Multi-response optimization and neural network modeling for parameter precision in heat reflux extraction of spice oleoresins from two pepper cultivars (Piper nigrum)

2.1 Materials

The standard grade dried black (4% moisture) and white peppers were bought from Malaysian Pepper Board (MPB) in Sarawak. The samples were then pulverized into powdery form using a Knife Mill Grindomix grinder (GM-200 model, Germany). The powdered samples were thereafter sieved into five different homogenous particle sizes (0.105 mm, 0.154 mm, 0.300 mm, 0.450 mm and 0.900 mm) and preserved in an airtight plastic.

2.2 Extraction of fixed oil through heat reflux technique

An accurately weighed pepper sample (25 g) was loaded into a 1 L volumetric flask containing distilled water with the method used by Mohan et al. (2013). The sample was then filtered and concentrated using rotary evaporator (BUCHI, R-200 model, Germany). The extraction yield was calculated on the dry basis and stored at 4 °C in a dark vial prior to further analysis.

2.3 Determination of specific energy absorbed

The energy absorbed by the system is an important variable to measure during the extraction process. A proper energy audit of the heat reflux system requires a balance between the total energy emitted and the energy absorbed by the water-sample. In this study, the energy absorbed by the water-pepper system consisted of the energy required in vaporizing the solvent plus the energy required for actual extraction as presented in the following sets of heat balance equations (Eqs. (1)(3)):

2.4 Modeling and optimization studies

2.5 Physicochemical characterization

3.1 Multi-response optimization

The results obtained from black pepper heat reflux extraction placed the optimum extraction condition at 240 min of extraction time, 0.154 mm of particle size, and 1:14 of feed-solvent ratio. The extraction yield and energy absorbed under the optimum condition were, 13.22 (mg extracts/25 g sample) and 285.60 J/min, respectively. However, in white pepper heat reflux extraction, the optimum extraction condition was attained at 180 min of extraction time, 0.105 mm of particle size of, and 1:18 of feed-solvent ratio. Under this condition, the total extracts and energy absorbed was 14.04 (mg extracts/25 g sample) and 264.82 J/min, respectively. These were the condition at which the extraction parameters jointly optimized the extraction yield and absorbed microwave energy (Table 2). Hence, by using the above conditions, an improved extraction yield is achievable at optimized energy absorption capacity.

3.2 Validation of orthogonal design results

The predicted optimal responses were validated using triplicate parallel tests as presented in Table 3. From the optimal yield and absorbed energy (y_b = 13.22 mg/g, Q_tb = 285.60 J/min) three confirmatory tests were conducted on the black pepper heat refluxation and the estimated X-goodness-of-fit obtained from the confirmatory tests was 0.014 and 0.042 from extraction yield and absorbed energy, respectively. Similarly, three parallel tests were also conducted from predicted optimal yield and absorbed energy in white pepper heat refluxation (y_w = 14.04 mg/g, Q_{t w} = 264.82 J/min). The estimated X-goodness-of-fit test values from white pepper refluxation were 0.078 and 0.071 for extraction yield and absorbed energy, respectively. In both cases, the X-goodness-of-fit test between the predicted and experimented responses clearly revealed a smaller X-values when compared with the 7.81 cut-off value for a 3-degree of freedom at 95% confidence level (Alara et al., 2019; Olalere et al., 2017; Subroto et al., 2015). This confirmed a no-significant difference between the predicted and actual optimum response values.

3.3 Statistical analysis of mean (ANOM)

The average response means, the extremum or delta difference (β_max − β_min) were computed with β_max and β_min being the largest and smallest average response means, respectively. The mean effects for each extraction factor were determined using the value generated from their extremum differences. The average means for the extraction yield and absorbed energy are presented in Table 4.

Fig. 1 (a) showed the individualistic effects of each extraction variable on the average yield. The maximum average yield obtained were 10.2 mg/g at level-3 of extraction time (240 min), 9.53 mg/g at level-2 of particle size (0.154 mm), and 9.05 mg/g at level-2 of feed-solvent ratio (1:16). Meanwhile, in Fig. 1 (b), the maximum average absorbed energy were 278.00 J/min at level-3 of extraction time (240 min), 274.60 J/min at level-2 of particle size (0.154 mm), and 276.90 J/min at level-2 of feed-solvent ratio (1:16). The order of contribution of the extraction parameters at different operating levels to the average yield and absorbed energy was A > C > B. The decreasing order of significance affirmed extraction time as the largest contributor to spice oleoresin extraction from black pepper. The higher time requirement in black pepper heat reflux could be attributed to the higher inter-molecular forces of attraction between their parenchymatous cells (Olalere et al., 2017). This indicated that the production of black pepper from an immature peppercorn has effect on the time requirement during the heat reflux extraction (Zaveri et al., 2010). However, particle size has the least effects on the extraction yield and energy absorbed in the heat reflux extraction of black pepper.

