You are on page 1of 11

Chapter 6

ANALYTICAL MODEL OF THE GRAPE DRYING

This chapter deals with the drying kinetics of the solar grape drying. It also elaborates,
about the best suitable mathematical model for the Thompson seedless grape solar
drying by using the forced convection. Diffusion models are the best suitable and
describing the drying phenomenon of the high moisture food product like grapes.
Nearly all such models present in the literature are tested for goodness of fit of grape
drying. Statistical analysis of all nine models to find the coefficient of determination
R2, reduced χ2and RMSE are also found out and compared to all tested models.

6.1 INTRODUCTION
Generally, in case of food product drying process, moisture effective diffusivity was
used to describe the drying characteristics. As very limited information available in
literature on the mechanism of moisture movement during drying and complexity of
the process, which may involve molecular diffusion, capillary flow, Knudsen flow,
hydrodynamic flow, surface diffusion and all other factors. The drying characteristics
in the falling rate period can be described by using Fick’s diffusion equation. This
equation can be used to evaluate effective diffusivity of spherical particles. These
characteristics can be investigated by effectively modeling the drying behavior of the
grapes in the solar dryer. It was considered that the whole drying process of grapes
occurs under the falling rate period and moisture diffusion controls the process. The
solution of the Fick’s second law equation can be given by Equation 6.11 with the
assumption that neglecting shrinkage, constant temperature and diffusion coefficients
and uniform initial moisture distribution,
̇ −
=
� � −

= { (− )+ (− )

+ (− )+ (− )

+⋯..+ (− )} → +∞

(6.11)

95 | P a g e
Where, is the moisture effective diffusivity (m2/s); r is volume equivalent diameter
of the grape samples, with 0.9 X 10-2 m as its value; t is the total drying time and n is
positive integer. For long drying time, the above can be simplified as Eq. (6.12) by
taking the first term of series solution,
̇ (6.12)
= (− )

By taking natural log of both sides we get following Eq. (6.12). The moisture
effective diffusivity can be calculated using the method of slopes. It is typically
determined by plotting the experimental drying data in terms of log versus time.
So the slope can be given from the Linear Regression of log versus time curves,
then the effective diffusion coefficients ( ) can be determined as follows;
̇ (6.13)
Ln = Ln ( )−

And,
(6.14)
= ℎ

The effective diffusivity ( is not constant parameter rather it depends on the


moisture content of the grapes. It can be estimated by an analysis of the experimental
drying data. The non-linear drying curve gives the estimated value of effective
diffusivity coefficient by method of slope (Aed et al., 2014; Mahesh, 2012; Yadav et
al., 2011; Shah, 2007).
0.1
0
-0.1
-0.2
-0.3 High
MC zone
In MR

-0.4 Medium In MR
-0.5 MC zone
-0.6
R² = 0.9939
-0.7 Low
MC zone
-0.8
0 5 10 15 20 25 30 35 40
Drying time (Hrs.)

Fig. 6.1 Ln(MR) Vs drying time to find the experimental value of effective
diffusivity

96 | P a g e
The method of slope estimates the coefficient of diffusivity at each moisture content
value (MR) and it assumes that it represents mass transfer by diffusion inside the
grape. From the experimental data with using Equation (6.14) and figure 6.1, the
experimental value of effective moisture diffusivity is calculated in the range of
1.15x10-7 to 2.1x10-7 (m2/s) with average value of 1.81x10-7(m2/s) ( with R2= 0.993).
Figure 6.1is the plot of Ln(MR) against the drying time for the sample
experiment. The analysis of this data reveals that this curve can divide into three parts
which represent different drying zones according to linearity observed. The names are
given to these zones as High Moisture Content (MC), Medium MC and Low MC zone
which is shown above in the figure 6.1. It is observed that the effective diffusivity
coefficient is high (2.1x 10-7 m2/s) in the High MC zone and it decreases to low of
(1.15x 10-7 m2/s) in the Low MC zone. It means that the moisture migration rate
decreases as moisture content decreases.
The mass loss due to evaporation at the skin of the grape is a function of the
partial vapour pressure difference between the skin layer pressure and working fluid
pressure passing over the grapes which is the driving force to cause the mass transfer.
This mass flux from the skin of the grapes is therefore a function wetness ( moisture
content at that instant) and the skin temperature which is considered to be the same as
that of working fluid temperature ( steady state condition).

