Agglomeration of turmeric powder and its effect on physico-chemical and microstructural characteristics

Journal of Food Engineering 120 (2014) 124–134

Contents lists available at ScienceDirect

Journal of Food Engineering journal homepage: www.elsevier.com/locate/jfoodeng

Agglomeration of turmeric powder and its effect on physico-chemical and microstructural characteristics K. Dhanalakshmi, Suvendu Bhattacharya ⇑ Food Engineering Department, Central Food Technological Research Institute, Council of Scientific and Industrial Research, Mysore 570020, India

a r t i c l e

i n f o

Article history: Received 26 April 2013 Received in revised form 15 July 2013 Accepted 19 July 2013 Available online 31 July 2013 Keywords: Agglomeration Turmeric powder Particle size distribution Image analysis Artificial neural network

a b s t r a c t Agglomerated foods have gained attention in recent years due to their convenience in use. Turmeric powder has been subjected to agglomeration process at different moisture contents (10–28%) and steaming times (0–60 min). Experimental cumulative particle size distribution data of agglomerated samples can be predicted well (0.951 6 r 6 0.999, p 6 0.01) with Rosin–Rammler–Bennett model. The functional properties related to hydration characteristics like wetting time (10–35 s) and sinking time (15–115 s) of agglomerated samples decrease with an increase in moisture content and/or steaming time. Microstructural observation shows that the non-agglomerated sample possesses spheroids and ellipsoids of different sizes. The size of agglomerates ranges between 50 and 160 lm; their shape varies from spheroid to elongated ellipsoids. Image analysis infers that the size related parameters increase with an increase in moisture content/steaming time. A four-layered artificial neural network having a structure of 2–10-8– 4 has been developed to predict the agglomeration process of turmeric powder. Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction Powdery food materials are frequently used for convenience in applications during transportation, handling, processing and for product formulations (Ghosal et al., 2010). A variety of food powders from different sources are used to serve specific purposes including improving sensory appeal and nutritional status of finished products. Occasionally, there is a need to modify the structure of food to achieve certain specific characteristics and convenience in use. Agglomeration is a physical phenomenon and can be described as the sticking of particulate solids, which is caused by short-range physical or chemical forces among the particles (Barbosa-Canovas et al., 2005). This phenomenon is triggered by specific processing conditions, or binders and substances those adhering chemically or physically to form bridges between particles (Pietsch, 2003). The main purpose of particle size enlargement by agglomeration is to regulate certain physical properties of food powders such as density, flowability, to improve dispersion and dissolution characteristics, and reduce the tendency of caking and dust formation (Mukherjee and Bhattacharya, 2006). Spices are widely used in food products to create the distinctive flavor and character that are representative of different cuisines. The delightful flavor and pungency of spices make them indispensable in the preparation of palatable dishes. In addition, they are used in the preparation of a number of pharmaceutical ⇑ Corresponding author. Tel.: +91 821 2513910; fax: +91 821 2517233. E-mail address: [email protected] (S. Bhattacharya). 0260-8774/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jfoodeng.2013.07.024

formulations (Peter, 2004). Turmeric, a commercially important spice, is mainly consumed as dry powder primarily for coloring because of its attractive yellow color and its associated therapeutic properties. It also imparts the characteristic flavor and preserves the freshness of the product prepared (Govindarajan and Stahl, 1980). The main biological activities of turmeric rhizome reported are anti-inflammatory, anti-microbial, anti-tumor and wound healing (Jayaprakasha et al., 2005). Turmeric powder is usually stored in bulk in opaque containers in which moisture absorption and light exposure are avoided. The aroma of turmeric is contributed by its volatile oil (e.g., terpeniods and aromatic compounds), while the color is attributed to the presence of diaryl heptanoids viz., curcuminoids (Govindarajan and Stahl, 1980). However, turmeric powder is not a free flowing sample. It also sticks to utensils when used for transferring during food preparations and forms lumps when added in large quantity in institutional cooking. The microencapsulated turmeric oleoresin powder can be obtained by spray drying of oleoresin using edible gum as a matrix (Kshirsagar et al., 2009). One of the demerits of spray-dried powders is their small particle size, typically in the range 10–100 lm in diameter. This small particle size may result in poor reconstitution properties, product separation during shipping and handling (when mixed with other ingredients), poor handling properties (e.g., flow and quantification), and dusting problems during manufacturing (Buffo et al., 2002). To overcome these problems, turmeric powder can be agglomerated directly. This will be an alternate process compared to the existing process of extraction of oleoresin, encapsulation followed by agglomeration. The agglomeration process

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134

125

Nomenclature a0, a1, a2, a11, a22, a12 model coefficients for the variables a⁄ redness (–) b⁄ yellowness (–) dpi mass mean mesh size (m) f(p) threshold (sigmoid) function of the independent input variable p (–) Fmax, Fmin Feret diameters (m) F(x) cumulative under size mass fraction (–) k consistency index (Pa sn) ⁄ L lightness (–) n flow behavior index (–) N number of data sets nR uniformity index of the particle size distribution (–) P perimeter (m) R correlation coefficient (–) S surface area (m2)

