Accepted Manuscript

Application of artificial neural network method for prediction of osmotic

pretreatment based on the energy and exergy analyses in microwave drying of
orange slices

Mohsen Azadbakht, Mohammad Vahedi Torshizi, Fatemeh Noshad, Arash Rokhbin

PII: S0360-5442(18)31998-4

DOI: 10.1016/j.energy.2018.10.017

Reference: EGY 13914

To appear in: Energy

Received Date: 04 June 2018

Accepted Date: 04 October 2018

Please cite this article as: Mohsen Azadbakht, Mohammad Vahedi Torshizi, Fatemeh Noshad,
Arash Rokhbin, Application of artificial neural network method for prediction of osmotic
pretreatment based on the energy and exergy analyses in microwave drying of orange slices,
Energy (2018), doi: 10.1016/j.energy.2018.10.017

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Application of artificial neural network method for prediction of osmotic

pretreatment based on the energy and exergy analyses in microwave drying of
orange slices
Mohsen Azadbakht*, Mohammad Vahedi Torshizi, Fatemeh Noshad, Arash Rokhbin
Department of Bio-System Mechanical Engineering, Gorgan University of Agricultural
Sciences and Natural Resources, Gorgan, Iran.

* Author for Correspondence: Email: azadbakht@gau.ac.ir

Tel & Fax: +981732440870

Abstract
In the present study, artificial neural network (ANN) method was applied for predicting osmotic
pretreatment based on the energy and exergy analyses in microwave drying of orange slices. For
this purpose, the oranges were cut into slices with a thickness of 4 mm and treated with salt (NaCl)
and distilled water solution (7% by weight) for 30, 60, and 90 min as osmosis pre-treatment. Then,
they were dried in three replicates using a microwave dryer and at three powers of 90, 360, and
900 W. The statistical analysis results showed that the osmosis time is significant for the energy
efficiency and exergy efficiency and specific exergy loss at 1% level. The highest energy and
exergy efficiency was observed at 900 W and in the osmosis time of 90 min. The highest energy
and exergy efficiency was observed at 42.1% and 31.08%, respectively. The maximum exergy loss
was seen at 360 W and osmosis time of 60 min. The osmosis time did not affect the specific energy
loss. The microwave power was statistically significant for all the parameters (energy and exergy)
such that with increasing the microwave power, the energy and exergy efficiency increased, while
the specific exergy and energy loss decreased. Overall, with increasing osmosis time and
microwave power, the energy and exergy levels of the microwave dryer increased. The maximum
coefficient of determination (R2) in a network containing 6 neurons in the hidden layer was 0.999
for energy efficiency, R2 = 0.871 for specific energy loss, R2 = 0.999 for specific exergy loss, and
R2 = 0.972 for exergy efficiency. This amount was seen in a network containing 4 neurons in the
hidden layer.
Keywords: microwave dryer, energy, exergy, osmosis, orange, pre-treatment
1. Introduction
Drying is one of the oldest methods for preserving agricultural and food products. One of the main
goals of drying agricultural products is moving the water in solids to the surface of the product to
a certain extent. Drying is performed in order to reduce the amount of microbial activity in the
product and minimize the amount of chemical interactions in order to minimize the damage
imposed on agricultural products[1][2]. Drying also increases the life of the product and, by
reducing the weight and volume of products, facilitates the packaging, transporting, and storage of
the products, leading to reduced costs. In addition to the above-mentioned cases, drying can control
the market of different products so that the desired product can be used in sensitive situations [3].
In practice, drying is a process that requires a high energy consumption due to the latent heat of
water evaporation. In this regard, 10% of the total energy consumption in the food industry is
related to drying food. Therefore, the drying of agricultural products reflects energy efficiency in
the production of food products and it is very important in industrial uses [4]. Microwave drying

1
 ACCEPTED MANUSCRIPT

is one of the important drying methods. Because of the better focus of energy on the product, the
removal of moisture is faster and, compared to other meth
compared to other drying methods [5,6].
Using microwave waves can reduce drying time by up to 50% or more, while it depends on the
type of product and the drying conditions in different methods. In the microwave drying, the
product is exposed to high-intensity electromagnetic waves. These high-frequency waves select
polar molecules (dipoles) and ions and coordinate them with a fast orientation of the electrical
field. In this orientation process, an adequate amount of heat is produced throughout the product
so that bipolar molecules such as water and salt are vibrated while microwave waves are passing
through inside the product and its biological tissue; as a result of these vibrated molecules, a high
heat content is generated [7]. The drying temperature and microwave power are two important
factors in microwave drying of agricultural products. These two factors significantly affect the
drying parameters such as drying time, drying curve, drying rate, drying efficiency, and final
product quality [8]. The effect of microwave power on drying of products has been examined by
many researchers. For example, Sharifian et al. (2015) evaluated the energy and mass transfer in
a microwave dryer on figs and found that the drying rate increases with increasing microwave
power, and the mass transfer increases when power increases [9]. In addition, Dadal et al. (2010)
examined the effect of the power levels and sample mass on drying okra with microwave. By
increasing microwave power, the drying time of okra was significantly reduced [10]. Darvishi et
al. (2014) analyzed the energy and exergy of white mulberries in the process of drying with a
microwave dryer and reported that the specific energy loss increases with increasing microwave
power. Additionally, energy efficiency was reduced by decreasing the moisture content and
microwave power. The best energy and exergy for white mulberry was observed at 100 W
microwave power [11]. Darvishi et al. (2016) conducted energy and exergy analysis, modeled
Kiwi slices with a microwave dryer, and reported that the energy and exergy efficiency increases
with increasing microwave power and decreasing the thickness of kiwi slices. Additionally, this
parameter decreased by reducing the moisture content of slices [12]. Moreover, nowadays,
improving the quality of agricultural products during the drying is of the highest importance. One
technique commonly used in this regard is pre-treatment [13], which can be applied to preserve
foods and improve the appearance of dried products [14].
According to the literature, osmosis pre-treatment is an easy, controllable, and good method for
fruit drying. It is also an economical and cost-effective method to remove part of the water from
plant and animal products by dipping it into solution [15]. In addition, the use of this method results
in a reduction in the drying time and improves the apparent quality of the samples after drying
[16].
The use of pretreatment in the drying of products was also reported in the following studies: Al-
Harahsheh et al. (2009) examined tomato drying in a microwave dryer with osmosis pretreatment
and reported that by increasing the dryer power and content of salt concentration, the rate of dryer
increased [17]. In addition, Azoubel et al. (2009) examined the effect of osmosis pre-treatment on
drying kinetics and quality of cashew apple and concluded that osmosis of cashew apples had a
higher rate of drying compared to non-pretreated fruits [18]. Moreover, the artificial neural
network (ANN) modeling is widely used in many fields. This method is of high efficiency in
solving the complex and non-linear equations in dryers [19]. Nikbakht et al. (2014) used the ANN
to predict the exergy and drying energy of pomegranate in a thin-layer dryer with microwave [20].
Aghbashlo et al. (2012) used an ANN to predict the performance of the spray dryer for oil and fish
[21]. Azadbakht et al also examined the exergy and energy of the fluid bed dryer using an ANN
for potato drying [22].
2
 ACCEPTED MANUSCRIPT

