The pretreatment was carried out at varying temperatures (60-90 °C) in a shaking incubator for varying times (60-120 min) using ZnCl₂.4H₂O/ Urea at varying concentration ratios (50-90% w/w) (Table 1) at a constant biomass to solvent ratio of 1:10 (5 g dry biomass will be added to 50 g of each solvent) and at a constant agitation speed of 270 rpm. The slurries obtained from each pretreatment method were recovered and separated into solid and liquid fractions by vacuum filtration.

The compositional analysis of biomass were analysed for hemicellulose, lignin, extractive and ash content as documented by the National Renewable Energy Laboratory (NREL) following Laboratory Analytical Procedures (LAP TP-510- 42618, 42619, 42622) [21-23].

The experimental design is shown in Table 1, the experimental data were analysed using RSM to fit the second order polynomial equation created by Design-Expert 7 software (Stat-Ease Inc., USA) The parameters were selected for the studies as follows: reaction temperature (60-90°C), ZnCl₂. 4H₂O/ Urea molar ratio (50- 90% w/w) and time (60-120 min). The variable ranges are shown in Table 1. The ranges were selected to keep the intervals within mild reaction conditions. Table 1 describes the five coded levels (−α, −1, 0, 1, α) and un-coded independent factors, the test ranges for the parameters and the overall experimental design. The CCD design made up of 20 experiments (8 factorial points, 6 centre points and 6 axial point) were carried out with three variables and each variable varied at three levels to allow for the evaluation of the curvature [24]. The value of α is the distance from the axial points to the central point α= (2n) 1/4, in which n is the number of independent factors, for three factors α=1.68 [25]. The factorial design experiments were run in duplicates to ensure reliability; the average error was calculated and used.

The experimental data was used to develop a quadratic polynomial model having % cellulose recovery (Y₁), % hemicellulose recovery in the liquid fraction (Y₂) and % lignin recovery (Y₃). The general regression model (Eq. (1)) was adopted.

(1)

Where A, B, C represent the symbols of the independent variables, temperature, time and solvent concentration, respectively. α₀; α₁, α₂ and α₃ are the regression coefficients for the linear terms of the expression and α₁₁, α₂₂, α₃₃α₁₂, α₁₃ and α₂₃ are the regression coefficients for the quadric terms of the equation. All experiments were done in duplicate and average values were calculated and used in the development of the model. The statistical analysis, plotting of the response surfaces and the optimization were performed using “Design Expert” software (Trial version 7.0.3, Stat-Ease, Inc., Minneapolis, USA). The fit quality of the polynomial equation was assessed by the coefficient of determination (R²), the significance of the regression and statistical coefficient was checked using t-test and F-test, respectively. The predictability of the developed model was then cross validated using laboratory experiments.

Numerical optimization by desirability function was carried out to optimize the operating conditions of the pretreatment. The desirability function technique can be used to determine the optimum conditions of a process which involves one or more responses. The desirability procedure first determines the independent variable levels that simultaneously produce the most desirable prediction for the responses, thereafter the overall desirability of the outcome is maximised based on the factors which can be controlled.

In numerical optimization, the desired goal for each factor and response was chosen. The goals defined were as follows: maximise, minimise, target, exact value (for factors only), none (for responses only) and within range. Maximum and minimum levels for each factor were chosen. The factors can also be weighted according to priority. The goals were combined to get a desirability function. The desirability function is an objective value ranging from zero (outside the goal) and one representing the goal, the program thus aims at maximising this function. The goal seeking step starts randomly from one point and makes its way to a maximal point through the steepest slope.

CCD was employed using the Design-Expert (Stat-Ease, Inc., Minneapolis, USA) software to study the simultaneous effects of reaction time, temperature and solvent concentration of pretreatment on cellulose recovery, hemicellulose recovery in the liquid fraction and lignin recovery. The experimental conditions were obtained by a complete three-factor experimental design, which included six replications of the centre point and six axial points and eight factorial points. Table 2 shows the series of conditions of the experiments conducted for each run.

The detailed recovery of the components of ZnCl₂.4H₂O/ Urea treated solid fraction for each run are shown in Table 2. The highest solids recoveries of cellulose for the pre-treatment were obtained in run 10 (39.55 min, 75 ^oC, 70/30% (w/w) ZnCl₂.4H₂O /urea) and 12 (90 min, 49.77 ^oC, 70/30% (w/w) ZnCl₂.4H₂O /Urea) with 87.6% and 86.1%, respectively. This shows that cellulose is not less affected by the severity of the pretreatment as opposed to hemicellulose. Hemicellulose recoveries were lowest at run 8 (120 min, 90°C, 90/10% (w/w) ZnCl₂.4H₂O/Urea) and for run 4 (90 min, 49.11 °C, 70/30% (w/w) ZnCl₂.4H₂O /Urea). These severities are considered high due to longer times and higher temperatures. This is in agreement with the various studies which state that hemicellulose is more susceptible to pretreatment conditions [26]. Additionally, the lowest lignin recovery for the trials were obtained during run 3 (60 min, 75, 50%), 4 (60 min, 100.23°C, 70/30% (w/w) ZnCl₂.4H₂O /Urea) 8 (120 min, 90°C, 90/10% (w/w) ZnCl₂.4H₂O /Urea).