involves the use of binders (e.g., moisture), applying pressure and increasing temperature (Palzer, 2011). Therefore, scope exists to monitor the physical and physico-chemical properties for which scientific data on agglomeration of turmeric powder are scarce. Earlier studies have been conducted on the characterization of turmeric powder with respect to its prime components like starch, volatile and nonvolatile components (Dhanalakshmi et al., 2011; Dhanalakshmi and Jaganmohan Rao, 2012). The results indicate that cured-dried turmeric powder exhibits a high yield of volatile and nonvolatile components (Dhanalakshmi and Jaganmohan Rao, 2012), which are the principal components of importance. Therefore, the objectives of the present study are to (a) conduct agglomeration of turmeric powder using water as a binder at different moisture contents and steaming times, (b) determine the physical, physico-chemical and functional characteristics, and morphological changes, and (c) development of an artificial neural network to predict the agglomeration process. 2. Materials and methods 2.1. Materials Fresh turmeric (Curcuma longa) rhizomes were procured from a local cultivator near Mysore, Karnataka, India. The rhizomes were manually cleaned with water to remove the adhering soil and extraneous matter (Dhanalakshmi et al., 2011). The fresh rhizomes were cured by employing the conventional method of cooking in excess of boiling water for 1 h; excess water was discarded and the rhizomes were dried in the shade for a week. These dried rhizomes were powdered in a pulverizer to obtain fine powder; temperature during grinding was less than 40 °C. This powder sample is referred as ‘cured-dried’ sample; the process was repeated twice. 2.2. Methods 2.2.1. Agglomeration of turmeric powder and experimental design Turmeric powder (6% moisture content) was mixed in a Hobart mixer (Model #1/BSP-BM7, Bakery Mixer, Malaysia) at room temperature (about 25 °C); water was sprayed from a spray gun onto turmeric powder to obtain samples with moisture contents of about 10%, 15%, 18%, 22%, 25% and 28%. The volume of powder sample was about 2.5 L and spray time was between 3 and 5 min. The materials were mixed at the lowest speed of rotation

x mesh size (m) xi mass fraction (–) xR size parameter (m) X1, X2 independent variables Y response function WExpt, WPred experimental and predicted values of target parameters Greek letters c_ shear-rate (s1) 4E total color difference (–) e random error of the regression model r shear-stress (Pa) v2 variance

for 30 min to ensure homogeneous product; moisture content was rechecked after this mixing step. Latter, the samples with different moisture contents were divided into seven batches. The first batch sample, after addition of moisture (without steaming), was subjected to granulation in a laboratory model granulator (Model # CMJ-8, Cadmach Machinery, Ahmedabad, India) and dried in a tray dryer at 50 °C for 8 h. The other batches were steamed in a pressure cooker at the ambient pressure for different time intervals (10–60 min). After steaming, these samples were granulated in the laboratory model granulator as mentioned earlier. These granules were also dried in the same tray dryer at 50 °C for 8 h. These samples were stored in double walled polyethylene bags at room temperature for further analysis within 24 h of sample preparation. The agglomeration process was repeated twice. 2.2.2. Physico-chemical characterization The non-agglomerated and agglomerated samples were used for the determination of mass mean particle size and particle size distribution employing the method of sieve analysis (McCabe et al., 2005). A set of standard sieves with apertures of 800, 710, 500, 355, 250, 150, 105 and 53 lm were stacked one upon the other in an ascending order of the aperture size. The sample to be tested was placed on the top sieve with aperture of 800 lm, and the sieving was performed for 20 min. The particles retained on each sieve were removed and weighed to calculate the mass fractions of particles. Particles that passed through the entire sieve sets were collected in the bottom pan of the stack. The mass mean mesh size (dpi) was calculated as the mass mean of two consecutive sieves used for analysis. Mass fraction (xi) was determined as the ratio of mass of sample retained on each sieve divided by the total mass. The mass mean particle/granule diameter was calculated based on the mass fraction using Eq. (1) (McCabe et al., 2005).

Mass mean diameter ¼

n X

xi dpi

ð1Þ

i¼1

Here, n was the number of sieves including a receiver (sieve i = 1 with an aperture of 0 lm). The particle size distribution of non-agglomerated turmeric powder sample (control) was determined. No sample was retained on sieves with apertures of 800, 710, 500, 355 and 250 lm. The maximum amount of particle was retained on the sieve with apertures of 105 lm. The mass of particles passing through the sieve with aperture 150 lm (particles with size less than the mass mean

126

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134

size of 127.5 lm) was determined in the agglomerated samples, and was reported as the extent of fines. Particle size distribution (PSD) data were fitted to Rosin– Rammler–Bennett (RRB) model (Macias-Garcia et al., 2004; Rosin and Rammler, 1933) (Eq. (2)).

 nR



FðxÞ ¼ 1  exp

x xR

ð2Þ

Here, x was mesh size (lm), F(x) was cumulative under size mass fraction; xR and nR were the function parameters known as the size parameter and uniformity index of the particle size distribution, respectively. The solver tool available in Microsoft Excel software (version 2003) was used to find the values of xR and nR, by employing the non-linear optimization method by minimizing the residual sum of squares (RSSs). Bulk density was determined by pouring the powder sample in a graduated measuring cylinder to a known volume (250 mL). The mass of the powder was determined, and bulk density was calculated by dividing the mass with corresponding volume. Triplicate measurements were performed and were reported as the untapped or loose bulk density. Wetting and sinking times were determined by glass sliding method as indicated by International Dairy Federation for milk powder (IDF, 1979) with marginal modification (Dhanalakshmi and Bhattacharya, 2012). The time required for the sample to wet completely was reported as wetting time. In the same experiment, the time required for sinking of the sample to the bottom of the beaker was noted, and reported as sinking time. The experiments were repeated thrice and results were reported as mean ± standard deviation (SD). 2.2.3. Rheological behavior A controlled stress rheometer with coaxial cylinder (Z41) attachment (Model # RS6000 Haake RheoWin, Thermo Scientific, Karlsruhe, Germany) was used to determine the flow behavior of the dispersions containing agglomerated turmeric powder samples. All rheological measurements were conducted at 25 °C on triplicate samples by employing a circulating water bath for control of temperature by ±0.1 °C (Bhattacharya, 1999). Dispersions containing 10% (w/w) turmeric powder (dry solid basis) were prepared and analyzed for their flow characterization (Dhanalakshmi et al., 2011). These samples were initially pre-sheared at a shearrate of 10 s1 for 30 s followed by a gradual increase of shear-rate up to 500 s1 in 60 s. Apparent viscosity was taken as the ratio of shear-stress and shear-rate when the latter was taken as 100 s1. Yield stress was determined by shearing at 5 s1 for 30 s followed by stopping the spindle, and allowing the sample to relax for 120 s. However, no detectable yield stress was observed for the samples. Hence, shear-stress and shear-rate data were fitted to the commonly employed power law model (Eq. (3)), and model parameters (k and n) were calculated by using the software provided by the instrument manufacturer. In the power law model (Eq. (3)), r was the shear-stress (Pa), c_ was the shear-rate (s1), k was the consistency index (Pa sn) and n was the flow behavior index (dimensionless). The suitability of the power law model was examined by determining the variance (v2) and correlation coefficient (r); the significance of r-values was judged at p 6 0.01.