The objective of this research is the energy and exergy performance analyses of microwave dryers
for drying orange slices under pre-treatment of osmosis in order to reduce the energy consumption
in the microwave dryer and increase the energy and exergy efficiency of the microwave with new
processes. For this purpose, the ANN was applied to verify the accuracy of the numbers obtained.
Additionally, the sensitivity coefficient test was applied to relate the energy and exergy factors to
microwave and pre-treatment.
2. Materials and methods
2.1. Sample preparation
Freshly harvested oranges (Tamson variety) were purchased from a local store in Gorgan city in
Iran and were kept at 10℃ in the laboratory. At the beginning of each experiment, the oranges
were washed and the slices were cut in a circular in a thickness of 4 mm and they were weighted.
Then, samples were pretreated by placing them in an osmosis solution (7 wt.%) containing 1000
gr of distilled water and 70 g salt (NaCl) for 30, 60, and 90 min. Drying process was performed in
a microwave dryer with 1.2, 4.8, and 12 W/g specific power density in the BioSystem Mechanics
Department of Gorgan University of Agricultural Sciences and Natural Resources (Fig. 1).
2.2. Experiment method
Slices were pretreated and placed in containers and dried at three powers of 90, 360, and 900 W.
The weight of oranges was measured using a 0.01 mg precision scale. The weight of each sample
was measured and recorded at a time interval of 1 min to reach a constant moisture. For each
treatment, the experiments were repeated in triplicate. Environmental conditions for testing were
conducted at a temperature of 20℃ and a relative humidity of 71%. First, the oranges were equal
to the slices of the same size, then the sample was placed inside the oven and the weight of the
sample was measured according to the standards. Then, using Eq. 1, the moisture content was
calculated [23].
𝑊 ‒ 𝑊e
𝑀𝐶 = (1)
𝑊
2.3. Energy analysis
Energy used in the drying and heating process is important for production processes in the
industrial and household sectors. However, the price of this energy is extremely expensive;
therefore, there is a strong incentive to invent processes that will use energy efficiently.
Currently, widely used drying and heating processes are complicated and inefficient and are
generally damaging to the environment. Thus, it is required to have a simplified lower-cost
approach replicable in a wide range of situations [24].
The mass and energy survival in the microwave dryers’ chamber is shown in Fig. 2. The general
relation of mass moisture survival is calculated using Eq. (2) [12].

∑𝑚𝑖𝑛 = ∑𝑚𝑜𝑢𝑡 (2)

According to Eq. 3, the initial mass of the sample is equal to the amount of water vapor removed
and the rate of dried sample mass.
𝑚𝑜 = 𝑚𝑒𝑤 + 𝑚𝑝 (3)

The mass of evaporated water is obtained using Eq. 4 [11].

3
 ACCEPTED MANUSCRIPT

𝑚𝑤𝑡 = 𝑚𝑑(𝑀0 ‒ 𝑀𝑡) (4)

The protected energy of the sensible heat, latent heat, and the thermal source of the microwave
were calculated using Eq. 5 and the input energy of the dryer was calculated using Eq. 6 [25]. In
Eq. 5, the energy loss is 𝑃𝑟𝑒𝑓 + 𝑃𝑡𝑟𝑎. Eq. 6 shows the input energy of the microwave. This
formula is composed of three parts, including absorbed energy, reflected energy, and passed
(
energy. In Eq. 6, (mCpT) ‒ (mCpT)
dp wp
)
+ 𝜆𝐾𝑚𝑤) equals to the absorbed energy of the
product.

𝑃𝑖𝑛 = 𝑃𝑎𝑏𝑠 + 𝑃𝑟𝑒𝑓 + 𝑃𝑡𝑟𝑎 (5)

Pin × t = ((mCpT)dp ‒ (mCpT)wp) + 𝜆𝐾𝑚𝑤 + 𝐸𝑟𝑒𝑓 + 𝐸𝑡𝑟𝑎 (6)

The latent heat of the orange samples is calculated using Eq. 7 [26].
𝜆𝐾
= 1 + 23exp ( ‒ 40𝑀𝑡) (7)
𝜆𝑤𝑓

The latent heat of free water evaporation was calculated according to Broker et al. using Eq. 8
[27].

= 2503 ‒ 2.386(𝑇 ‒ 273)𝜆𝑤𝑓 (8)

The thermal capacity is a function of the moisture content and can be calculated through Eq. 9
[28].
𝑀𝑡
𝐶𝑃 = 840 + 3350 × ( ) (9)
1 + 𝑀𝑡

The thermal efficiency of the dryer is calculated using Eq. 10 [29].

𝑒𝑛𝑒𝑟𝑔𝑦 𝑎𝑏𝑠𝑜𝑟𝑝𝑡𝑖𝑜𝑛
𝜂𝑒𝑛 = (10)
𝑃𝑖𝑛 × 𝑡
The specific energy loss was measured using Eq. 11 [11]

𝐸𝑖𝑛 ‒ 𝐸𝑎𝑏𝑠 𝑃𝑖𝑛 × 𝑡

𝐸𝑙𝑜𝑠𝑠 = 𝑜𝑟 𝐸𝑙𝑜𝑠𝑠 = (1 ‒ 𝜂𝑒𝑛) × (11)
𝑚𝑤 𝑚𝑤

2.4. Exergy analysis

4
 ACCEPTED MANUSCRIPT

With the onset of the energy crisis, energy and exergy (the maximum useful work that comes
from a certain amount of available energy or from the flow of materials) analyses are among the
leading thermodynamic research works. In the exergy analysis, the main purpose is to determine
the location and amount of irreversible production during the various processes of the
thermodynamic cycle and the factors affecting the production of this irreversibility. In this way,
in addition to evaluating the performance of various components of the thermodynamic cycle,
methods to increase cycle efficiency are also identified [30].

The general exergy equilibrium in the microwave chamber is as follows [12].

The amount of exergy transmitted due to evaporation in the drying chamber was calculated using
Eq. 14 [31]
𝑇0
𝑒𝑥'𝑒𝑥𝑎𝑝 = (1 ‒ ) × 𝑚𝑤𝑣𝜆𝑤𝑝 (14)
𝑇𝑝

where mwv is calculated using Eq. 15 [12].