Close

Fig. 1

Average mean effects-black pepper (a) average yield black pepper (b) average energy black pepper.

Export to PPT

In white pepper heat refluxation, the maximum average yield was 11.90 mg/g at level-2 of extraction time (180 min), 11.56 mg/g at level-1 of particle size (0.105 mm), and 12.22 mg/g at level-2 of feed-solvent ratio (1:16) as illustrated in Fig. 2(a). Furthermore, maximum average energy absorbed was 241.50 J/min at level-2 of extraction time (180 min), 247.70 J/min at level-2 of particle size (0.154 mm), and 254.00 J/min at level-2 of feed-solvent ratio (1:16) as shown in Fig. 2(b). The extraction factor with the highest extremum difference was declared as having the most contributing effect on the response settings (Abbass et al., 2016). In this case, the feed-solvent ratio gave the highest contribution to the extraction yield and absorbed energy. The extracting solvent (water) therefore provided a good interaction with the pepper matrix with a decreasing order of C > A > B. A threshold solvent volume of 400 ml provided a strong interaction between the solvent and pepper sample (Olalere et al., 2017; Olawale, 2012). The result therefore indicated that feed-to-solvent ratio (C) has the most significant contribution to the extraction of fixed oil from white pepper. However, the particle size has the least contribution to the extraction yield and energy absorbed by white pepper in heat reflux extraction.

Close

Fig. 2

Average mean effects-white pepper (a) average yield white pepper (b) average energy white pepper.

Export to PPT

3.4 ANN prediction

In the ANN model prediction, the network inputs were the extraction time (A), particle size (B) and feed-solvent ratio (C). The extraction yield and absorbed energy are the output variable in black and white heat reflux extraction. The Gradient Descent and Levenberg–Marquardt training algorithms were hidden layers used in conjunction with tansigmoid transfer function. The 3-4-5-1 topology was employed and this signifies that the model has 3 input layers, 4 neurons in the first hidden layer, 5 neurons in the second hidden layer and one output layer. The networks were trained severally for each response variable until both the coefficient of regression for training and prediction (between the simulated responses and expected responses) increases.

3.4.1

3.4.1 ANN prediction using Gradient Descent algorithm

The overall regression coefficient of 0.9370 and root mean squared error of 0.8849 was obtained for the optimum yield from black pepper. While an overall regression coefficient and root mean squared error of 0.9387 and 0.2565 were generated for the optimum yield from white pepper. Similarly, the overall regression coefficient and root mean squared error was 0.9959 and 0.7725 for optimal energy absorbed in black pepper heat reflux. Whereas, for the optimum yield in white pepper heat reflux, the overall regression coefficient and root mean squared error was 0.9114 and 0.3206 respectively. The result shows a good agreement with the optimality of Taguchi experimental design data as presented in Table 5.

Table 5 Network training configuration and results (Gradient Descent).

Sample name	Output	Transfer function	Model topology	MSE	RMSE	R2
Black pepper	yb	tansig	3-4-5-1	0.7831	0.8849	0.9370
	Qtb	tansig	3-4-5-1	0.5968	0.7725	0.9959
White pepper	yw	tansig	3-4-5-1	0.0658	0.2565	0.9387
	Qtw	tansig	3-4-5-1	0.1028	0.3206	0.9114

Fig. 3 is the post regression plot of predicted results and the expected trained network response. The primary aim of training and retraining the network is to provide a better function approximation which can enable it predict similar problem (Aggarwal, 2015). The regression value for the training process must therefore be closer to unity (≈1) for a better prediction. For this study, the higher regression coefficients obtained between the network outputs is an indication of good training in the supervised neural learning process neural network. Moreover, the root mean square error obtained from the training process went through a marginal reduction as the number of iterations was increased. The smaller MSE and RMSE values further validated a good training and better prediction (Thakker et al., 2016).

Close

Fig. 3

Post regression analysis with Gradient Descent algorithm (a) average yield black pepper (b) average energy black pepper (c) average yield white pepper (d) average energy white pepper.