6.2 DRYING KINETICS

Drying kinetics is often presented by measuring the average product moisture content
as a function of time. This relationship is known as a drying rate. During the initial
drying stages, excess moisture on the product surface results in a rapid rate of
moisture removal. Subsequent drying of the material depends on the product-
dependent rate, at which internal water migrates to the product surface via diffusion
(Chen et al., 2009).

Several experiments on drying kinetics of the grapes were conducted using the
developed solar ETC dryer. The drying behaviour of the Thompson seedless grapes
was studied under the uncontrolled environmental conditions. The drying experiments
were performed under the forced convection mode. The various mathematical models
which are extensively suitable for high moisture content food product such as grapes

97 | P a g e
are tested, in order to find the best suitable drying model. MATLAB program was
made to calculate the model constants. (Appendix B) These nine different
mathematical models are solved for their constants and listed in the Table 6.1.
The various experiments are carried out and non dimensional moisture ratio
(MR) is found out. Nine different well known mathematical models are tested for
curve fitting. The Regression Analysis was performed using the statistical computer
program. The goodness of fit of the tested mathematical models to the experimental
data was evaluated from the coefficient of determination R2and the reduced Chi-
Square χ2 between the predicted and experimental values.

Table: 6.1 Values of mathematical model constants


Model Name Model constants for Model constants for
Set I Set II
Newton k= 0.04459 k=0.04038
 MR = e − kt
Page k=0.02756 k=0.03074
 MR = exp(-ktn ) n=1.161 n=1.09
Henderson and Pabies a=1.052 a=1.032
 MR = a.e-kt k=0.04755 k=0.04217
Logarithmic a=1.117 a=1.038
 MR = a.e- kt+ c k=0.04128 k=-0.006966
c=-0.07868 c=0.04163
Two term a=-0.06471 a=1.068
 MR = a.e-ko+ b.e-k1t b=1.065 b=-0.06991
k0=20.59 k0=0.04396
k1=0.0483 k1=0.3477
Two-term exponential a=1.708 a=0.8546
 MR = ae − k t+(1− a) e - k a t k=0.06196 k=0.0435
Wang and Sing a= -0.03823 a=-0.03629
MR =1+ at + bt2 b=0.0004343 b=0.00042
Diffusion approach a=-0.1465 a=-0.06821
 MR=a.e −kt+(1−a)e−k bt b= 0.1936 b=0.1276
k=0.2684 k=0.3443
Modified Henderson & Pabis a=13.64 a=1.71
 MR=a.e−kt+b.e−gt+c.e-ht b=-12.92 b=-0.9182
c= 0.5395 c=0.8023
k=0.03381 k= 0.02747
g=0.03406 g=0.02666
h=0.2727 h=0.9161

The Root Mean Square Error (RMSE) is also calculated to validate the predicted data
with experimental data (Aed, 2014; Ayyappan, 2012; Shah, 2007). The higher the R2

98 | P a g e
values and the lower the χ2 values, the better is the goodness of fit (Ertekin and
Yaldiz, 2004).The reduced Chi-Square and RMSE can be calculated by following
Equations 6.15 To 6.17. Custom made programs in MATLAB14B is used to find the
values which are shown in following Table6.2.