𝑚𝑡 + ∆𝑡 + 𝑚𝑤𝑣𝜆𝑤𝑝
𝑚𝑤𝑣 = (15)
∆𝑡

Specific exergy loss was calculated using Eq. 16 [11]:

𝑇
𝑒𝑥 = 𝐶𝑝[(𝑇 ‒ 𝑇0) ‒ 𝑇0ln ( )] (16)
𝑇0

Exergy efficiency for each dryer system – as the exergy rate used in drying the product to the
exergy of drying source supplied to the system – is calculated by Eq. 17 [32]
𝑒𝑥𝑒𝑟𝑔𝑦 𝑎𝑏𝑠𝑜𝑟𝑝𝑡𝑖𝑜𝑛
𝜂𝑒𝑛 = × 100 (17)
𝑃𝑖𝑛 × 𝑡

The specific exergy loss was calculated using Eq. 18 [27].

(18)

5
 ACCEPTED MANUSCRIPT

In this research, the source of temperature and pressure in the environment was 20℃ and 101.3
MPa, respectively.
𝐸𝑋𝑖𝑛 ‒ 𝐸𝑋𝑎𝑏𝑠
𝐸𝑋𝑙𝑜𝑠𝑠 = (18)
𝑚𝑤
2.4. Mass Transfer Parameters

2.4.1Drying Kinetics
The change of moisture in orange slice during the drying process was expressed as the moisture ratio defined
as[33]:

𝑀𝑡 ‒ 𝑀𝑒
𝑀𝑅 = ( )
𝑀0 ‒ 𝑀
𝑒
(19)

2.4.2. Moisture Diffusivity

The effective moisture diffusivity is an important transfer property required in the modeling of various drying
food processes
The Fick’s diffusion equation developed for solid objects with spherical geometry by Crank (1975) was applied
to the experimental data on the assumption that there are a uniform initial moisture distribution and negligible
external resistance. For this purpose, Eq. 20 is used [27]:

∞ 𝐷
6 1 2 2 𝑒𝑚
𝑀𝑅 = 2 ∑
𝜋 𝑛 = 0𝑛
2
exp ( ‒ 𝑛 𝜋
𝑟
2
𝑡)
(20)

Simplifying Eq. 20 by taking the first term of the series solution gives [12]:
6 𝐷
2 𝑒𝑚
𝑀𝑅 = 2
exp ( ‒ 𝜋 2 𝑡) (21)
𝜋 𝑟

Effective diffusivity is also typically calculated using the slope of Eq. (22); when a natural logarithm of MR
versus time was plotted, a straight line was obtained with a slope of:
𝐷
2 𝑒𝑚
𝜋 2
(22)
𝑟

2.5. Mass Transfer Coefficient

Kaya et al. (2010) described a procedure to determine the mass transfer coefficient as follows [34]:

𝑉𝑝
ℎ𝑚 =‒ ln (𝑀𝑅) (23)
𝐴𝑝 × 𝑡

For a symmetrically heated sphere, V/A is equal to the radius [35], hence for a given orange slice, Eq. (24) can
be simplified to:
𝑟
ℎ𝑚 =‒ ln (𝑀𝑅) (24)
𝑡

6
 ACCEPTED MANUSCRIPT

2.6. ANN Modeling

In this research, the multilayer perceptron (MLP) ANN was used for modeling the energy and
exergy of the microwave dryer in order to predict energy efficiency, specific energy loss, exergy
efficiency, and specific exergy loss by one hidden layer and 4, 6, and 8 neurons using the
NeuroSolutions 5 software. Hyperbolic tangent linear activation functions (Eq. 25), which are the
most common type of activation functions, were used in the hidden input and output layer. In this
paper, the Levenberg-Marquardt algorithm was applied to teach the network. Additionally, 75%
of the data were used for training, 10% were used for network evaluation (Validating Data), and
15% were used for testing the network (Testing Data) (Table 2). The three microwave power (90
, 360 and 900 W) and the three osmotic time (30, 60, and 90 min) were considered as network
inputs and the energy efficiency, specific energy loss, exergy efficiency, and the specific exergy
loss were the considered network outputs. In Fig. 3, the input and output numbers of the network
are shown. Three repetitions were considered to achieve the minimum error rate and maximum
network stability as a mean of 1000 Epoch for the network. The error was estimated using an
algorithm with back propagation error. Statistical parameters including Root Mean Square Error
(RMSE), R2, and Mean Absolute Error (MAE) were calculated for inputs and relationships using
the formulas shown in Table 1.

2.6. Statistical Analysis

The oranges were dried in the microwave at three powers of 90, 360, and 900 and three osmoses
of 30, 60, and 90 min and the obtained numbers were sorted and calculated using the Excel
software. All experiments were performed in triplicate and the results were analyzed using a
factorial experiment in a completely randomized design with SAS statistical software.
3. Results and discussion
The analysis of variance (ANOVA) results of orange slices drying in different microwave powers
for energy efficiency, specific energy loss, specific exergy loss, exergy efficiency, effective
moisture diffusivity and mass transfer coefficient are shown in Table 3. According to the results,
the power of the microwave was significant for energy efficiency, specific exergy loss, exergy
efficiency and mass transfer coefficient at 1% level and significant for specific energy loss at 5%
level. Additionally, according to Table 3, the results obtained for osmosis have shown the
significance level of 1% for energy efficiency, specific exergy loss, and exergy efficiency and
mass transfer coefficient and non-significance for the specific energy loss in osmosis state. The
interaction effect of energy efficiency, exergy efficiency and effective moisture diffusivity and
mass transfer coefficient at 5% level and the interaction effect of a specific exergy loss are
significant at the 1% level. Thus, we compared the means with the LSD test, which its results are
shown in Tables 3, 5, and 6.

