Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 39
Estimation of production efficiency in the extraction of essential oil
from orange peel using neural networks
Estimación de la eficiencia productiva en la extracción de aceite esencial a
partir de la cáscara de la naranja mediante redes neuronales
Sandra Elvira Fajardo Muñoz
1
; Anthony Josue Freire Castro
2
; Michael Isaac Mejía Garzón
3
Received: 28 / 07 / 2023 – Received in revised form: 12 / 09 / 2023 – Accepted: 06 / 11 / 2023
– Published: 19 / 03 / 2024
Research
Articles
X
Review
Articles
Essay
Articles
* Author for correspondence.
Abstract
In this work, a feedforward Artificial Neural Network (ANN) with 9 hidden layers and backpropagation (BP) training algorithms and Levenberg-
Marquardt (LM) weight adjustment algorithms were used for the prediction of oil extraction yield from the orange peel (Citrus sinensis), For training and
validation, data were used in the amount of load in grams as an input variable and the oil yield in percentage as an output variable, which were obtained
in the distillation technique by steam entrainment using the Clevenger trap. Different architectures were studied by varying the number of neurons in the
hidden layer, finding that the ANN with 9 neurons provided the best fit of the experimental data, which indicates greater efficiency and accuracy compared
to the other architectures analyzed. Regarding the experimental data, the percentage mean square error (MSE%) and the determination coefficient
,
were evaluated, finding for the ANN values of MSE%=0.0040 and
=0.9929, proving that the hypothesis research is true. These results show the
efficacy and potential of using neural networks for modeling and prediction of orange oil extraction performance within the domain of training data.
Keywords: Artificial neural networks, backpropagation algorithm, convergence, topology, extraction, essential oil.
Resumen
En este trabajo, se utilizó una Red Neuronal Artificial (RNA) feedforward con 9 capas ocultas y algoritmos de entrenamiento backpropagation (BP) y de
ajuste de pesos Levenberg- Marquardt (LM) para la predicción del rendimiento de extracción de aceite a partir de la cáscara de naranja (Citrus sinensis),
para el entrenamiento y validación, se emplearon los datos en cantidad de carga en gramos como variable de entrada y el rendimiento de aceite en
porcentaje como variable de salida, los cuales se obtuvieron en la técnica de destilación por arrastre de vapor usando la trampa Clevenger. Se estudiaron
distintas arquitecturas variando el número de neuronas en la capa oculta, encontrando que la RNA con 9 neuronas brindaba el mejor ajuste de los datos
experimentales, lo que indica mayor eficacia y exactitud frente a las otras arquitecturas analizadas. Con respecto a los datos experimentales, se evaluó el
error cuadrado medio porcentual (ECM%) y el coeficiente de determinación R^2, encontrándose para la RNA valores de ECM%=0.0040 y R^2=0,9929,
comprobando que la hipótesis de investigación es verdadera. Estos resultados muestran la eficacia y potencialidad del uso de las redes neuronales para el
modelado y predicción del rendimiento de extracción de aceite de naranja dentro del dominio de los datos de entrenamiento.
Palabras Claves: Redes neuronales artificiales, algoritmo backpropagation, convergencia, topología, extracción, aceite esencial.
1. Introduction
The essential oils industry has experienced tremendous
growth due to technological advances and industrial
revolutions. The juice and nectar market generates a large
amount of waste from fruit pulp, which represents a burden
on the environment. In the canton of Las Naves, province
of Bolívar, orange production is significant, and the waste
generated can be used to obtain value-added products, such
as essences, perfumes, shampoos, soaps, among others,
from the essential oils present in the peel.
In Chemical Engineering, it is important to develop
mathematical models to predict the efficiency of physical
or chemical separation processes. In this regard, artificial
neural networks (ANNs) are gaining popularity as
1
Universidad de Guayaquil; sandra.fajardom@ug.edu.ec ; Guayaquil; Ecuador.
2
Universidad de Guayaquil; anthony.freirec@ug.edu.ec ; Guayaquil; Ecuador.
