the variation in pressure is the barely driving force for permeation. For transportation operation exclusively by convection, the flow rate is proportionate to the pressure, and is in accordance with Darcy’s law. The water permeability (L_h) of the membrane was determined by the water flux at different applied pressures. The water permeability (L_h) of the membrane was found to be 244.87 L/h∙m² bar.

Applied pressure and feed concentration are the important variables that affect the treatment process in terms of permeate flux and rejection. Hence, the effect of these parameters was investigated as a function of time.

Figure 14(a) shows the permeate flux profiles with time for various applied pressures (0.5 - 2 bar) with feed particle size and concentration of 2 µm and 100 NTU respectively. It can be seen from the figure that the permeate flux declines sharply at the initial period and becomes gradual there-after in the microfiltration run.

The reasons for the flux reductions are owing to concentration polarization on the surface of membrane and blocking of pore in the porous ceramic structure. In addition, it is noticed that the permeate flux increases with increasing applied pressures due to the higher driving forces acted on the membrane. However, the permeate flux profile on increasing applied pressure does not follow the linear trend (Figure 14(b)). This is due to concentration polarization surface that creates further resistances to transport liquid through the membrane [31]. The rate of flux decline is higher at higher applied pressure due to the immediate formation of a clay layer on the surface of the membrane, which makes membrane fouling faster while increasing the applied pressure.

Figure 15 presents the retention of the clay particles with time for various applied pressures. It is noticed that the retention increases with an increase in applied pressure. This occurs due to the fact that higher pressures assist the enrichment of the clay cake formed on the filter leading to pore blockage. The highest retention was obtained at 2 bar with permeability of 184.5 L/h∙m² bar with a retention of 96% for suspensions of 100 NTU decreasing it to 4 NTU. The obtained value of turbidity is conformed to the standards for turbidity value set by the WHO (2012) for potable water which is of 5 NTU [3].