Table 3: ANOVA results of energy efficiency, specific lost energy lost, specific lost exergy and
exergy efficiency, effective moisture diffusivity and mass transfer coefficient under different
powers and osmoses

Specific exergy
Parameter
Energy efficiency Specific energy loss loss Exergy efficiency

7
 ACCEPTED MANUSCRIPT

Mean Mean Mean Mean

Square F Value Square F Value Square F Value Square F Value
27.93*
Microwave power 730.98 * 34.36 12.40* 142.85 23.24** 535.03 322.17**
osmotic 192.57 7.36** 4.27 1.54 ns 51.71 8.41** 126.72 76.31**
Microwave power ×osmotic 117.24 4.48* 3.88 1.40 ns 56.47 9.19** 9.71 5.85*
ERROR 25.94 2.77 6.14 1.66
Effective moisture diffusivity . Mass transfer coefficient
Microwave power 4.423×10-8 2.12ns 0.00012773 9.56**
osmotic 5.571×10-8 2.67ns 0.00007570 5.67**
Microwave power ×osmotic 7.354×10 -8 3.53* 0.00004468 3.35*
ERROR 2.082*10-8 0.00001336

3.1. The effect of power and osmosis on energy efficiency

Based on Table 3, an interaction effect of power and osmosis on energy efficiency is significant at
the level of 5%. Fig. 4 shows the interaction of these parameters on energy efficiency. According
to the results obtained, energy efficiency increased significantly with increasing the power of the
microwave. The obtained results are in line with those of Dadalı et al. (2010) on the drying of okra
in a microwave dryer [10]. In addition, based on the results obtained, with increasing the amount
of osmosis time, the energy efficiency increased significantly. This increase can be explained by
the fact that with increasing the osmosis time, the amount of mass reduction in orange increased,
leading to an increase in the orange dry matter and removed water from oranges. This result is due
to the penetration of sodium and chloride ions to orange tissue during the sample immersion.
Moreover, as NaCl crystals penetrate into cellular parts of the immersed product, cells are
stimulated and the sample contraction is reduced. In addition, the presence of sodium chloride in
the oranges tissue caused a larger amount of unstable water in the tissue of this product, which is
like an ionic solution with better microwave heat characteristic than water, leading to increased
absorption of heat in oranges. Besides, with increasing the osmosis time, temperature absorption
of orange slices also increased, leading to the faster removal of water from the product. It leads to
a reduction in drying time and an increase in the energy efficiency of the microwave dryer,
consequently. The obtained results are in line with those of Al-Harahsheh et al. (2009) on tomato
drying in a microwave dryer [17]. The obtained R2 was also desirable for all drying powers. Table
2 presents the values of R2 for the obtained equations.
Similar capital letters represent non-significance in the same power while similar small letters
represent the non-significance in the same osmosis time.
3.2. The effect of power on specific energy loss
Based on Table 3, the power of the microwave was significant at 5% probability level. Fig. 5 shows
the results obtained. The maximum amount of the specific energy loss is observed at the power of
90 W (7.15 MJ) and the minimum amount of specific energy loss is observed in a power of 900
W (1.24 MJ). As the specific energy loss is inversely related to the water removed from the
product, by increasing the amount of water removed from the product, the amount of specific
energy loss declines. The obtained results are consistent with those of Darvishi et al. (2016) on
kiwi drying in a microwave dryer [12]
3.3. The effect of power and osmosis on exergy efficiency

8
 ACCEPTED MANUSCRIPT

Based on Table 3, an interaction effect of power and

the level of 5%. Also, Fig. 6 shows the interaction of these parameters on exergy efficiency.
The maximum exergy efficiency was observed in a power of 900 W and osmosis of 90 min
(31.08%) while the minimum exergy efficiency was obtained in a power of 90 W and the osmosis
time of 30 min (6.7%). With increasing the power of the microwave, the exergy efficiency
increased significantly. This result can be explained by the fact as power increase, the temperature
of the microwave chamber dryer also increases and the product mass is removed faster, leading to
reduced orange drying time. This reduction in time and faster mass removal, finally, increases the
exergy efficiency of the microwave dryer. For exergy efficiency, the useful power is highly
important and by reducing drying time, the useful power increases. Moreover, according to Fig. 6,
the osmosis time of oranges has a significant effect on the exergy efficiency. Based on this figure,
with increasing the osmosis time, the amount of exergy efficiency increased, indicating a better
heat exchange at higher osmosis times.
The capital letters represent significance in the same power and small letters represent
significance in the same osmosis
As shown in Table 3, the R2 values are adequate for all drying powers.
3.4. The effect of power and osmosis on a specific exergy loss
According to Table 3, the interaction effect of power and osmosis time is significant at 1% for a
specific energy loss. Fig. 7 presents the interaction of these parameters on the specific exergy loss.
The maximum amount of specific exergy loss was observed in a power of 90 W and osmosis time
of 60 min (19.875 MJ) while the minimum specific exergy loss was observed in a power of 900
W and osmium time of 90 min (3.71 MJ). Given these results, the amount of specific exergy loss
is reduced by increasing the power.
Orange drying time is increased at lower temperatures, leading to the increased exergy entering
the microwave chamber and the reduced specific exergy loss. These results are in line with those
of Darvishi (2017) on soybean drying in a microwave dryer [27]. Moreover, based on the results,
the osmosis time did not have a significant effect on the amount of specific exergy loss.
The capital letters represent significance in a similar power and small letters represent
significance in the same osmosis
As shown in Table 4, the obtained R2 values are adequate for powers of 360 and 900 W but very
low for the power of 90 W.

3.5. Effective moisture diffusivity

According to Table 3, the interaction effect of power and osmosis time is significant at 5% level
for an effective moisture diffusivity. Fig. 8 presents the effect of these parameters on the effective
moisture diffusivity. The maximum amount of effective moisture diffusivity was observed in a
power of 900 W and osmosis time of 90 min (52.87×10-5 m²/s) while the minimum specific exergy
loss was observed in a power of 90 W and osmium time of 60 min (47.41×10-6 m²/s). Given these
results, the amount of effective moisture diffusivity is increased by increasing the power. It can be
argued that by increasing the content of heat generated, the activity of the water molecules in the
product is also increased. In addition, receiving the heat by molecules and performing the kinetic
motion between them leads to a selecting a higher moisture diffusivity when a high power is

9
 ACCEPTED MANUSCRIPT

applied for drying. Furthermore, according to the results, at 900 watts, the highest effective
moisture diffusivity was observed. In this power, the energy consumed and specific energy loss
was at best. According to the results obtained for the effective moisture diffusivity, it can be stated
that the best power and osmotic time were 900 watts and 90 min, respectively. The obtained results
are consistent with those of Darvishi et al. (2017) on SOYBEAN drying in a microwave dryer
[27].

3.6. Mass transfer coefficient

According to Table 3, the effect of microwave power and osmotic time is significant at 1% and
interaction effect of power and osmosis time is significant at 5% level for a mass transfer
coefficient. Fig. 9 presents the interaction of these parameters on the mass transfer coefficient.
The maximum mass transfer coefficient was observed in the power of 900 W and osmosis time
of 90 min while the minimum specific exergy loss was observed in power of 90 W and osmium
time of 90 min. According to Fig. 9, at 90 W, no significant difference is seen among any of the
osmosis times. In addition, for other powers, no significant difference is seen between osmosis
times of 30 and 60 min but at 90 min osmosis time, a significant difference is observed in all
drying conditions.