3
Universidad de Guayaquil; michael.mejia@ug.edu.ec ; Guayaquil; Ecuador.
modeling tools due to their high capacity. Researchers are
exploring how to apply ANNs to develop new PID
controllers, inspired by the functioning of biological
neurons in terms of learning and memory. ANNs can learn
from examples and generalize to solve various problems,
even with incomplete or erroneous data.
In this research work, it is proposed to conduct a laboratory-
scale experiment to obtain essential oils from orange peel,
specifically from the species Citrus Sinensis L. In addition,
the programming and mathematical calculation tool Matlab
will be used to develop an ANN model that will allow a
more accurate prediction of the efficiency of the process.
Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 40
In recent years, the growth of technological activity has
been alarming, as human beings need resources to satisfy
their needs and desires. Technology refers to the set of
technical and scientific knowledge that enables the creation
and design of objects and services to satisfy human needs.
In this context, the concept of network arises, which refers
to a set of interconnected entities that allow the flow of
material and non-material elements between their
connection points [1].
In search of improvements, humans have shown interest in
understanding the functions of the brain and have
developed technological tools to emulate its functions. The
brain is an information processor with complex and special
characteristics. Its main function is to process large
amounts of sensory information immediately, combine and
compare it with stored information, and respond
appropriately to new situations [2].
.
In this study, an artificial neural network model is
developed to mimic the information processing capabilities
of the brain. Conventional computers are limited in their
ability to interact with complex data and variable
environments, which makes neural networks useful for
solving problems where traditional algorithms are not
effective. These models can be applied to unit operations in
Chemical Engineering, such as the extraction of essential
oils from orange waste [3].
In Ecuador, large amounts of waste are generated,
including organic waste, and orange peel represents a
potential source of value-added products, such as essential
oils. These oils have applications in various industries, such
as pharmaceuticals, food, and cosmetics. The extraction of
essential oils is carried out by methods such as steam
entrainment distillation with the Clevenger trap [4].
In the food industry, the use of natural additives is
increasingly valued over synthetic additives due to health
concerns. Natural flavorings are especially appreciated as
they can enhance the sensory experience of foods. Essential
oil extraction methods allow obtaining natural fragrances
that can be used as aromatic additives [5].
In summary, this study focuses on the development of an
artificial neural network model to predict the effectiveness
of essential oil extraction from orange peel. The potential
of orange waste as a source of added value and the
importance of natural additives in the food industry are
highlighted. In addition, the extraction methods used, such
as steam distillation, are mentioned.
The orange tree (Citrus sinensis) is a fruit tree belonging to
the Rutaceae family. Its fruit is the sweet orange, of globose
or oval shape with a diameter of 6-9 cm. It has a slightly
rough orange rind and a pulp without oily vesicles, and its
seeds are white. The tree reaches a height of three to five
meters, with a rounded crown and regular branches. It has
a single straight and cylindrical trunk that changes color
from green to gray. The leaves are evergreen, of medium
size and elongated, with rounded base and ending in a
point. The flowers appear solitary or in clusters in the axils
of the leaves [6].
The orange tree is native to tropical and subtropical areas
of Asia and has spread throughout North Africa,
southeastern Europe, and the Americas due to its
introduction by Europeans in the 16th century. The flowers
of the orange tree are used to obtain essential oils that are
used in perfumery and also have medicinal applications.
In summary, the orange tree is a fruit tree with specific
characteristics, whose fruit is the sweet orange. Its
geographical distribution has expanded thanks to human
intervention, and the flowers of this tree have important
uses in the perfume and medicine industry [7].
Fig. 1. Oranges (Citrus sinensis) in the canton of Las Naves,
Bolivar province
Source: [8]
The level of carbohydrates in orange peel residues is
80.8%. According to the carbohydrates identified are
pectin’s 30-50%, sugars (sucrose, fructose, glucose),
hemicellulose, 10-20% and cellulose 20-40% [9].
