• Tidak ada hasil yang ditemukan

Transesterification experiments

Ultrasound–Assisted Biodiesel Production Using Heterogeneous Base

3.3 Results and discussion

3.3.3 Transesterification experiments

Feedstock optimization: The pseudo–component mixture design with 36 experimental sets and corresponding triglyceride conversion with standard deviation is shown in Table 3.1. In feedstock optimization the esterified oils are blended in different volumetric propositions. From Table 3.1, it can be seen that, no single oil feedstock has significant dominance in blends. The highest conversion of triglyceride was obtained with experimental set 11, followed by set 1 and set 10. Some major observations from these experiments are as follows:

1. As the volume of Castor oil increases above 20% in the mixed feedstock, the transesterification yield reduces significantly. The decrease in transesterification yield attributed to high viscosity of castor oil, which increases the overall feedstock viscosity.

This is manifested in terms of mass transfer limitations between oil and methanol phase.

This result also suggests that in order to enhance the transesterification yield, volume fraction of castor oil in mixed feedstock should be kept minimum (< 0.1 v/v).

2. Blending of Jatropha oil at moderate to higher volumetric ratio (> 0.275 v/v) has adverse impact on transesterification yield. Reduction in transesterification yield is mainly due to its high acid value even after esterification reaction. Thus low to moderate (< 0.275 v/v) fraction of Jatropha oil to should be blended to enhance the transesterification yield.

3. Waste cooking oil in mixed oil feedstock shows mixed response on transesterification yield. In some experimental sets, moderate or higher volume fraction of waste cooking oil shows good transesterification yield, while in other experimental sets it resulted in lower transesterification yields. This mixed reaction depends on the volume fractions of other oils especially volume fraction of Castor, Jatropha and Palm oil. Hence, the moderate volume fraction (0.1 to 0.25 v/v) of waste cooking oil is

favourable for higher yield of transesterification reaction.

4. Blending of Rubber seed oil to mixed oil feedstock improved the yield of transesterification reaction. This could be attributed to its relatively low viscosity, which helps in mixing of other oils and methanol with each other. The highest yield of transesterification was obtained at 50% volume of rubber seed oil in feedstock blend.

5. Addition of Palm oil to mixture feedstock showed effect as that of waste cooking oil, i.e. for some experimental sets with higher volume fraction of palm oil, enhanced the transesterification yield and for some experimental sets, it reduced the yield. Palm oil has second highest viscosity after Castor oil among all oils used in feedstock. On the other hand, the acid value of esterified palm oil is low (0.95 mg KOH/g), which helps promoting the transesterification reaction. However, for volume fraction of palm oil > 0.2, overall viscosity of feedstock increases leading to lower mass transfer between oil and methanol. Therefore, moderate volume fraction (0.1 to 0.2 v/v) of Palm oil in blended feedstock results higher transesterification yield.

On the basis of above observations and experimental results, the optimum volume fractions of different oils in mixed oil feedstock were selected as experimental set 11. The experimental set 1 and 10 are also good alternatives for large scale transesterification process in case of non–availability of any oil feedstock.

Process optimization: After initial screening of different oil mixtures for transesterification reaction, an attempt was made to maximize the yield of transesterification reaction by varying the operating parameters such as alcohol: oil molar ratio, operating temperature of reaction and addition of catalyst to reaction mixture. The process optimization was done using Box–Behnken statistical design coupled with response surface methodology. Statistical optimization helps in understanding the interaction between two operating parameters.

The experimental transesterification yield and the model predicted yield for 15 experimental sets of Box–Behnken statistical design are tabulated in Table 3.2B. The experimental yield consisted of the average of two runs and corresponding standard deviation, as the experiment for each set was carried out in duplicate to validate reproducibility of results. A quadratic equation was fitted to the experimental data using coded values of process parameters as follows:

2 2 2

86.34 8.49 8.68 14.46 21.50 16.66 22.54 6.65 9.08 0.64 Y C M T C M T C M T C MT

From Table 3.2 (B), the experimental results showed a close match with model–

predicted values of triglyceride conversion, indicating that the model prediction matches well to the experimental results. This statement is also strengthened by the regression coefficients values, viz. R2 = 0.9998; R2 (predicted) = 0.9979; R2 (adjusted)

=0.9995. The ANOVA (analysis of variance) of the fitted model predicted various coefficients of the quadratic model such as linear, square and interaction coefficients.

p– and t– values of these coefficients are listed in Table 3.3. From ANOVA predicted coefficients, large (absolute) t–stat value and p–value < 0.05 indicated the consequence of the coefficient and thus, corresponding process parameter. F–values related to coefficients of linear, interaction and quadratic variables, indicated the importance of the individual effect of corresponding optimization variable and the magnitude of interaction between them. From ANOVA results the p–values of all coefficients are <

0.05, which indicates all the process parameters have significant impact on process optimization. Based on p–values of interactions, the interaction between catalyst loading and reaction temperature is most significant followed by the interaction between molar ratio and catalyst, while the temperature and molar ratio showed the least interaction. The F–value and p–value of Lack–of–Fit were 2.01 and 0.351 respectively, which denotes that Lack–of–Fit is not significant with compared to pure

error or in other way, the model is significant.

Table 3.3: Statistical analysis of experimental results

(A) Estimated regression coefficients for triglyceride conversion %

Term Coefficients SE coeff t–stat p–value

Constant () 86.34 0.288 299.75 0

Catalyst (C) 8.49 0.1764 48.12 0.002

Molar ratio (M) 8.68 0.1764 49.2 0

Temperature (T) 14.46 0.1764 81.96 0

Catalyst × Catalyst (C2) –21.5 0.2596 –82.79 0.008 Molar ratio × Molar ratio (M2) –16.66 0.2596 –64.17 0 Temperature × Temperature (T2) –22.54 0.2596 –86.82 0 Molar ratio × Catalyst (MC) 6.65 0.2495 26.66 0.006 Temperature × Catalyst (TC) 9.08 0.2495 36.4 0 Temperature × Molar ratio (TM) 0.64 0.2495 2.57 0.01

(B) Analysis of variance (ANOVA) for transesterification reaction Source Degrees of freedom Sq SS F–value p–value

Regression 9 7367 3288.56 0

Linear 3 2850.85 3817.78 0

Square 3 4008.2 5367.67 0

Interaction 3 507.95 680.23 0

Residual Error 5 1.24 – –

Lack–of–Fit 3 0.93 2.01 0.35

Pure Error 2 0.31 – –

Regression 14 7368.24 – –

R2 = 99.98%; R2 (pred) = 99.79%; R2 (adj) = 99.95%

The response surface plots for the quadratic model are shown in Fig. 3.5, which indicated the interaction between two process parameters on transesterification yield, with retaining the third parameter at its centre point. The surface plots are graphical presentations of quadratic equation. The top (dark) colour region displays to the maximum transesterification yield. The surface plots indicated the strong interaction between the parameters as same predicted by ANOVA analysis.

The optimum set of parameters predicted by the quadratic model was: catalyst loading = 7% (w/w); molar ratio (alcohol/oil) = 11.68:1; temperature = 332 K. The validation experiment was performed at optimum set of parameters. Triglyceride

conversion in the validation experiment was 92.35 ± 1.08%, which is very close to the triglyceride conversion of 92.03% predicted by the model.

(A) (B)

(C)

Figure 3.5: Contour plots depicting interactions among parameters for statistical optimization of transesterification process (A) molar ratio vs catalyst loading; (B) temperature vs catalyst loading and (C) molar ratio vs temperature