The experimental data was analysed using the Design-Expert software and a second-order polynomial equation was obtained. The equation included linear, quadratic and interactive terms for the variables tested. The polynomial equations, describing the cellulose recovery (Y₁), hemicellulose recovery (Y₂) and lignin recovery (Y₃) as a function of reaction time (A), temperature (B) and solvent concentration (C) of ZnCl₂.4H₂O/ Urea MHS pretreatment are shown in Eq. (2), Eq. (3) and Eq. (4):

(2)

(3)

(4)

Eqs. (2 - 4) are predictive regression models that could be used to explain the main effect of the individual variables and as well as the interactive effect on the responses considered. Furthermore, it could be explained that the negativity of the regression coefficients indicates (as a rule of thumb) the negative effect of the factors on the responses (decrease the response). For example, solvent concentration (C) decreases cellulose recovery (Y₁). Whereas the positivity of the regression coefficients indicates (as a rule of thumb) the positive effect of the factors on the responses (enhancement in the quantity of the response).

3.3.3. Pareto Analysis

The percentage effect of each of the factors was done by the Pareto analysis [16]. The analysis calculates the percentage effect (Pi) of each factor on the response, according to the following equation:

(5)

Where βi is the regression coefficient of each process variable.

Figs. (1, 2 and 3) shows the Pareto plot. As seen in Fig. (1) under the ZnCl₂.4H₂O/ Urea pretreatment, the most significant factor in cellulose recovery are B (Temperature) and C (Solvent concentration) in both linear and interactive forms. The interactive form (AB) however has the highest influence of 69.07%. Hemicellulose recovery in the liquid fraction was mainly affected by temperature (linear term), followed by interactive term (BC), solvent concentration (linear term), time (linear term) and interaction of time and temperature. Finally, lignin recovery was affected primarily and almost entirely by temperature (linear term) with 88.72% effect. Followed by solvent concentration (linear term) at 4.7%. The highest interactive term was temperature*solvent concentration (BC) with 2.2% effect on lignin recovery.

3.3.4. Model Fitting

The reliability of the predicted models was additionally tested by comparing the predicted and experimental response results. The observed recovery from the experiment and the predicted values are shown and represented as a parity plot in Fig. (4) with the R² values for each response. In each case, there was satisfactory correlation between the predicted values and observed values from the experiment which was R²=0.9021, 0.8531, 0.8920 for Figs. (4A, 4B and 4C), respectively. The values indicate close agreement between the experimental and predicted values for each response. In Figs. (4A, 4B, 4C), 9.8%, 14.69% and 22.3% of the total variation, respectively, was not explained by the model.

Fig. (1). Pareto chart showing the percentage effect of each factor in cellulose recovery.

Fig. (2). Pareto chart showing the percentage effect of each factor on hemicellulose recovery in the liquid fraction.

Fig. (3). Pareto chart showing the percentage effect of each factor on lignin recovery.

The three-dimensional response surface plot described by the above-mentioned polynomial Eq. (2), Eq. (3) and Eq. (4) were fitted to the experimental data as represented in Fig. (4). One of the variables was kept constant (at midpoint) while the other two variables were varied according to the experimental ranges determined previously. The shapes of the response surfaces indicate the type and amount of the interaction between the different independent variables [31]. The surface plots for the cellulose, hemicellulose and lignin recovery are shown in Fig. (5A, 5B and 5C). Fig. (5A) shows that the pretreatment temperature had the major increasing effect on cellulose recovery. As seen in Fig. (5A), longer times result in higher cellulose recovery only for the pretreatment at higher temperatures while the effect of longer pretreatment time was negligible at lower pretreatment temperatures. At higher solvent concentrations, cellulose recovery increased, however, as the concentration increased, the effects of increasing temperature became negligible after 77 ^oC. Overall, the cellulose recovery was maximum at 100.23% with only a small increase when time is increased to approximately 90 min.

The response surface plots for the hemicellulose recovery in the liquid fraction are shown in Fig. (5B). The results showed that temperature also has a large influence on hemicellulose recovery in the liquid fraction. Fig. (5B) shows a decrease in the recovery of hemicellulose in the liquid fraction with increasing concentration until 75% where an increase is observed thereafter. The largest recovery in the liquid fraction is observed at higher concentration and longer time. In Fig. (5B), it was observed that at lower concentrations, the effect of temperature is negligible. While at lower temperatures, the effect of concentration is negligible. However, at higher temperatures and concentrations, we notice an increase in the hemicellulose recovery in the liquid fraction. As expected, the recovery of hemicellulose is highly dependent on the severity of the experiment. This is in agreement with various other studies which have shown a dependence of hemicellulose on severity compared to the other components [32]. The graph of the interaction between concentration and temperature (Fig. 5B) gives the maximum recovery of 24.6% at 100% concentration, 100.23 ^oC and time at the midpoint of 75 minutes; these are the best conditions for achieving high yields of recovered hemicellulose in the liquid fraction. This validates the results of the Pareto chart in Fig. (2) where Temperature* Solvent concentration (BC) had the highest interactive influence on the response.