Table 1. Physicochemical composition of orange peels
Main Components
(%)
Dry Matter
90,00
Protein
6,00
Carbohydrates
62,70
Fats
3,40
Fiber
13,00
Ash
6,90
Minerals
(%)
Calcium
2,00
Magnesium
0,16
Phosphorus
0,10
Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 41
Vitamins
(mg/Kg)
Niacin
22,00
Riboflavin
22,20
Amino acids
(%)
Arginine
0,28
Lysine
0,20
Tryptophan
0,06
Source: [10]
Table 1 shows the physicochemical composition of the
orange peel, in which it is analyzed that the main
components, such as dry matter, protein, carbohydrates,
fiber and ash are found in greater proportion and those
called traces, such as certain minerals, vitamins and amino
acids in smaller proportion. These experimental data will
help us to know the yield at the moment of extracting the
essential oil.
1.1. Industrial uses and applications
Orange has several industrial uses due to its beneficial
properties. It reduces low-level cholesterol and possesses
bioflavonoids with anticarcinogenic properties that help
prevent breast and colon cancer. Orange peel contains
vesicles with essential oils that provide characteristic
aromas and act as a defense against pests [11].
1.2. Essential oils
Orange essential oil is widely used in the manufacture of
products for human consumption. Its fungicidal
characteristics make it useful in the manufacture of insect
repellents and pesticides. It is also used in the manufacture
of soft drinks, syrups, vitamin complexes, perfumes, eau de
cologne, soaps, and other products [12].
Essential oils are volatile organic compounds obtained
from plants, bacteria, or fungi. They are used in cosmetics,
food, pharmaceuticals, and other industrial processes that
require aromas and essences. They have various biological
properties, such as antioxidant, anti-inflammatory,
antimicrobial, anticancer and lipid-lowering properties.
They can be extracted by methods such as distillation, cold
pressing, hydro-diffusion, supercritical fluids, and
microwave radiation [13].
Essential oils are complex mixtures of more than 100
different components, including aliphatic compounds,
phenylpropanes, monoterpenes and sesquiterpenes.
In summary, orange has industrial uses due to its beneficial
properties, especially in the production of essential oils.
These oils are used in various industrial sectors and have
important biological properties [14].
1.3. Methods of Extraction of Essential Oils
According to Fennema [15], , it is important to define the
extraction method as this will directly influence the quality
and quantity of the essential oil obtained.
Kirk Donald and Othmer, mentioned by Guevara [16], state
that there are a great number of techniques where the
extraction of the essences from the raw materials that
contain them is achieved. Their choice will depend on
characteristics such as:
a) Characteristics of the raw material.
b) Volatility of the essence.
c) The percentage of essence in the plant.
d) The purity and quality characteristics to be obtained.
1.4. Artificial Neuron
The artificial neuron was designed to "emulate" the basic
operating characteristics of the biological neuron. In
essence, a set of inputs is applied to the neuron, each of
which represents an output of another neuron. Each input is
multiplied by its corresponding "weight" or weighting
analogous to the degree of connection of the synapse. All
weighted inputs are summed, and the level of excitation or
activation of the neuron is determined [17]. A vector
representation of the basic functioning of an artificial
neuron is given by the following equation
( 1)
Where:
•
•
•
Fig. 2. Artificial neuron model
Source: [18]
The most commonly used activation functions are the
Sigmoid and Hyperbolic Tangent function expressed in
Table (3).
Table 2. Activation functions
Sigmoid
Hyperbolic Tangent
Source: Xabier Basogain Olabe, Artificial Neural
Networks and their Applications, [19]
Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 42
In Table 2, we can observe the most used F functions, are
the Sigmoid and Hyperbolic Tangent function since these
return an output that will be generated by the neuron given
an input or set of inputs, i.e. each of the layers that make up
the neural network have an activation function that will
allow to reconstruct or predict.