3.7. Artificial neural network (ANN)

A multi-layered perceptron (MLP) neural network model was used to predict energy efficiency,
specific energy loss, exergy efficiency, and specific exergy loss. The duration of placing orange
samples in osmosis and microwave power were considered as network inputs and energy
efficiency, specific energy loss, exergy efficiency, and specific exergy loss were considered as
network outputs. As lower error value was obtained using the hyperbolic tangent activation
(transfer) function, it was applied in both the hidden layer and the output layer. Based on the test
method, 15% of the data were used for training, by which the network could learn the relationships
between inputs and outputs well, 10% were used to test the network, and 75% were used to
evaluate the network (Table 5).
The results showed that the ANN with 6 neurons in the hidden layer for energy efficiency (R2 =
0.999) and specific energy loss (R2 = 0.871) and the neural network with 4 and 6 neurons in the
hidden layer for exergy efficiency (R2 = 0.972) and specific exergy loss (R2 = 0.999) can predict
energy efficiency, specific energy loss, exergy efficiency, and specific exergy loss in different
osmosis times and microwave powers (Table 5). In addition, the neural network with 8 neurons in
the hidden layer has the highest R2 value, after the mentioned layers, for energy efficiency, specific
energy loss, and exergy efficiency. For the specific exergy loss, the highest value was obtained
with a structure with 4 neurons in the hidden layer, following by the structure with 6 neurons in
the hidden layer.

Table 6 shows the best network between input data and the data simulated by the network for each
of the neurons in the hidden layer. Smaller epochs suggest that the number of neurons in the layer
successfully learned by the neural network compared to other neurons.
As shown in Table 6, the best network for energy efficiency at training step (Run = 1, Epoch = 21)
in the 6-neuron state in the hidden layer reaches a constant value after about 21 generations of

10
 ACCEPTED MANUSCRIPT

error. In addition, the best network for the specific energy loss in training (Run = 1, Epoch = 13)
in 8-neuron state in the hidden layer reaches a constant value after about 19 generations of errors.
For exergy efficiency of training value (Run = 1, Epoch = 14), in the 6-neuron state in the hidden
layer and for the specific exergy loss (Run = 1, Epoch = 17), an appropriate overlap was obtained
between the mean square error of the input numbers and the numbers simulated by the network.
To examine the effect of input parameters and identify the most influential factor, the sensitivity
analysis test was performed on the considered networks. As shown in Fig. 10, the osmosis time in
an ANN with 6 neurons in the hidden layer was considered as the most effective factor in
predicting energy efficiency. In addition, in this network, microwave power in hidden layers with
8 neurons had the highest sensitivity.
The results of the sensitivity analysis for the specific energy loss are shown in Fig. 11. Based on
this figure, the highest sensitivity was obtained for the microwave power and the osmosis time in
the hidden layers with 4 neurons. In fact, in the specific energy loss, with increasing the number
of neurons in hidden layers, the sensitivity of input parameters to output data decreased.
For the exergy efficiency, the sensitivity analysis is shown in Fig. 12. Based on this figure, the
highest sensitivity for the microwave power was obtained in the hidden layers with 4 neurons
while the maximum sensitivity coefficient for the osmosis time was obtained in hidden layers
with 4 neurons. Based on the sensitivity analysis for energy efficiency, the sensitivity of
microwave power is reduced with increasing the number of neurons in the hidden layers, while
the sensitivity of osmosis time increases with increasing the number of neurons in the hidden
layers.
Fig. 13 illustrates the sensitivity coefficient for the specific exergy loss. As can be seen, the
highest sensitivity for the microwave power was calculated in hidden layers with 6 neurons while
the highest sensitivity for osmosis time was calculated in the hidden layers with 4 neurons.
Additionally, in all three neuron numbers of 4, 6, and 8, the microwave power sensitivity
coefficient for the specific exergy loss was higher than that of osmosis time.
4. Conclusion
Microwave power plays an important role in determining the characteristic of orange drying. An
increase in the power of the microwave increases the energy and exergy efficiency drying,
leading to the reduced drying time. Osmosis pre-treatment has no significant effect on energy
and exergy loss, and power plays a more important role in these two factors. In addition, osmosis
pre-treatment increases the absorption of heat in orange, leading to an increase in the energy and
exergy efficiency during drying. Also, increasing the osmosis time and microwave power had a
significant effect on the amount of energy and exergy. Based on the results obtained, osmosis
time had more effect on the energy efficiency than the exergy efficiency.According to the results
obtained for the effective moisture diffusivity and mass transfer coefficient, the best power and
osmotic time were 900 W and 90 min. Based on the results obtained from the ANN, the highest
R2 value for energy efficiency, specific energy loss, and exergy efficiency is seen in the ANN
with 6 neurons while for the specific exergy loss, the highest R2 is seen in the network with 4
neurons in the hidden layers. Data obtained from the network and the initial data obtained from
the experiment overlap for the energy efficiency, the specific exergy loss, and the exergy
efficiency in a network with 6 neurons in the hidden layer. Meanwhile, for the specific exergy
loss, the data in a network with 4 neurons in the hidden layer overlap. With this number of
neurons in the hidden layer, the fastest type of learning was seen in the ANN. Based on the
results obtained, the sensitivity coefficient shows the highest sensitivity to osmosis neural
network with 4 neurons in the hidden layers. However, this value for the exergy efficiency was
11
 ACCEPTED MANUSCRIPT

obtained in the hidden layer with 8 neurons. The maximum microwave power sensitivity
coefficient for the energy efficiency and specific exergy loss was obtained in a network with 6
neurons in the hidden layer while for exergy efficiency and specific energy loss, it was obtained
in a network with 4 neurons in the hidden layer. Given the results obtained for R2, RMSE, and
Epoch, it can be stated that the ANN designed in this work has the ability to predict the energy
efficiency, specific energy loss, exergy efficiency, and specific exergy loss at an acceptable level
for orange.