r ¼ kðc_ n Þ

ð3Þ

2.2.4. Color measurement The color of non-agglomerated and agglomerated turmeric powder samples was determined by employing a colorimeter (Model # LABSCAN XE, Hunter Associate Laboratory, Virginia, USA). The illuminant employed was C (average daylight) and view angle was 10°; the visible range of 400–700 nm was employed and the diameter of the measuring port was 10 mm. The color

parameters like L⁄ (lightness), a⁄ and b⁄ values were determined as per the Commission Internationale de’Eclairage (CIE) Lab method (Hutchings, 1994). The total color difference (DE) was also calculated keeping the standard glazed white plate (supplied by the equipment manufacturer) as the reference for comparison. The DE was given by the following equation:

DE ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   2 2 ðLo  L Þ þ ðao  a Þ2 þ ðbo  b Þ

ð4Þ



Here, Lo , ao and bo were the values of standard white, while L⁄, a⁄ and b⁄ were the corresponding values for the sample under examination. All color measurements were conducted on triplicate samples. 2.2.5. Microstructure The microstructure of non-agglomerated and agglomerated turmeric samples was examined by employing a scanning electron microscope (SEM) (Model#435VP, Leo Electron Microscopy, Cambridge, UK). The samples were dried in a hot air oven, maintained at a temperature of 50 °C for 5 h. The dried samples were mounted on metal stubs with double adhesive conducting tapes, and were coated with a thin film of gold employing a sputter coater. Microscopic examination was conducted at an accelerating voltage of 15 kV at magnifications of 500 and 1000 to observe the morphology, shape and size of the samples. Representative photomicrographs are presented here. 2.2.6. Imaging and image analysis The photomicrographs, obtained in the SEM studies, were analyzed using the image analysis software (IMAGEJ1.45s, National Institutes of Health, Maryland, USA). The two dimensional basic parameters like major and minor axes (X and Y), surface area (S), Feret diameters (Fmax and Fmin) and perimeter (P) were obtained by employing this software. The shape parameters like roundness, equivalent diameter, elongation and compactness were calculated using Eqs. (5)–(8) (Russ, 2011). The average values of 10 measurements are reported here.

Roundness ¼

4S

Equivalent diameter ¼

Elongation ¼

ð5Þ

pF 2max

F max F min

Compactness ¼

qffiffiffiffiffiffiffiffiffi 4 p S F max

rffiffiffiffiffiffi 4S

p

ð6Þ

ð7Þ

ð8Þ

2.2.7. Development of artificial neural network (ANN) The concept of ANN is being increasing applied in different food processing operations and product development (Bhattacharya and Patel, 2007; Hernandez et al., 2008; Singh et al., 2009). The physical and physico-chemical measurements provided 72 data sets (shown in Table 1 as mean ± SD of 24 experimental sets), which were divided randomly into 21 and 3 sets, for training the network and for verification of the trained network, respectively. The ANN software, developed earlier by using Turbo C++, based on analogies to biological neurons and tested with published data (Bhattacharya and Patel, 2007), was employed; the methodology of back propagation network and the normalization (between 0 and 1) of data were used. The sigmoid function represented by Eq. (9) was used as the transfer function to convert summed values of input nodes to node values for the next layer (Jensen, 1994).

127

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134 Table 1 Experimental values used for training of artificial neural network (ANN) and validation. Experiment No.

Control 1 2 3 4 5 6 7* 8 9 10 11 12 13 14* 15 16 17 18 19 20 21* 22 23 24 *

Input variable

Output function

Moisture content of sample (%)

Steaming time (min)

Extent of fines (%)

Bulk density (kg m3)

Wetting time (s)

Sinking time (s)

6 10 10 10 10 15 15 15 15 18 18 18 18 22 22 22 22 25 25 25 25 28 28 28 28

0 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60

97.6 ± 4.1 35.2 ± 1.6 83.4 ± 4.1 78.5 ± 3.1 81.0 ± 3.8 80.9 ± 3.2 81.6 ± 2.5 77.1 ± 3.2 77.1 ± 2.6 73.2 ± 3.4 74.2 ± 2.9 76.0 ± 2.1 74.6 ± 1.8 71.5 ± 1.5 71.3 ± 3.2 64.8 ± 2.5 67.9 ± 1.8 71.4 ± 1.9 67.0 ± 2.6 45.6 ± 1.2 34.4 ± 0.9 68.3 ± 2.1 39.6 ± 1.1 15.3 ± 0.7 11.0 ± 1.8

542.0 ± 3.1 535.1 ± 2.7 522.4 ± 3.9 474.7 ± 2.1 469.6 ± 2.5 519.2 ± 2.8 491.9 ± 2.5 476.7 ± 1.9 450.8 ± 2.3 497.8 ± 1.2 476.5 ± 3.2 473.2 ± 1.8 458.5 ± 4.1 480.5 ± 1.4 470.4 ± 1.2 456.6 ± 4.1 447.0 ± 2.7 468.2 ± 1.3 439.9 ± 1.2 432.9 ± 4.1 419.2 ± 2.7 433.1 ± 2.4 427.7 ± 2.3 398.4 ± 2.9 381.0 ± 4.9