Export to PPT

3.4.2

3.4.2 ANN prediction using Levenberg–Marquardt training algorithm

Table 6 is the results obtained from ANN prediction per input variable using Levenberg–Marquardt training algorithm. The overall regression coefficient and root mean squared error of 0.9551 and 0.5701 respectively were generated from optimum yield in black pepper heat refluxation. While an overall regression coefficient and root mean squared error of 0.9845 and 0.5708 respectively for optimum yield in white pepper heat refluxation. Similarly the overall regression coefficient and root mean squared error was 0.9398 and 0.7258 for optimal energy absorbed in black pepper heat refluxation. Whereas for the optimum yield in white pepper heat refluxation, the overall regression coefficient and root mean squared error was 0.9546 and 0.4525 respectively. The summary of the overall regression coefficient and root mean squared error for each response in black and white pepper heat refluxation. Table 6 consists of results obtained from the Levenberg–Marquardt algorithm.

Table 6 Network training configuration and results (Levenberg-Marquardt).

Sample name	Output	Transfer function	Model topology	MSE	RMSE	R2
Black pepper	yb	tansig	3-4-5-1	0.3250	0.5701	0.9551
	Qtb	tansig	3-4-5-1	0.5268	0.7258	0.9398
White pepper	yw	tansig	3-4-5-1	0.3258	0.5708	0.9845
	Qtw	tansig	3-4-5-1	0.2048	0.4525	0.9546

Fig. 4 is the post regression analysis with a comparison between the predicted experimental results and the expected. The results obtained gave regression values closer to unity (≈1) and lower root mean square errors. The proximity of the regression coefficient and lower root mean square errors showed good training and a better prediction.

Close

Fig. 4

Post Regression analysis with Levenberg Marquardt algorithm (a) average yield black pepper (b) average energy black pepper (c) average yield white pepper (d) average energy white pepper.

Export to PPT

3.5 Chemical composition

The relative abundance of each classes of compound identified from GC–MS analysis of the black (BPOE) and white (WPOE) extracts are presented in Fig. 5. A total of 19 components were identified in GC–MS analysis of BPOE and the major components were piperine (37.85%), caryophyllene (8.97%), sabinene (5.76%), limonene (4.51%), and β-bisabolene (1.66%). Moreover, the extracted oleoresin consisted of 39.38% alkaloids, 11.06% monoterpenes hydrocarbons, 2.23% oxygenated monoterpenes and 13.19% sesquiterpenes. These suggested that the alkaloids have the highest concentration of bioactive compounds, followed by monoterpenes and sesquiterpenes. Moreover, α-Thujene, guineensine, 1-cinnamoyl piperidine, piperettine, α-selinene, α-cubebene were found to be in trace amount as reported by Singh et al. (2013). However, in WPOE a total of 25 bioactive compounds were identified with piperine (42.29%), caryophyllene (10.36%), sabinene (7.94%), limonene (6.47%), and β-bisabolene (2.42%) as major components. The white pepper extract contained 44.36% of alkaloids, 16.21% of monoterpenes hydrocarbons, 3.97% of oxygenated monoterpenes, and 14.03% of sesquiterpenes. The higher concentration of compounds such as piperine, caryophyllene, sabinene, limonene and β-bisabolene in white pepper oleoresin extracts (WPOE) is therefore an indication of its higher nutritional and functional characteristics than black pepper oleoresin extracts (BPOE) as illustrated in Fig. 5.

Close

Fig. 5

Percentages of different classes of compounds extracted from black and white pepper.

Export to PPT

3.6 Morphological characteristics

The medullary and peripheral bundles are the two striking features in the structure of black and white pepper (Ravindran, 2000). It composed of layer of stone like elongated parenchymatous cells which are neatly packed in a fresh sample as shown in Fig. 6 (a) and (c). The cellulosic cell wall of an untreated fresh pepper sample first experienced an induction process in which the outer vascular bundles swell due to the penetration effect of solvent. As a result of the localized heating effects on the pepper matrix, the membrane structure of the cellulose cell wall becomes disoriented thereby resulting in a subsequent rupture. The elongated stone like cells then become broken into smaller forms which lead to a reduced surface area. Fig. 6 (a) and (c) the ordered thin-wall stone-like elongated cells of fresh black and white pepper, respectively. Fig. 6 (b) and (d) are the ruptured cells after exposure of white pepper to the conventional localized heat.

Close

Fig. 6

SEM-monograph (a) Black P.nigrum at pre-extraction (b) Black P.nigrum at post-extraction (c) White P.nigrum at pre-extraction (d) White P.nigrum at post-extractionheating.

Export to PPT