References
[1] Bjarnadottir SG, Lunde K, Alvseike O, Mason A, Al-Shamma’a AI. Assessing quality
parameters in dry-cured ham using microwave spectroscopy. Meat Science 2015;108:109–
14. doi:10.1016/j.meatsci.2015.06.004.
[2] Mukred A, Singh D. Estudo Geral repository of University of Coimbra. International
Journal of Business Information Systems 2017;24:261–84.
[3] Azadbakht M, Vehedi Torshizi M, Aghili H, Ziaratban A. Thermodynamic analysis of
drying potato cubes in a fluidized bed dryer. CARPATHIAN JOURNAL OF FOOD
SCIENCE AND TECHNOLOGY 2017;9:96–106.
[4] Azadbakht M, Vahedi torshii M, Ziaratban A, Aghili H. Energy and exergy analyses during
eggplant drying in a fluidized bed dryer. Agricultural Engineering International: CIGR
Journal 2017;19:177–82.
[5] Wray D, Ramaswamy HS. Novel Concepts in Microwave Drying of Foods. Drying
Technology 2015;33:769–83. doi:10.1080/07373937.2014.985793.
[6] Sharma GP, Prasad S. Specific energy consumption in microwave drying of garlic cloves.
Energy 2006;31:1585–90. doi:10.1016/j.energy.2005.08.006.
[7] Akoy EOM, Von Höresten D. Microwave Drying of Mango Slices at Controlled
Temperatures. International Journal of Innovation and Applied Studies ISSN
2015;12:2028–9324.
[8] Li Z, Raghavan GSV, Orsat V. Temperature and power control in microwave drying.
Journal of Food Engineering 2010;97:478–83. doi:10.1016/j.jfoodeng.2009.11.004.
[9] Sharifian F, Nikbakht AM, Arefi A, Modarres Motlagh A. Experimental assessment of
energy and mass transfer in microwave drying of fig fruit. Journal of Agricultural Science
and Technology 2015;17:1695–705.
[10] Dadalı G, Kılıç Apar D, Özbek B. Microwave Drying Kinetics of Okra. Drying
Technology 2007;25:917–24. doi:10.1080/07373930701372254.
[11] Darvishi H, Zarein M, Minaei S, Khafajeh H. Exergy and energy analysis, drying kinetics
and mathematical modeling of white mulberry drying process. International Journal of
Food Engineering 2014;10:269–80. doi:10.1515/ijfe-2013-0065.
[12] Darvishi H, Zarein M, Farhudi Z. Energetic and exergetic performance analysis and
modeling of drying kinetics of kiwi slices. Journal of Food Science and Technology
2016;53:2317–33. doi:10.1007/s13197-016-2199-7.

12
 ACCEPTED MANUSCRIPT

[13] Mozumder N, Rahman M, Kamal M, Mustafa A, Rahman M. Effects of Pre-drying

Chemical Treatments on Quality of Cabinet Dried Tomato Powder. Journal of
Environmental Science and Natural Resources 2012;5:253–65.
doi:10.3329/jesnr.v5i1.11590.
[14] Hii CL, Ogugo JF. Effect of pre-treatment on the drying kinetics and product quality of
star fruit slices. Journal of Engineering Science and Technology 2014;9:122–34.
[15] Mehdi M, Shahi N, Sabetghadam M, Nemat N. The effect of osmotic and ultrasound pre-
treatment on some physicochemical features of avocado. International Journal of
Pharmaceutical Research & Allied Sciences 2016;5:121–31.
[16] Revaskar VA, Pisalkar PS, Pathare PB, Sharma GP. Dehydration kinetics of onion slices
in osmotic and air convective drying process. Research in Agricultural Engineering
2014;60:92–9.
[17] Al-harahsheh M, Al-muhtaseb AH, Magee TRA. Chemical Engineering and Processing :
Process Intensification Microwave drying kinetics of tomato pomace : Effect of osmotic
dehydration. Chemical Engineering and Processing: Process Intensificatio 2009;48:524–
31. doi:10.1016/j.cep.2008.06.010.
[18] Azoubel PM, El-Aouar ÂA, Tonon RV, Kurozawa LE, Antonio GC, Murr FEX, et al.
Effect of osmotic dehydration on the drying kinetics and quality of cashew apple.
International Journal of Food Science and Technology 2009;44:980–6.
doi:10.1111/j.1365-2621.2008.01783.x.
[19] Özdemir MB, Aktaş M, Şevik S, Khanlari A. Modeling of a convective-infrared kiwifruit
drying process. International Journal of Hydrogen Energy 2017;42:18005–13.
doi:10.1016/j.ijhydene.2017.01.012.
[20] Nikbakht AM, Motevali A, Minaei S. Energy and exergy investigation of microwave
assisted thin-layer drying of pomegranate arils using artificial neural networks and
response surface methodology. Journal of the Saudi Society of Agricultural Sciences
2014;13:81–91. doi:10.1016/j.jssas.2013.01.005.
[21] Aghbashlo M, Mobli H, Rafiee S, Madadlou A. Energy and exergy analyses of the spray
drying process of fish oil microencapsulation. Biosystems Engineering 2012;111:229–41.
doi:10.1016/j.biosystemseng.2011.12.001.
[22] Azadbakht M, Aghili H, Ziaratban A, Vehedi Torshizi M. Application of artificial neural
network method to exergy and energy analyses of fluidized bed dryer for potato cubes.
Energy 2017;120:947–58. doi:10.1016/j.energy.2016.12.006.
[23] Yogendrasasidhar D, Pydi Setty Y. Drying kinetics, exergy and energy analyses of Kodo
millet grains and Fenugreek seeds using wall heated fluidized bed dryer. Energy
2018;151:799–811. doi:10.1016/j.energy.2018.03.089.
[24] Jindarat W, Rattanadecho P, Vongpradubchai S. Analysis of energy consumption in
microwave and convective drying process of multi-layered porous material inside a
rectangular wave guide. Experimental Thermal and Fluid Science 2011;35:728–37.
doi:10.1016/j.expthermflusci.2010.11.008.
[25] Jindarat W, Rattanadecho P, Vongpradubchai S. Analysis of energy consumption in
microwave and convective drying process of multi-layered porous material inside a
13
 ACCEPTED MANUSCRIPT

rectangular wave guide. Experimental Thermal and Fluid Science 2011;35:728–37.