60.0 ± 3.0 35.0 ± 1.8 32.0 ± 1.6 30.0 ± 1.4 30.0 ± 1.2 31.0 ± 1.3 28.0 ± 1.2 27.0 ± 1.8 26.0 ± 2.0 29.0 ± 0.9 27.0 ± 1.1 25.0 ± 1.9 24.0 ± 2.1 28.0 ± 1.5 24.0 ± 0.6 22.0 ± 1.2 22.0 ± 1.6 25.0 ± 1.3 22.0 ± 0.6 18.0 ± 0.8 19.0 ± 1.8 23.0 ± 1.1 18.0 ± 2.3 15.0 ± 2.1 10.0 ± 1.9

90.0 ± 4.5 115.0 ± 4.3 108.0 ± 1.2 96.0 ± 4.8 90.0 ± 4.2 95.0 ± 2.7 86.0 ± 3.1 83.0 ± 3.9 80.0 ± 4.4 88.0 ± 2.9 75.0 ± 3.7 73.0 ± 3.3 65.0 ± 4.4 80.0 ± 4.1 68.0 ± 4.5 58.0 ± 4.0 55.0 ± 3.4 78.0 ± 2.7 65.0 ± 1.8 44.0 ± 2.6 40.0 ± 3.2 63.0 ± 3.7 32.0 ± 1.2 28.0 ± 1.3 15.0 ± 0.8

Indicates data sets used for validation of the developed network.

f ðpÞ ¼

1 : 1 þ ep

ð9Þ

Here, f(p) was the threshold (sigmoid) function of the independent input variable p. The object oriented programming (OOP) consisted of a collection of simple elements (neurons), placed in different hidden layer(s). The ANN program was modeled to receive the input parameters (moisture content of feed and steaming time), and yielded the desired output parameters (extent of fines, bulk density, and wetting and sinking times) through a process of training or learning. The training process (Baughmen and Liu, 1995) consisted of preliminary trials using all the randomly selected 21 data sets (Table 1) with 0.0001 as the error tolerance, 0.5 as learning rate and 5,000 iterations to obtain the error on network predictions by varying the number of hidden layers (1, 2, and 3) and number of neurons (4–10) in each of the hidden layers (Bizot, 1983). The number of hidden layers (1–3) followed by the total number of neurons in the hidden layers was tentatively decided based on the error values. The number of neurons in each hidden layer was then varied along with the variation in learning rate (0.01– 1.0) and the number of iterations (up to 50,000) to have the optimum network structure with a minimum error. The optimized network was trained to have the final weights (w), which were stored for the next operation. Finally, the network was validated for remaining three unused experimental data input sets (Table 1) and compared with the experimental results of the targeted parameters. The root mean square (RMS) error (Eq. (10)) was computed to know the error of the developed ANN system.

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u N uX  2 RMS errorð%Þ ¼ 100t W Expt  W Pred =W Expt =N

ð10Þ

n¼1

Here, WExpt and WPred were the experimental and predicted values of target parameters, respectively, and N was the number of data sets (n = 1, 2, 3, . . . , N).

2.2.8. Statistical analysis All measurements were carried out in triplicates (unless specifically mentioned), and data were expressed as arithmetic mean ± standard deviations (SD). Duncan’s Multiple Range Test (DMRT) was applied to determine the existence of significant difference at p 6 0.05. Multiple regression analysis (quadratic polynomial model, Eq. (11)) was employed to obtain 3-D wire-mesh plot for easy visualization of the effect of independent variables on the response functions (Little and Hills, 1978).

Y ¼ a0 þ a1 X 1 þ a2 X 2 þ a11 X 21 þ a22 X 22 þ a12 X 1 X 2 þ e:

ð11Þ

Here, Y was the response function, X1 and X2 were independent variables, a0, a1, a2, a11, a22 and a12 represented the model coefficients for the variables, and e was the random error of the regression model. 3. Results and discussion 3.1. Physical characterization Turmeric powder has been agglomerated by adding different levels of moisture as binder and steaming for different time intervals. The effectiveness of the agglomeration process is reported in terms of the extent of fines present in the sample, and the mass mean particle diameter of the agglomerates as a function of moisture content and steaming time (Fig. 1). The extent of fines is between 8% and 87%; an increase in moisture content usually decreases the extent of fines. A complex curvilinear trend has been observed for the extent of fines (Fig. 1A). At low moisture content, the extent of fines increases with steaming time and reaches the maximum value; at higher moisture contents like 25–28%, the extent of fines decreases markedly with an increase in steaming time. The desirable lowest extent of fines is obtained when the steaming time and moisture content are at their highest levels. The mass mean particle diameter increases from 194 to 365 lm with an increase in moisture content, but steaming time offers a marginal effect (Fig. 1B). The extent of fines and the mass mean particle

128

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134

Fig. 1. Effect of moisture content of feed and steaming time on: (A) extent of fines, and (B) mass mean particle diameter of agglomerates.

15%

18%

22%

25%

28%

Mass fraction (%)

10%

Particle size (µm) Fig. 2. Effect of moisture content of feed (10%, 15%, 18%, 22%, 25% and 28%) and steaming time on the particle size distribution of agglomerated turmeric samples.

diameter depend on two opposing phenomena. Moistening followed by steaming allows the starch rich powder to undergo the phenomenon of gelatinization to offer a sticky outer surface; hence, binding of particles improves and particle size increases with a simultaneous decrease in the extent of fines. However, during the post-agglomeration drying step, the weakly bonded particles undergo the process of attrition (breakage of already formed agglomerates) such that the fines are again formed. The particle size distribution (PSD) curves of agglomerated turmeric samples are shown in Fig. 2. The PSD curve of

non-agglomerated (control) sample shows a single mode (indicating the most commonly occurring particle diameter) at about 200 lm. The agglomerated sample with 10% moisture content of feed also shows the single mode (Fig. 2). However, samples with moisture content of feed between 15% and 25% show the PSD curve with bimodal characteristics. It means that the progressive growth of particles favors bimodal characteristics. An increase in steaming time up to 40 min offers insignificant change (p 6 0.05) in the pattern of the curves. However, above 40 min of steaming, the particle size increases significantly particularly in samples having moisture