{∑�= MR� − MR , � MR� − MR ,� }


Ṙ = (6.15)
√{∑�= MR� − MR ,� ∑�= MR� − MR ,� }

{∑�= MRexp� − MR ,�
χ = (6.16)

= {√∑ MRexp� − MR ,� } (6.17)


�=

Table: 6.2 Values of statistical analysis


Model Name R2 χ2 RMSE
Newton 0.9853 0.001027 0.03125
Page 0.993 5.19162e-05 0.02071
Henderson & Pabis 0.987 4.29772e-05 0.0221
Logarithmic 0.989 1.13128e-07 0.02393
Two term 0.9783 0.007111 0.02356
Two-term exponential 0.9665 0.001356 0.06098
Wang and Singh 0.9826 0.000563 0.02251
Diffusion approach 0.9847 3.94521e-04 0.0192
Modified Henderson & Pabis 0.9003 0.004435 0.1875

By comparing all the eight models which gives the highest R2, the lowest reduced χ2
and the lowest RMSE is considered as the best describing model. By comparison
these values from Table 6.2, it is observed that Page’s model describe accurately the
drying process of Thompson seedless grapes with coefficient of determination value
R2=0.993, reduced χ2=5.19x10-5 and RMSE = 0.02071. The values of model constants
are in the range of k=0.02645 to 0.0312 and n=1.17 to 1.087 for all the experiments.

99 | P a g e
1.2
R2 = 0.993
1 χ2 = 5.19162e-05
RMSE = 0.02071
0.8 Actual MR
Experimental MR
0.6
Page Model MR

0.4

0.2

0
0 10 20 30 40
Drying Time (Hrs.)

Fig. 6.2 Comparison of actual moisture ration with the Page’s model

Above figure 6.2 shows the comparison of the Page’s model and actual data plotted
against the drying time for a sample test. The coefficient of determination, R2 = 0.993,
reduced χ2 = 5.19162e-05to relate the dependency of the independent variables on
each other and the RMSE = 0.02071 is observed as the best suitable amongst all the
values. Whereas the values for all these statistical parameters to all the experiments
are shown in following table:

Table: 6.3 Statistical parameter values for Page’s model

Parameter Range/ Values Average Best Suitable


Value
Coefficient of 0.989 to 0.993 0.9905 0.993
Determination(R2)
Reduced Chi-Square (χ2) 6.1442e-04 to 1.1261 e-06 7.4543 e-05 5.19162e-05
Root Mean Square Error 0.0187to 0.0562 0.0342 0.02071
(RMSE)

From figure 6.2 it is clear that the Page model is describing the drying behavior of the
Thompson seedless grapes very accurately. The coefficient determination value (R2)
is observed 0.993 whereas, the in figure 6.3 the comparison is shown of the actual
data along with Newton’s model plotted against the drying time for same
experimental test readings.

100 | P a g e
1.2

1 R2 = 0.9853
χ2 = 0.001027
0.8
RMSE = 0.03125
Moisture Ratio MR

0.6

0.4
Newton model
0.2 Experimental data

0
0 5 10 15 20 25 30 35 40
Drying time (Hrs.)

Fig. 6.3 Comparison of actual moisture ration with the Newton’s model

The coefficient of determination, R2 = 0.9853, reduced χ2 = = 0.001027 to relate the


dependency of the independent variables on each other and the RMSE = 0.03125 is
observed as the best suitable amongst all the values. Whereas the values for all these
statistical parameters to all the experiments are shown in the following Table,

Table: 6.4 Statistical parameter values for Newton’s model

Parameter Range/ Values Average Best Suitable


Value
Coefficient of 0.97 to 0.9855 0.9789 0.9853
2
Determination(R )
Reduced Chi-Square (χ2) 1.825e-03 to 3.126 e-04 1.1432 e-03 0.001027
Root Mean Square Error 0.02324 to 0.06734 0.04223 0.03125
(RMSE)

101 | P a g e
1.2

1
R2 = 0.987
χ2 = 4.29772e-05
RMSE = 0.0221
Moisture Ratio MR 0.8

0.6

0.4

Handerson and Pabis model


0.2
Experimental data
0
0 10 20 30 40
Drying time (Hrs.)

Fig. 6.4 Comparison of actual moisture ration with the Henderson & Pabis

model

Above figure 6.4 shows the comparison of the Henderson & Pabis model and actual
data plotted against the drying time for a sample test. The coefficient of
determination, R2 = 0.987, reduced χ2 = 4.29772e-05 to relate the dependency of the
independent variables on each other and the RMSE = 0.0221 is observed as the best
suitable amongst all the values. Whereas the values for all these statistical parameters
to all the experiments are shown in following Table,