doi:10.1016/j.expthermflusci.2010.11.008.
[26] Abdelmotaleb A, El-Kholy MM, Abou-El-Hana NH, Younis MA. Thin layer drying of
garlic slices using convection and combined (convection - infrared) heating modes. Misr J
Ag Eng 2009;26:251–81.
[27] Darvishi H. QUALITY, PERFORMANCE ANALYSIS, MASS TRANSFER
PARAMETERS AND MODELING OF DRYING KINETICS OF SOYBEAN. Brazilian
Journal of Chemical Engineering 2017;34:143–58. doi:10.1590/0104-
6632.20170341s20150509.
[28] Brooker DB, Bakker-Arkema FW, Hall W. Drying and storage of grains and oilseeds. Van
Nostrand Reinhold, New York 1992;49:450.
[29] Soysal Y, Öztekin S, Eren Ö. Microwave Drying of Parsley: Modelling, Kinetics, and
Energy Aspects. Biosystems Engineering 2006;93:403–13.
doi:10.1016/j.biosystemseng.2006.01.017.
[30] Mokhtarian M, Tavakolipour H, Kalbasi-Ashtari A. Energy and exergy analysis in solar
drying of pistachio with air recycling system. Drying Technology 2016;34:1484–500.
doi:10.1080/07373937.2015.1129499.
[31] Sarker MSH, Ibrahim MN, Abdul Aziz N, Punan MS. Energy and exergy analysis of
industrial fluidized bed drying of paddy. Energy 2015;84:131–8.
doi:10.1016/j.energy.2015.02.064.
[32] Dincer I, Sahin AZ. A new model for thermodynamic analysis of a drying process.
International Journal of Heat and Mass Transfer 2004;47:645–52.
doi:10.1016/j.ijheatmasstransfer.2003.08.013.
[33] Koukouch A, Idlimam A, Asbik M, Sarh B, Izrar B, Bostyn S, et al. Experimental
determination of the effective moisture diffusivity and activation energy during convective
solar drying of olive pomace waste. Renewable Energy 2017;101:565–74.
doi:10.1016/j.renene.2016.09.006.
[34] Kaya A, Aydın O, Dincer I. Comparison of experimental data with results of some drying
models for regularly shaped products. Heat and Mass Transfer 2010;46:555–62.
doi:10.1007/s00231-010-0600-z.
[35] Torki-Harchegani M, Ghanbarian D, Sadeghi M. Estimation of whole lemon mass transfer
parameters during hot air drying using different modelling methods. Heat and Mass
Transfer 2015;51:1121–9. doi:10.1007/s00231-014-1483-1.
[36] Azadbakht M, Vehedi Torshizi M, Aghili H, Ziaratban A. Application of Artificial Neural
Network (ANN) in Drying Kinetics Analysis for Potato Cubes. CARPATHIAN
JOURNAL OF FOOD SCIENCE AND TECHNOLOGY 2018;10(2):96–106.
[37] Azadbakht M, Vahedi torshii M , Ziaratban A, Ghajarjazi E. Application of Artificial
Neural Network (ANN) in predicting mechanical properties of canola stem under shear
loading. Agricultural Engineering International: CIGR Journal 2016;18:413-425.

[38] Khoshnevisan B, Rafiee Sh, Omid M. Prediction of environmental indices of Iran wheat
production using artificial neural networks. International Journal of Energy and
14
 ACCEPTED MANUSCRIPT

Environment 2013;4:339–48.

Figures caption:

Fig. 1. Diagram of microwave drying system

Fig. 2. Volume control of microwave system
Fig. 3. ANN schematic
Fig. 4. Effect of osmotic pre-treatment and microwave power on energy efficiency
Similar capital letters represent non-significance in the same power and similar small letters
represent the non-significance in the same osmosis time.
Fig. 5. Effect of microwave power on specific energy loss
The similar capital letters mean non-significant difference.
Fig. 6. Effect of osmotic pre-treatment and microwave power on exergy efficiency
The similar capital letters represent non-significance in the same power, and similar small letters
represent the non-significance in the same osmosis time
Fig. 7. Effect of osmotic pre-treatment and microwave power on specific exergy loss
The similar capital letters represent non-significance in the same power while the similar small
letters represent the non-significance in the same osmosis time.
Fig. 8. Effect of osmotic pre-treatment and microwave power on effective moisture diffusivity
Fig. 9. Effect of osmotic pre-treatment and microwave power on mass transfer coefficient
Fig. 10. Sensitivity coefficient for energy efficiency
Fig. 11. Sensitivity coefficient for specific energy loss
Fig. 12. Sensitivity coefficient for exergy efficiency
Fig. 13. Sensitivity coefficient for specific energy loss

Nomenclature
Cp Heat capacity (J/kg K)
Dem Effective moisture diffusivity (m2/s)
E Energy,(J)
Eloss Specific energy loss (J/kg water)
Eabs Exergy absorbed (J)
Ein Exergy input (J)
Eref Exergy reflected (J)

15
 ACCEPTED MANUSCRIPT

Etra Exergy transmitted (J)

EX Exergy (J)
ex Specific exergy (J/kg water)
EXabs Exergy absorbed (J)
EXin Exergy input (J)
exloss Specific exergy loss (J/kg water)
exexap Exergy of evaporation water (J/kg water)
EXref Exergy reflected (J)
EXtra Exergy transmitted (J)
hm Mass transfer coefficient (m/s)
m Mass (kg)
MR moisture ratio
m0 Initial mass of sample (kg)
md Mass of dry sample (kg)
mwt Mass of water evaporated (kg)
MC Moisture content (kg water/kg dry matter)
M0 Initial moisture content (kg water/kg dry matter)
MR Moisture ratio (dimensionless)
Mt Moisture content at any time (kg water/kgdry matter)
Me Equilibrium moisture content (kg water/kg dry matter)
P Microwave power (W)
Pabs Microwave power absorbed (W)
Pin Microwave power emitted by the magnetron
Pref Microwave power reflected (W)
Ptra Microwave power transmitted (W)
t Time (s)
T Temperature (K)
T0 Ambient temperature (K)
Ws Sample dry solid, g
W W Sample total weight, g
ηen Energy efficiency (%)
ηex Exergy efficiency (%)
λk Latent heat of sample (J/kg)
λwf Latent heat of free water (J/kg)
Subscripts
in Input
out Output
dp Dry product
en Energy
exe Exergy
abs absorbed
ref
tra transmitted
wp Wet product
W evaporate Water

16
 ACCEPTED MANUSCRIPT

Figures:

Fig. 1. Diagram of microwave drying system

Fig. 2. Volume control of microwave system

17
 ACCEPTED MANUSCRIPT

Fig. 3. ANN schematic

Energy efficiency(%)

50 aA
45 90W 360W 900W
40 aAB
35 bA
30 aB
bB
25
bC cA
20
15 cA cA
10
5
0
30min 60min 90min
Osmotic time(min)

Fig. 4. Effect of osmotic pre-treatment and microwave power on energy efficiency

Similar capital letters represent non-significance in the same power and similar small letters
represent the non-significance in the same osmosis time.

18
 ACCEPTED MANUSCRIPT

Specific energy loss(MJ)

10
9
8 A

7 B
6
5
4
3
2 C
1
0
90 Microwave Power(W)
360 900

Fig. 5. Effect of microwave power on specific energy loss

The similar capital letters mean non-significant difference.
Exergy efficiency(%)

90W 360W 900W

40
35
30
aB bA aA
25 aB
bB
20 bB
15 cAB cA
10 cB
5
0
30min 60min 90min
Osmotic time(min)

Fig. 6. Effect of osmotic pre-treatment and microwave power on exergy efficiency

The similar capital letters represent non-significance in the same power, and similar small letters
represent the non-significance in the same osmosis time

19
 ACCEPTED MANUSCRIPT

Specific exergy loss(MJ)

90 360 900

25
aA
20 aAB

15
aA
bB bA
10 aB
cA
cA aB
5

0
30 60 90
Osmotic time(min)

Fig. 7. Effect of osmotic pre-treatment and microwave power on specific exergy loss
The similar capital letters represent non-significance in the same power while the similar small
letters represent the non-significance in the same osmosis time.