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134

129

Table 2 Rosin–Rammler-Bennet (RRB) model parameters for particle size distribution of agglomerated turmeric samples. Moisture content of sample (%)

Steaming time (min)

RRB model parameters Size parameter (xR) (lm)

Distribution parameter (nR) (–)

6 10

– 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60

195 242 201 373 332 228 227 213 235 240 229 232 264 265 236 249 235 240 269 304 333 338 345 346 410

4.6 6.2 1.9 7.4 3.0 5.5 6.0 1.4 2.9 3.2 5.9 1.7 2.1 2.5 4.9 8.0 4.2 6.9 2.9 3.1 2.9 2.7 2.7 3.4 4.7

15

18

22

25

28

a

Correlation coefficienta (r) (–)

0.995 0.997 0.987 0.999 0.997 0.994 0.996 0.960 0.980 0.990 0.993 0.973 0.984 0.951 0.991 0.996 0.983 0.970 0.974 0.992 0.996 0.996 0.994 0.990 0.992

Significant at p 6 0.01.

Fig. 3. Particle size distribution data of agglomerated turmeric sample with 28% moisture content at different steaming times fitted to Rosin–Rammler–Bennet (RRB) model.

content between 10% and 15%. It may thus be inferred that moisture content of the feed plays a major role on particle size increase of turmeric samples during the agglomeration process, while only a marginal effect is exhibited by steaming time. In case of corn starch as a model system and binders like water and pre-gelatinized starch, the particle size has been reported to increase with an increase in the concentration of binders (Dhanalakshmi and Bhattacharya, 2012). Rosin–Rammler–Bennett (RRB) model (Eq. (2)) parameters such as size parameter (xR) and uniformity index (nR) have been computed (Table 2). The high correlation coefficient (r = 0.951–0.999, p 6 0.01) indicates the suitability of RRB model (Fig. 3). The size parameter (xR) for non-agglomerated sample is 195 lm, while it

Fig. 4. Effect of moisture content of feed and steaming time on physico-chemical characteristics like: (A) bulk density, (B) wetting time, and (C) sinking time of agglomerated turmeric samples.

varies between 201 and 410 lm for agglomerated products; it increases with an increase in moisture content of feed and the change is marginal in case of steaming time. A low value of nR indicates scattered distribution, while higher values mean steeper curves with an increasingly uniform particle size distribution (Prasher, 1987). For non-agglomerated sample, the value of nR is 4.6, while it varies between 1.4 and 8.0 for agglomerated samples.

130

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134

Table 3 Rheological parameters of agglomerated turmeric powder dispersions containing 10% solid. Moisture content of feed (%)

Steaming time (min)

Apparent viscosity* (mPa s)

Power law model parameters Flow behavior Index (n) (–)

Consistency index (k) (mPa sn)

10

0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60

3.32 ± 0.04a 4.21 ± 0.65b 6.05 ± 0.51c 2.69 ± 0.45d 5.09 ± 0.41e 2.74 ± 0.42f 2.63 ± 0.04d 3.08 ± 0.08a 3.30 ± 0.19a 2.28 ± 0.08b 2.18 ± 0.07c 2.47 ± 0.38ac 2.22 ± 0.91d 2.30 ± 0.04e 5.31 ± 0.17a 3.80 ± 0.04b 2.35 ± 0.24c 3.62 ± 0.95bde 2.57 ± 1.02e 2.70 ± 0.39c 3.92 ± 0.11d

1.06 ± 0.02a 0.98 ± 0.06b 1.03 ± 0.01b 0.83 ± 0.04c 0.97 ± 0.02d 0.93 ± 0.03bd 1.12 ± 0.07e 1.05 ± 0.01a 1.01 ± 0.02a 1.14 ± 0.03b 0.97 ± 0.07ac 1.02 ± 0.03c 0.93 ± 0.02d 1.14 ± 0.01b 0.98 ± 0.02a 1.01 ± 0.03b 1.22 ± 0.14c 1.00 ± 0.06d 1.22 ± 0.37e 1.15 ± 0.05f 1.09 ± 0.04g

2.39 ± 0.28a 4.63 ± 1.95b 5.08 ± 0.55c 11.10 ± 5.88d 5.70 ± 1.10e 4.23 ± 1.46f 1.34 ± 0.569e 2.30 ± 0.034a 3.00 ± 0.15b 1.00 ± 0.20a 2.90 ± 0.10c 1.50 ± 0.20c 2.16 ± 0.31b 1.10 ± 0.04d 5.50 ± 0.46a 3.57 ± 0.60b 8.32 ± 0.66c 3.23 ± 1.31d 1.99 ± 0.26cde 1.00 ± 0.34f 2.52 ± 0.58be

15

28

Variance (–)

Correlation coefficient (r) (–)

0.02 0.04 0.04 0.74 0.03 0.32 0.03 0.02 0.03 0.05 0.16 0.08 0.35 0.04 0.02 0.01 0.03 0.01 0.11 0.03 0.05

0.99 0.99 0.99 0.97 0.99 0.97 0.99 0.99 0.99 0.99 0.97 0.99 0.98 0.98 0.99 0.99 0.98 0.99 0.99 0.97 0.99

Data in the same column with different superscripts differ significantly at p 6 0.05 according to DMRT. reported at a shear-rate of 100 s1.

*

However, no clear trend has been identified for nR with moisture content of feed and/or steaming time. 3.2. Physico-chemical and functional characteristics The effect of moisture content and steaming time on the physico-chemical characteristics of agglomerated turmeric powder (Table 1) shows that an increase in moisture content from 10% to 28%, and steaming time from 0 to 60 min decrease the bulk (loose) density (Fig. 4); it is in the range of 396 and 530 kgm3 for agglomerated samples (Fig. 4A) while it is 542 kgm3 for non-agglomerated sample. The moistening and steaming process enhances the adhesion characteristics to form porous and irregular shaped agglomerates (discussed latter) causing the bulk density to decrease. The functional properties related to hydration characteristics like wetting time (10–35 s) and sinking time (15–115 s) of the agglomerated samples decrease significantly with an increase in moisture content and/or steaming time (Fig. 4B and C). This may be attributed to the decrease in bulk density in addition to increased surface area of agglomerated samples due to porous structure. As low wetting and sinking times are the desired features for any convenience or instant foods, the usefulness of agglomeration process for turmeric powder is justified.