This type of artificial neuron model ignores many of the
characteristics of biological neurons. Among them is the
omission of delays and synchronism in the generation of
the output. However, despite these limitations, the
networks constructed with this type of artificial neuron
present qualities and attributes with some similarity to
those of biological systems [20].
1.5. Structure of Artificial Neural Networks (ANN)
Artificial neural systems mimic the hardware structure of
the nervous system. Each neuron performs a mathematical
function. Neurons are grouped in layers, constituting a
neural network. A given neural network is tailored and
trained to perform a specific task. Finally, one or more
networks, plus interfaces with the environment, make up
the overall system [21].
In biological neural networks, neurons correspond to the
processing elements. Interconnections are made by output
branches (axons) that produce a variable number of
connections (synapses) with other neurons or with other
parts such as muscles and glands. Neural networks are
systems of simple, highly interconnected processing
elements.
Fig. 3. Hierarchical structure of a system based on artificial
neural networks.
Source: [22]
Figure 3 shows the complexity of a neural system, since the
output obtained from the network is the result of abundant
feedback loops along with nonlinearities of the processing
elements and adaptive changes of its parameters. [23].
Formally, a neural or connectionist system is composed of
the following elements:
• A set of elementary processors or artificial neurons.
• A connectivity pattern or architecture.
• A dynamic of activations.
• A learning rule or dynamic.
• The environment in which it operates.
2. Materials y methods
The Citrus Sinensis orange tree was obtained in the
province of Bolivar, canton Las Naves, 88 km northwest of
Guaranda. This region of the country has a beneficial
tropical forest for the development of this species. Three
sacks of this citrus material were collected and immediately
transferred to the city of Guayaquil, Guayas province, for
reconditioning and subsequent extraction of the essential
oil.
Fig. 4. Raw material reception.
2.1. Characterization of the orange species Citrus
sinensis.
1. Degree of maturity: The pinto fruit was
evaluated, determining the maturity index, which is
expressed by the °Brix/valuable acidity ratio. This indicates
that as the °Brix/valuable acidity ratio increases, ripening
advances directly [24].
2. Biometric determination: Ten medium pintonas
oranges were taken at random, and we proceeded to
calculate the percentage of peel and pulp, it was determined
by weighing separately the pulp from the peel, and using
the following formulas, we obtain the %peel and %pulp.
( 2)
( 3)
Where:
Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 43
2.2. Extraction by steam distillation using the Clevenger
Trap
The peel, once weighed with 200 g, 300 g, 350 g, 400 g and
500 g loads, was fed to the equipment, placing 1,250 liters
of distilled water in the container. The steam that is
generated drags all the volatile components on the surface
of the peel and then, with the help of the refrigerant, it
condenses into a mixture of water and essential oil. The
time established was 30 minutes of distillation once the first
drop of condensate has fallen.
2.3. Experimental Design
Fig. 5. Experimental design for the extraction of essential
oil from orange peel by the steam distillation method using
the Clevenger trap
2.4. Statistical test
Since for each initial load of peel a minimum of 5
extractions are going to be performed, a small sample size
is considered since , therefore, we will use the t-
statistic to construct the confidence intervals.
2.5. Construction of the predictive mathematical models
To model the relationship between the amount of initial
charge and the yield of the essential oil extraction process,
we proceeded to make a scatter diagram to analyze what
type of relationship exists between the 2 variables.
With the Matlab program, several trend lines were
reviewed, such as: linear, a polynomial of degree 2 and a
logarithmic one, where their predictive capacity was
subsequently evaluated with the verification load of 350 g.
2.6. Construction of the artificial neural network (ANN)
A multilayer perceptron network was implemented to
predict the yield of the orange peel essential oil extraction
process as a function of the amount of initial load. Two
important factors were considered to ensure the success of
the neural model: the number of hidden layers and the
number of neurons per layer.
A hidden layer and an output layer were used in the
network, with a sigmoidal activation function for the
hidden layer and a linear function for the output layer.