Osmotic Time(min)
30 60 90
Effective moisture diffusivity

0.0006 Aa
0.0005
0.0004 aB
0.0003 aB
bA bA bA
0.0002 bB cB
0.0001 cC
0
90 360 900
Microwave power(W)

Fig. 8. Effect of osmotic pre-treatment and microwave power on effective moisture diffusivity

20
 ACCEPTED MANUSCRIPT

Osmotic Time(min)
mass transfer coefficient (m/s) 30 60 90

0.035
Aa
0.03
0.025 bA
aB
aB
0.02 bA aA aB
cA bB
0.015
0.01
0.005
0
90 360 900
Microwave power(W)

Fig. 9. Effect of osmotic pre-treatment and microwave power on mass transfer coefficient

8 Microwave power

7 Time osmotic
6
5
Sensitivity

4
3
2
1
0
8 6 4
Neuron in hidden layer

Fig. 10. Sensitivity coefficient for energy efficiency

21
 ACCEPTED MANUSCRIPT

0.9 Microwave power

0.8 Time osmotic

0.7
0.6
0.5
Sensitivity

0.4
0.3
0.2
0.1
0
8 6 4
Neuron in hidden layer

Fig. 11. Sensitivity coefficient for specific energy loss

12 Microwave power

Time osmotic
10

8
Sensitivity

0
8 6 4
Neuron in hidden layer

Fig. 12. Sensitivity coefficient for exergy efficiency

22
 ACCEPTED MANUSCRIPT

8 Microwave power

7 Time osmotic

6
5
Sensitivity

4
3
2
1
0
8 6 4
Neuron in hidden layer

Figure 13. Sensitivity coefficient for Specific energy loss

Table 1: Neural Network Relationships

Formula
Formula Reference
Number

𝑥 ‒𝑥
𝑒 ‒𝑒
Tanh = (25) [36]
𝑥 ‒𝑥
𝑒 +𝑒

∑𝑛 (𝑃𝑖 ‒ 𝑂𝑖)
2
𝑖=1
2 R2 = 1- (26) [37]
(𝑃𝑖 ‒ 𝑂)

23
 ACCEPTED MANUSCRIPT

∑𝑛 (𝑃𝑖 ‒ 𝑂𝑖)
2
𝑖=1
1‒ 2
r= (27)
(𝑃𝑖 ‒ 𝑂)

2
(𝑃𝑖 ‒ 𝑂𝑖)
RMSE = ∑𝑛 (28) [32]
𝑖=1 𝑛

∑𝑛 |𝑃𝑖 ‒ 𝑂𝑖|
𝑖=1
MAE = (29) [38]
𝑛

Tables:

Table 2: Optimization values for artificial neural network parameters

The number
Number of of
Type of activation Testing Validating Training
hidden Learning rule hidden
function data % data % data %
layers layer
neurons
Levenberg
1 Hyperbolic tangent 4 15% 10% 75%
Marquardt
Levenberg
1 Hyperbolic tangent 6 15% 10% 75%
Marquardt
Levenberg
1 Hyperbolic tangent 8 15% 10% 75%
Marquardt

Table 4: Regression equations of power (90, 360 and 900 W) for Specific exergy loss
Microwave Power Equating R²

24
 ACCEPTED MANUSCRIPT

90 y = 6.1096x0.9259 0.548
360 y = 9.3684x-0.104 0.9379
900 y = 15.882e0.3322x 0.9901

Table 5: Error values in predicting experimental data using optimal artificial neural network

4 neuron in hidden 6 neuron in hidden 8 neuron in hidden

Error
layer layer layer

Energy efficiency
Mean squared error 40.7364295 4.010991123 2.988120859
Normalized Mean squared
error 0.387597681 0.022064854 0.119134058
Mean absolute error 4.032143844 1.417448017 1.489954289
Correlation coefficient 0.831286093 0.999303134 0.977049886
Specific energy loss
Mean squared error 4.337581878 0.025695416 0.866819665
Normalized Mean squared
error 39.32863229 0.028049053 0.221324525
Mean absolute error 1.618505148 0.106840226 0.658209612
Correlation coefficient 0.417441698 0.991636439 0.94131972
Exergy efficiency
Mean squared error 2.861528643 2.041341749 5.169678018
Normalized Mean squared
error 0.063233888 0.792429787 0.264937076
Mean absolute error 1.56647252 1.255905643 1.989219103
Correlation coefficient 0.972033224 0.495847176 0.96768871
Specific exergy loss
Mean squared error 29.67488181 0.010963853 29.00512733
Normalized Mean squared
error 0.636353366 0.000271428 0.812034038
Mean absolute error 2.754316991 0.07534787 2.748302448
Correlation coefficient 0.77300422 0.999934906 0.750517268

Table6: Some of the best MLP neural network topologies to predict test values

Energy efficiency
3 neuron in hidden layer 5 neuron in hidden layer 7 neuron in hidden layer
Cross Cross Cross
Training Training Training
Validation Validation Validation
Run # 1 1 1 2 1 3
Epoch # 38 33 21 5 27 6
25
 ACCEPTED MANUSCRIPT

Minimum
0.021512597 0.00194432 0.023753914 0.021932085 0.025436747 0.001191989
MSE
Final MSE 0.021512597 0.001970578 0.023753914 0.038460095 0.025436747 0.008555116
Specific energy loss
Run # 1 2 1 3 1
Epoch # 13 4 57 14 19
Minimum
0.029177034 0.004976503 0.041720161 0.001642759 0.026481779
MSE
Final MSE 0.029177034 0.104359341 0.041720161 0.01981361 0.026481779
Exergy efficiency
Run # 1 1 1 1 1 2
Epoch # 35 17 14 7 29 4
Minimum
0.001758156 0.000763416 0.002044987 0.000826504 0.001330489 0.003554999
MSE
Final MSE 0.001758156 0.009132225 0.002044987 0.007230043 0.001330489 0.007237963
Specific exergy loss
Run # 1 1 1 1 1 2
Epoch # 23 23 17 12 47 18
Minimum
0.00788163 2.63784E-05 0.032262786 0.000125721 0.008362538 0.000251749
MSE
Final MSE 0.00788163 2.63813E-05 0.032262786 0.000164817 0.008362538 0.000252867

26
 ACCEPTED MANUSCRIPT

 An increase in the power increases the energy and exergy efficiency drying.
 Osmosis pre-treatment has no significant effect on energy and exergy loss.
 Osmosis pre-treatment, leading to an increase in the energy and exergy efficiency.
 Increasing the time and power have effect on the amount of energy and exergy.
 The ANN designed has the ability to predict the energy and exergy items.