The flow behavior index varies between 0.93 and 1.22 indicating these samples to be shear-thinning, Newtonian and shearthickening for conditions of n values to be <1, equal to 1 and >1, respectively. The isolated turmeric starch dispersions have been identified as shear-thickening, while cured-dried samples behave as shear-thinning (Dhanalakshmi et al., 2011). The power law equation is suitable to correlate shear-stress and shear-rate as the correlation coefficients (r) are high (P0.97) and the variance values are low (0.01–0.35). It is thus inferred that agglomeration of turmeric powder is a complex process and the rheological parameters of the dispersions do not show any clear trend with moisture content and steaming time. 3.4. Color parameters Color of non-agglomerated and agglomerated turmeric samples has been reported in terms of L⁄, a⁄, b⁄ and the total color difference (individual data not presented). The non-agglomerated as well as agglomerated turmeric powder samples have a bright orange–yellow color as a⁄ and b⁄ values are high, and b⁄  a⁄. The lightness

3.3. Flow behavior The flow characteristics of agglomerated turmeric powder dispersions prepared with different moisture contents and steaming times are shown in Table 3. The apparent viscosity (reported at a shear-rate 100 s1) and consistency index do not show any clear trend with moisture content and steaming time. The apparent viscosity of non-cured turmeric powder is 4.15 mPa s, while it is 7.57 mPa s for cured-dried samples (Dhanalakshmi et al., 2011). However, in the present study, the apparent viscosity values (2.18–6.05 mPa s) are lower than these two samples. It is possible that the steaming process for wet powder may induce changes such that the non-starchy components (protein, crude fiber, fat and minerals) present in turmeric powder binds with starch that also affect the flow behavior. The settling of particles during measurement may also be responsible.

Fig. 5. Microstructure of cured-dried turmeric powder.

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134

(L⁄) is an indication of the extent of light reflected compared to the incident light. A maximum of 10.6% decrease in L⁄ values (significant at p 6 0.05) has been observed due to agglomeration. The a⁄ values (an indication of the redness of sample) show a decrease up to 9.2% while b⁄ (an indication of the yellowness of sample) decreases to a maximum of 11.7%. Hence, there is a marginal increase in the total color difference (DE) to a maximum of 4.4%. However, the changes in these color parameters are significant (at p 6 0.05) indicating a minor change in the color of the samples occurs due to agglomeration. Among the color parameters, the lightness (L⁄) and yellowness (b⁄) are the most affected items but their decrease is limited to a maximum of 12%. The possible reason for the decrease in color parameters is the presence of light-sensitive curcuminoid pigments (responsible for the characteristic yellow color) in turmeric powder (Govindarajan and Stahl, 1980).

131

3.5. Microstructural features The microstructure of raw and agglomerated turmeric samples has been focused to know the size, shape and alignment of particles due to the application of agglomeration process. The surface morphology and microstructure of the dried granulated particles influence the physical and functional characteristics of the powder. Agglomeration process produces granules resulting from many individual particles, which appear to be bonded together by binders like water (steam) and gelatinized starch. The control sample (non-agglomerated turmeric powder) shows particles having different sizes and shapes (Fig. 5). The maximum size (dimension) of particle is between 50 and 160 lm while shape varies from spheroid to elongated ellipsoids. Further, the surface of particles appears to be rough possibly due to the curing process that induces starch gelatinization and subsequent retrogradation in addition to

Fig. 6. Photomicrograph (left column) and processed images (right column at same magnification) of turmeric powder agglomerated with 10% moisture content (A and B), and agglomerated with 18%, 22% and 28% moisture followed by 10 min steaming (C and D, E and F, and G and H). Scale bar indicates 20 lm for all these 8 figures.

132

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134

ched out amylose during the gelatinization process possibly acts as the binding material. The role of amylose here is important as turmeric starch contains about 48% amylose (Dhanalakshmi et al., 2011). 3.6. Image analysis

Fig. 7. Magnified view (2000) of agglomerated turmeric powder as shown in Fig 6G.

damages caused during the size reduction process. The use of water alone at 18% level produces only a few agglomerated particles of small sizes (Fig. 6A) while higher moisture contents (22% and 28%) followed by steaming (10 min) favor agglomeration process, and progressive increase in size of the agglomerates has been noticed (Figs. 6C, E and G). The size of such agglomerates can be as high as 300 lm as they are composed of many particles. However, a hollow/porous structure is formed that favors easy passage for water migration during subsequent hydration. The magnified view of the photomicrograph Fig. 6G i.e. Fig. 7 shows a cavity, which appears to be created by several particles joined by a binder; the lea-

The non-agglomerated and agglomerated turmeric powder samples have been subjected to image analysis to analyze their structural features (Fig. 6 B, D, F and H). These processed images have been used to obtain the size related parameters (such as surface area, perimeter, equivalent diameter, roundness, elongation and compactness; Table 4). The surface area increases with moisture content and steaming time, which is prominent for feed containing higher moisture content like 28%. This trend is also true for perimeter and equivalent diameter (Table 4). The roundness for unprocessed (control) sample is lowest (0.64), while the agglomerated samples show higher values (0.66–0.86). It means that elongated raw powder tends to form spherical shape due to agglomeration. The elongation is the ratio of the largest to that of the smallest dimension of the same particle or agglomerate; it is 1.57 for control sample, and 1.20–1.59 for agglomerated products. However, no clear trend has been noticed on the effect of moisture content and steaming time on the shape parameters like roundness, compactness and elongation. 3.7. Artificial neural network and validation The experimental values of physico-chemical parameters for different combinations of moisture content and steaming time (Table 1) have been randomly assigned into two groups of 21 and 3 data sets for training and validation of the network, respectively.