Several networks were constructed by varying the number
of neurons in the hidden layer, between 2 and 10, to find
the best architecture. The backpropagation algorithm was
used for supervised training, due to its fast tuning and easy
application. The Levenberg-Marquardt algorithm was used
to adjust the weights of the connections between the
neurons in each laye.
Training and validation were performed using different
amounts of load (200g, 300g, 400g and 500g) as the input
variable, and average performances in percentage as the
output variable. Seventy percent of the data was used for
training, while the remaining 30% was used for validation.
Training continued until the error in the validation data
reached a minimum value. To evaluate the predictive
capability of the network, a midpoint within the
experimental range was used, in this case, 350g.
The effectiveness of the neural network was evaluated
using two indicators: the percent mean square error
(MSE%) and the quadratic coefficient of determination
(R^2). These indicators were calculated using specific
expressions and allow measuring the accuracy of the
network in estimating process performance values on test
data.
In summary, a neural network was used to predict the yield
of the orange peel essential oil extraction process. The
parameters of the network were adjusted, and precision
indicators were used to evaluate its predictive capacity. The
results obtained allow more accurate estimation of the
process yields as a function of the amount of initial charge.
(4)
(5)
Where
represents the experimental value,
represents the value predicted by the network,
the
value of the average response and N the total number of
data. In turn these two parameters are important to buy the
performance of the different proposed ANN architectures
to estimate which one is the best.
3. Results
Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 44
Table 3, how’s the maturity indicator of the orange that was
selected.
Table 3. Indicator of the degree of maturity of the orange
species Citrus sinensis
Color Fruit
pH
°Brix (%)
A.T(%)
I.M
Pintona
(yellowish)
4,84
1,9
0,81
2,35
A.T= Titratable acidity
I.M= Maturity Index
Source: Own elaboration
The ratio between °Brix/titratable acidity is called the
maturity index, and for the development of this work, the
maturity index of the Pinotona orange was 2.35.
3.1. Statistical analysis.
The behavior of the yields of essential oil extraction (%),
by the steam distillation technique, with amounts of load
(200, 300, 400, 400, 500) grams, can be seen in Table 4.
Table 4. Averages of essential oil %yield as a function of
loading (g)
Size (
Loading amounts (g)
200
300
400
500
1,5
0,0625
0,0971
0,1473
0,1674
As shown in Figure 6, for a given shell size, the greater the
amount of load introduced into the system, the higher the
yield of the process, with an average value of 0.1674% for
a 500 g load.
Fig. 6. Bar chart of %Yield Vs Amount of husk
3.2. Mathematical model using the curve fitting
technique.
Figure 7 shows a scatter diagram of the yields obtained in
the experiment as a function of the amount of husk (g) used,
which will help us to determine what type of relationship
exists between these two variables. Figure 9 shows that we
proceeded to analyze the data with three mathematical
models: linear, polynomial of degree 2 and logarithmic,
together with their coefficient of determination (
Fig. 7. Scatter plot of (%) Yield vs Amount of shell (g)
Source: Own elaboration
Fig. 8. it of the experimental data with different
mathematical models and presentation of their R^2
Table 5 shows each of the equations of the 3 mathematical
models presented in the previous figure with their
respective R^2 , and we can see that the polynomial
adjustment of degree 2 has the highest coefficient of
determination compared to the other 2 adjustments
performed.
Table 5. Redictive comparison of the performance of the
different adjustments performed.
Fitting
Equation
%Yield
predicted
with a load
of 350 g
Linear
%Yield=0,0004*(load (g))-
0,0096
0,9792
0,1304
Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 45
Logarithmic
% Yield=0,1196*ln (load
(g))-0,5755
0,9795
0,1251
Polynomial
% Yield=-3
+0,0006*(load
(g))-0,0477
0,9862
0,1255
However, Table 6 shows the error between the
experimental value of the yield and the value predicted by
each of the mathematical models with the 350 g load. The
average experimental value of the yield for this load was
0.1158%, and we can see that the logarithmic model is the
best since it presents a lower error, although its
is the
second best.