Table: 6.5 Statistical parameter values for Henderson & Pabis model

Parameter Range/ Values Average Best Suitable


Value
Coefficient of 0.971 to 0.987 0.982 0.987
2
Determination(R )
Reduced Chi-Square (χ2) 3.4334e-03 to 5.5754 e-05 3.897 e-04 4.29772e-05
Root Mean Square Error 0.0110 to 0.0671 0.0431 0.0221
(RMSE)

102 | P a g e
1.2

1 R2 = 0.989
χ2 = 1.13128e-07
0.8 RMSE = 0.02393
Moisture Ratio MR

0.6

0.4

Logarithmic model
0.2
Experimental data
0
0 5 10 15 20 25 30 35 40
Drying time (Hrs.)

Fig. 6.5 Comparison of actual moisture ration with the Logarithmic model

Above figure 6.5 shows the comparison of the logarithmic model and actual data
plotted against the drying time for a sample test. The coefficient of determination, R2
= 0.989, reduced χ2 = 1.13128e-07 to relate the dependency of the independent
variables on each other and the RMSE = 0.02393 is observed as best suitable amongst
all the values. Whereas the values for all these statistical parameters to all the
experiments are shown in following Table,

Table: 6.6 Statistical parameter values for Logarithmic model

Parameter Range/ Values Average Best Suitable


Value
Coefficient of 0.974 to 0.989 0.98103 0.989
Determination(R2)
Reduced Chi-Square (χ2) 2.3134e-05 to 1.3256 e-07 1.5624 e-06 1.13128e-07
Root Mean Square Error 0.0134 to 0.0876 0.05682 0.02393
(RMSE)

Whereas, other models like, Two term, Two term exponential, Diffusion and modified
Henderson & Pabis model deviate much with respect to the actual data and these are
less accurate. The comparison of all these models together is shown in the following
figure 6.6.

103 | P a g e
Pages
1.05
Newton
0.95
Logarthmic
Moisture Ratio MR 0.85
Two term
0.75
0.65
Two term Exponential

0.55 Diffusion
0.45 Henderson &Pabis
0.35 Wang & Sing
0.25
0.15
1 6 11 16 21 26 31 36
Drying time (Hrs.)

Fig. 6.6 Comparison of different mathematical models of drying

It is concluded from the statistical analysis that, Page’s model describes the drying of
Thompson seedless grapes accurately with forced convection by using the solar ETC
system.
The corresponding values of drying constants ‘k’ and ‘N’ for this system using
the Page’s model are 0.0275 and 1.161 respectively with very high coefficient of
determination (R2 =0.993) nearly equal to 1.
1.2

1
Predicted MR (Page's Model)

y = 0.9929x + 0.0024
R² = 0.9937
0.8

0.6

0.4
Comparision of
Mrexp & MRpre
0.2

0
0 0.2 0.4 0.6 0.8 1 1.2

Experimental MR ( Actual)

Fig. 6.7 Comparison of MR predicted Vs MR experimental

104 | P a g e
From figure 6.7 it is clear that the predicted values calculated by the mathematical
model have describing completely the drying phenomenon. The predicted values of
MR are almost matching on all experimental values with coefficient of determination
R² = 0.993.

6.3 CONCLUDING REMARKS

Drying kinetic study of the Thompson seedless grapes is carried out in this chapter.
Effective moisture diffusivity is calculated experimentally by using the Fick’s
equation and observed data in the form of log MR plotting against the drying time.
The various nine mathematical models, suitable for high moisture content food
product like grapes, which are present in the literature, are studied for goodness of fit.
This explains the drying behaviour of the grapes most accurately. The values of all the
drying constants of these models are calculated for all experimental tests by using the
MATLAB program. Selective data is tabulated in this chapter. It is concluded that
Page’s model describe most accurately the drying behaviour of the Thompson
seedless grapes under uncontrolled environmental conditions, by using current solar
dryer setup.

105 | P a g e

You might also like