Table 4 Size and shape parameters of non-agglomerated and agglomerated turmeric samples Values are reported as mean ± SD. Moisture content (%)

Steaming time (min)

Surface area (mm2)

Perimeter (lm)

Roundness (–)

Equivalent diameter (lm)

Elongation (–)

Compactness (–)

6 (Control) 10

– 0 30 60 0 30 60

1.3 ± 0.1 1.4 ± 0.2 1.5 ± 0.3 1.5 ± 0.4 1.7 ± 0.4 2.1 ± 0.7 2.4 ± 0.4

132 ± 10 210 ± 23 299 ± 43 518 ± 152 504 ± 106 648 ± 69 691 ± 32

0.64 ± 0.06 0.86 ± 0.16 0.82 ± 0.13 0.66 ± 0.17 0.70 ± 0.08 0.67 ± 0.04 0.85 ± 0.15

37.9 ± 1.8 42.0 ± 3.0 44.1 ± 4.0 43.0 ± 5.0 46.8 ± 6.0 52.1 ± 0.9 68.0 ± 7.1

1.57 ± 0.15 1.20 ± 0.24 1.23 ± 0.19 1.59 ± 0.39 1.44 ± 0.17 1.51 ± 0.10 1.45 ± 0.21

0.56 ± 0.03 0.92 ± 0.09 0.45 ± 0.04 0.29 ± 0.04 0.29 ± 0.02 0.27 ± 0.01 0.26 ± 0.08

28

Input layer

Hidden layers

Output layer

Extent of fines (%) Moisture content (%) Steaming time (min)

Bulk density (kgm-3) Wetting time (s) Sinking time (s)

Fig. 8. Four-layered neural network structure (2–10-8–4) relating process parameters and physico-chemical attributes of agglomerated turmeric samples.

133

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134 Table 5 Validation of results using artificial neural network. Experiment No.

7 14 21

Experimental Predicted Experimental Predicted Experimental Predicted

Input variables

Response function

Moisture content (%)

Steaming time (min)

Extent of fines (%)

Bulk density (kg m3)

Wetting time (s)

Sinking time (s)

15

40

22

20

28

0

77.1 73.8 71.3 71.8 68.3 71.7

476.7 478.2 470.4 462.8 433.1 457.3

27.0 27.8 24.0 24.7 23.0 23.9

83.0 78.1 68.0 73.0 63.0 70.4

The configurations used for training the first network are one and two hidden layers with the total number of neurons in the hidden layers varying between 4 and 10. The number of iteration and learning rate (b) are kept constant at 5000 and 0.5, respectively. The error obtained from different networks is shown in Annexure 1. Based on the initial estimated error values, a network having two hidden layers containing 10 and 8 neurons have been tentatively decided with an error of 0.066 (Annexure 1). Latter, the learning rate (b) has been varied from 0.2 to 0.7. However, no appreciable reduction has been achieved by varying b and even after assigning values greater than 0.7 (Annexure 2). Hence, the learning rate (b) of 0.5 and iterations of 5000 has been decided; the corresponding error of the network is 0.066. In the next stage, the number of iterations has been varied up to 50,000. Considering the error values, which are approximately stable between 25,000 and 500,000 iterations, a value of 30,000 has been finalized. Hence, the finalized network has a structure of 2–10-8–4 (Fig. 8) with a learning rate of 0.5 and number of iterations of 30,000, which has an error of 0.017. The remaining three data sets (not used during the training process) have been used for validation and their comparison is shown in Table 5. The RMS error during validation is 7.6%, which appears to be satisfactory for predicting a complex biological process like agglomeration of turmeric powder. In the present study, the second order regression equations (Eq. (11)) are used to understand the effects of independent variables on the response functions; a moderate fit for all the graphs has been obtained as the multiple correlation coefficients (R) are between 0.84 and 0.91 (significant at p 6 0.01). However, regression equations have their own limitations like inadequate optimization and accurate prediction. Hence, artificial neural network (ANN) has been used which is a stronger tool for prediction of target parameters. The finalized network has a RMS error of 7.6% which appears satisfactory for predicting a complex biological process like agglomeration of turmeric powder. However, ANN does not show the effect of individual independent variables. It is, thus, inferred that a combination of both is desirable to have a detailed picture of the complex process like agglomeration. The regression based 3D graphs show the effect of variables while ANN acts as a good tool for prediction.

4. Conclusions The turmeric powder samples are subjected to agglomeration process by varying the moisture content of the feed and steaming time, and the physical, physico-chemical and morphological characteristics of the product have been determined. The mass mean diameter of the particles or granules increases with an increase in moisture content of the feed. Physico-chemical properties like bulk density, and wetting and sinking times of the agglomerated samples decreases with an increase in moisture content of the feed. Steaming time shows only a marginal effect on these parameters.

Marginal decrease also occurs in the brightness of the agglomerated turmeric samples. The microstructural observation shows that the non-agglomerated sample possesses spheroids and ellipsoids of different sizes having uneven outer surfaces. The image analysis infers that the size related parameters like surface area, perimeter and equivalent diameter increase with an increase in moisture content and/or steaming time, while shape related parameters like roundness, elongation and compactness fail to provide any trend. A four-layered artificial neural network (ANN) having a structure of 2–108–4 has been developed with a root mean square error of 7.6% to predict the quality attributes of the agglomerated turmeric powder.

Acknowledgements The first author wishes to thank the Council of Scientific and Industrial Research (CSIR), New Delhi, India for awarding the Senior Research Fellowship (SRF) to conduct the Ph.D. research programme. The authors thank Mr. Krishna Murthy and Mr. K.G. Girish of Food Engineering Department for their assistance in helping sieve analysis and analysis of results.

Appendix A. See Tables 6 and 7.