Table 6. Comparison between the different models by
calculating the error with the experimental value obtained
Adjustment
%Predicted yield with a
350 g charge
Error (%)
Linear
0,9792
0,1304
12,6079
Logarithmic
0,9795
0,1251
8,0310
Polynomial
0,9862
0,1255
8,3765
Source: Own elaboration
3.3. Artificial neural network model.
Table 7 shows the optimal number of neurons in the hidden
layer, which was determined by a trial-and-error process,
minimizing the difference between the experimental values
and those predicted by the network at the verification point.
The ECM% and
values of each of the predictions made
with different numbers of neurons in the hidden layer are
also shown.
Table 7. Efficiency predictive capacity with different
numbers of neurons in the hidden layer.
Number of
neurons in the
hidden layer
Predicted
performance
value
EMC (%)
2
0,0990
0,8522
0,0314
3
0,0941
0,5165
0,0503
4
0,1070
0,9452
0,0110
5
0,1634
0,7618
0,2295
6
0,0983
0,9528
0,0339
7
0,1170
0,9574
0,0034
8
0,1599
0,9806
0,1975
9
0,1187
0,9929
0,0040
10
0,1217
0,9984
0,0067
ANN with a 9-neuron architecture in the hidden layer was
found to provide the best prediction.
Fig. 9. Schematic diagram of the optimal ANN model
Source: Own elaboration
3.4. Comparison of the values predicted by the
logarithmic mathematical model, ANN and
experimental.
Figure 10 shows the essential oil extraction yield data as a
function of the amount of orange peel, the fitting curve of
the mathematical model and the values predicted by ANN
(artificial neural network).
Fig. 10. Experimental values of Yield (%), logarithmic
model and ANN
Source: Own elaboration
In addition, it can be seen that the data predicted by the
artificial neural network present a high correspondence
with the fitting curve provided by the mathematical model,
showing the capacity of the network to capture linear and
nonlinear interactions associated with the extraction
process.
And as shown in Table 8, the neural network presents a high
predictive capacity at the verification point compared to the
other mathematical models.
Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 46
Table 8. Demonstration of the predictive ability of the
neural network against the other models.
Model
%Rendimiento
predicho con una
carga de 350 g
Error
(%)
Linear
0,9792
0,1304
12,6079
Logarithmic
0,9795
0,1251
8,0310
Polynomial
0,9862
0,1255
8,3765
Artificial Neural
Network
0,9929
0,1187
2,4431
4. Conclusions
The characterization of orange peel was achieved in the
laboratory accredited by ISO 17025 Analytical
Laboratories in its most important physical-chemical
parameters such as degree of maturity, humidity, ash
content, reducing sugars. The results obtained were of great
importance since the extraction process was developed with
this type of raw material, and the ANN predictive model
will only work for extractions carried out with this type of
peel under the physical-chemical conditions established in
this work.
The steam distillation technique using the Clevenger trap
yielded an average extraction yield of 0.1680% for the 500
g initial charge. Therefore, it is within the range of 0.5-0.8%
yields with this type of raw material using this extraction
method as they may oscillate depending on the variety, fruit
maturity stage and extraction method used. The optimal
ANN model developed with Matlab R2019a software was
the one that had 9 neurons in the hidden layer, had as input
variable the amount of cargo in grams and as output
variable the yield in percentage; this model showed higher
accuracy against the other analyzed architectures. This can
be demonstrated by the parameter values of the coefficient
of determination,
and the percent root mean
square error, , therefore, the research
hypothesis was proven to be true.
It was possible to observe the predictive capacity of the
network to relate the variables, demonstrating its potential
of artificial intelligence for the modeling and prediction of
physical processes, as could be observed in the application
that was given in the present work, its results were better
and closer to the experimental ones compared to the
mathematical adjustments that are commonly performed in
the Chemical Engineering career to relate the different
process variables.
Referencias
[
1
]
M. Moreno, «Redes Neuronales Artificiales aplicadas al Análisis de
Datos,» tesisenred.net, pp. 17-31, 2002.