Fig. (5C) represents the effects of temperature and time on lignin recovery at a constant solvent concentration of 75%. An increase in temperature resulted in a linear increase in lignin recovery. The length of time had no influence on the lignin recovery at lower temperatures; however, the lignin recovery was maximal at shorter times. Results in Fig. (5C) shows that an increase in time caused a linear decrease in lignin recovery at low solvent concentrations, however at higher solvent concentrations; there is a slight decrease in lignin recovery with increasing time. Solvent concentration at shorter times resulted in a decrease in lignin recovery until approximately 62.50% thereafter, an increase in lignin recovery was observed. Fig. (5C) shows the effect of temperature and concentration on lignin recovery at a constant time of 75 min. Solvent concentration has a negligible effect on the recovery of lignin at lower temperatures. There is however a slight increase in lignin recovery with increasing concentration. Temperature has a linear effect on lignin recovery as the increase in temperature results in an increase in the response at both high and low solvent concentrations. Figure 4.8C shows the highest amount of recovery (69%) and this agrees with the Pareto chart (Fig. 3) where BC is shown to have the highest interactive effect on lignin recovery

Fig. (4). Parity plot showing distribution of predicted vs. actual values of responses. (A) Cellulose recovery. (B) Hemicellulose recovery in the liquid fraction. (C) Lignin recovery.

Fig. (5). Response surface plot showing effect of pretreatment conditions on (A) Cellulose recovery. (B) Hemicellulose recovery (C) Lignin recovery. Midpoint conditions 75%/25% (w/w) ZnCl2.4H2O/ Urea, 75 °C, 75 min.

Fig. (6). Desirability ramp of cellulose recovery, hemicellulose recovery in liquid fraction and lignin recovery.

The mathematical model which was developed from the experimental data can be used for the effective prediction of the operating conditions which should be used for the optimization of the model responses. The strategy that was used can determine the pretreatment temperature, pretreatment time and solvent concentration, which leads to maximum responses. Desirability Function (DF) assigns a scale-free value to all responses in the problem by the so-called individual desirability functions and then combine these values by usually taking their geometric mean to obtain an overall desirability function as a single objective [27]. This multiple response numerical method was applied for the optimization of any combination of three factors (time, temperature, concentration). Fig. (6) was obtained from design expert analysis and demonstrates the desirability values of the numerical optimization procedure in which the criteria were set as follows: “in range” for time (60 min), “in range” for temperature (70-120°C), “in range” for concentration (50-90%) the objective was to optimize (maximize) the responses, cellulose recovery, hemicellulose recovery in the liquid fraction and lignin recovery to analyse economically viable optimal conditions. A variety of solutions were found however the best local maximum was shown as a desirability plot (Fig. 6). The desirability plot shown in Fig. (6), provides the optimized treatment conditions of 90 min, 120°C and concentration of 71.32% (w/w) ZnCl₂. 4H₂O/ Urea. At this condition, cellulose recovery was 99.03%, hemicellulose recovery in the liquid fraction was 27.18% and lignin recovery was 72.43% and desirability was 0.902. These optimum values were cross validated experimentally which resulted in 91% cellulose recovery, 29% hemicellulose recovery in the liquid fraction and 68% lignin recovery at the same conditions. These values indicate the suitability and accuracy of the model. Apparently, the recovery of cellulose needs different conditions compared to hemicellulose recovery in liquid fraction and lignin recovery. These were the best conditions to favourably optimize all three responses.

The optimized value for the solvent concentration is 71.32% (w/w) ZnCl₂.4H₂O/ Urea. This low solvent concentration is advantageous as it reduces the viscosity of the solvent. This reduces some challenges, which are usually encountered with higher concentrations of MHS. The optimized temperature of 120°C is higher than the glass transition temperature of both lignin and hemicellulose, which are 40°C and from 50°C – 100°C, respectively. This temperature is thus expected to delignify the biomass as well as increase the dissolution of hemicellulose to an extent [33]. The pretreatment time of 90 min is desirable as it is expected to reduce the degradation of the polysaccharide components (cellulose and hemicellulose) which results in an increase in polysaccharide yield and decrease in fermentation inhibitors. The obtained desirable conditions differ largely from those obtained in the pretreatment of wheat straw using ionic liquid pretreatment. The conditions found in the study performed by Fu and Mazza were 158°C, 49.5% ionic liquid concentration and 216 min [34]. The large difference in the optimal conditions is caused by the responses for the optimization being based on the recovery of fermentable sugars in the pretreatment of wheat straw as opposed to whole fractions of the components as calculated in this study [34].