Table 6 Error of ANN during training of the network using different layers and number of neurons. Number of hidden layer

1

2

Number of neurons in different hidden layers First layer

Second layer

4 6 8 10 4 4 4 4 6 6 6 6 8 8 8 8 10 10 10 10

– – – – 4 6 8 10 4 6 8 10 4 6 8 10 4 6 8 10

Error

0.167 0.138 0.132 0.126 0.158 0.134 0.169 0.148 0.099 0.084 0.082 0.105 0.097 0.075 0.077 0.074 0.076 0.067 0.066 0.067

134

K. Dhanalakshmi, S. Bhattacharya / Journal of Food Engineering 120 (2014) 124–134 Table 7 Error of ANN by varying learning rate (b). Learning rate (b)

Error

0.2 0.3 0.4 0.5 0.6 0.7

0.070 0.066 0.066 0.066 0.097 0.078

References Barbosa-Canovas, G.V., Rivas, E.O., Juliano, P., Yan, H., 2005. Size enlargement. In: Food Powders: Physical Properties Processing and Functionality. Kluwer Academic/Plenum, New York, pp. 175–198. Baughmen, D.R., Liu, Y.A., 1995. Neural network. In: Bioprocessing and Chemical Engineering. Academic Press, California. Bhattacharya, S., 1999. Yield stress and time-dependent rheological properties of mango pulp. Journal of Food Science 64, 1029–1033. Bhattacharya, S., Patel, B.K., 2007. Simulation of coating process: rheological approach in combination with artificial neural network. Journal of Texture Studies 38, 555–576. Bizot, H., 1983. Using the GAB model to construct sorption isotherms. In: Jowitt, R., Escher, F., Hallström, B., Meffert, H.F.T., Spiess, W.E.L., Vos, G. (Eds.), Physical Properties of Food. Allied Science Publishers, London, pp. 43–54. Buffo, R.A., Probst, K., Zehentbauer, G., Luo, Z., Reineccius, G.A., 2002. Effects of agglomeration on the properties of spray-dried encapsulated flavors. Flavour and Fragrance Journal 17, 292–299. Dhanalakshmi, K., Bhattacharya, S., 2012. Flow and functional characterization of corn starch powder in presence of water and pregelatinized starch. Journal of Food Process Engineering 35, 887–897. Dhanalakshmi, K., Jaganmohan Rao, L., 2012. Comparison of chemical composition and antioxidant potential of essential oil from fresh, dried and cured turmeric (Curcuma longa) rhizomes. Industrial Crops and Products 38, 124–131. Dhanalakshmi, K., Jaganmohan Rao, L., Bhattacharya, S., 2011. Turmeric powder and starch: selected physical, physico-chemical and microstructural properties. Journal of Food Science 76, C1284–C1291. Ghosal, S., Indira, T.N., Bhattacharya, S., 2010. Agglomeration of a model food powder: effect of maltodextrin and gum Arabic dispersions on flow behavior and compacted mass. Journal of Food Engineering 96, 222–228.

Govindarajan, V.S., Stahl, W.H., 1980. Turmeric-chemistry, technology, and quality. CRC Critical Reviews in Food Science and Nutrition 12, 199–301. Hernandez, J.A., Heyd, B., Trystram, G., 2008. Prediction of brightness and surface area kinetics during coffee roasting. Journal of Food Engineering 89, 156–163. Hutchings, J.B., 1994. Food Colour and Appearance. Blackie Academic and Professional, London, pp. 159–163. IDF, 1979. IDF Standard 87. Belgium: International Dairy Federation. Jayaprakasha, G.K., Jaganmohan Rao, L., Sakariah, K.K., 2005. Chemistry and biological activities of C. longa. Trends Food Science and Technology 16, 533– 548. Jensen, B.A., 1994. Expert systems – neural networks. In: CF, Moor, BG, Liptak (Eds.), Instrument Engineers Handbook, Process, Control, third ed. Butterworth Heinemann Ltd., UK, pp. 48–54. Kshirsagar, A.C., Yenge, V.B., Sarkar, A., Singhal, R.S., 2009. Efficacy of pullulan in emulsification of turmeric oleoresin and its subsequent microencapsulation. Food Chemistry 113, 1139–1145. Little, T.M., Hills, F.J., 1978. Agricultural Experimentation: Design and Analysis. John Wiley and Sons, New York, pp. 247–266. Macias-Garcia, A., Cuerda-Correa, E.M., Diaz-Diez, M.A., 2004. Application of RosinRammler and Gates-Gaudin-Schumann models to the particle size distribution analysis of agglomerate cork. Materials Characterization 52, 159–164. McCabe, W.L., Smith, J.C., Harriot, P., 2005. Unit Operations of Chemical Engineering, seventh ed.. McGraw-Hill, Boston, pp. 967–1000. Mukherjee, S., Bhattacharya, S., 2006. Characterization of agglomeration process as a function of moisture content using a model food powder. Journal of Texture Studies 37, 35–48. Palzer, S., 2011. Agglomeration of pharmaceutical, detergent, chemical and food powders similarities and differences of materials and processes. Powder Technology 206, 2–17. Peter, K.V., 2004. Introduction. In Handbook of Herbs and Spices, vol. 2. Woodhead Publishing, Cambridge, pp. 1–8. Pietsch, W., 2003. An interdisciplinary approach to size enlargement by agglomeration. Powder Technology 130, 8–13. Prasher, C.L., 1987. Particle shape, size and surface. In Crushing and Grinding Process Handbook. John Wiley and Sons, Chichester, UK., pp. 46–118. Rosin, P., Rammler, E., 1933. The laws governing the fineness of powdered coal. Journal of the Institute of Fuel 7, 29–36. Russ, J.C., 2011. Characterizing shape. In: The Image Processing Handbook, sixth ed.. CRC Press, Taylor and Francis Group, Florida, pp. 597–623. Singh, R.R.B., Ruhil, A.P., Jain, D.K., Patel, A.A., Patil, G.R., 2009. Prediction of sensory quality of UHT milk – a comparison of kinetic and neural network approaches. Journal of Food Engineering 92, 146–151.