[
2
]
E. A. Heredia, «NARANJA ESTUDIO AGROINDUSTRIAL DEL
ECUADOR,» 4 Julio 2008. [En línea]. Available:
https://issuu.com/mipro/docs/naranja.
[
3
]
A. Badui, «APLICACIÓN DEL LECHO POR FUENTE-
FLUIDIZADO “SPOUT-FLUID BED” CON SÓLIDOS INERTES
A LA MICROENCAPSULACIÓN DE ACEITE ESENCIAL DE
NARANJA,» sappi.ipn.mx/, pp. 1-45, 2008.
[
4
]
L. Balboa, «Obtención Experimental de aceite esencial y
subproducto a partir de la cáscara de naranja (Citrus Sinensis),»
repositorio.umsa.bo, pp. 1-10, 2019.
[
5
]
J. M. Huayta, «EFECTO HIPOLIPEMIANTE DEL ACEITE
ESENCIAL DE CÁSCARA DE NARANJA (Citrus sinensis) EN
RATTUS NORVEGICUS VARIEDAD WISTAR,»
repositorio.unsa.edu.pe, pp. 1-25, 2017.
[
6
]
P. A. Reyes-Ortiz y Colaboradores, «1 PRODUCCIÓN DE
NARANJA (Citrus sinensis L. Osbeck) VARIEDAD VALENCIA
TARDÍA BAJO MANEJO ORGÁNICO EN SAN PABLO,
PAPANTLA DE OLARTE; VERACRUZ.,» 13 Junio 2017. [En
línea]. Available:
https://www.researchgate.net/publication/320110143_PRODUCCI
ON_DE_NARANJA_Citrus_sinensis_L_Osbeck_VARIEDAD_V
ALENCIA.
[
7
]
A. Contreras-Oliva y Colaboradores, «CALIDAD
FISICOQUÍMICA, SENSORIAL Y NUTRICIONAL DE
NARANJAS CV. VALENCIA RECUBIERTAS CON
QUITOSANO,» 15 Agosto 2012. [En línea]. Available:
http://www.scielo.org.mx/pdf/agro/v46n5/v46n5a2.pdf.
[
8
]
Mejía y Freire, «La naranja está barata,» El Diario Comercio
Ecuador, vol. 1, nº 1, p. 1, 2021.
[
9
]
A. C. Trejo y A. C. Krauser, «Química de los carbohidratos,» 16
Enero 2013. [En línea]. Available:
https://www.researchgate.net/publication/286640492_Quimica_de_
los_Carbohidratos.
[
1
0
]
A. Germán, «Producción de enzimas pectinasas por actinomycetos
en cultivo sumergido utilizando pectina y cáscara de naranja,»
sisbib.unmsm.edu.pe, pp. 4-6, 2014.
[
1
1
]
A. E. Q. Valle, «UTILIZACIÓN DE RESIDUOS DE CÁSCARA
DE NARANJA PARA LA PREPARACIÓN DE UN
DESENGRASANTE DOMÉSTICO E INDUSTRIAL,»
repositorio.uisek.edu.ec, pp. 13-22, 2009.
[
1
2
]
A. Castañeda, «Actividad antimicrobiana del aceite de naranja
residual,» Scielo, pp. 1-6, 2018.
[
1
3
]
L. Mancilla, «Estudio del aceite esencial de la cáscara de la naranja
dulce (Citrus sinensis, variedad Valenciana) cultivada en Labateca
(Norte de Santander, Colombia),» Redalyc.org, pp. 4-7, 21 Febrero
2007.
[
1
4
]
D. Nuñez, «OBTENCIÓN DE ACEITES ESENCIALES A PARTIR
DE CORTEZA DE NARANJA "CITRUS SINENSIS" VARIEDAD
VALENCIANA,» Researchgate.net, pp. 6-13, Julio 2019.
[
1
5
]
0. Fennema, «Introducción a la Química de los Alimentos,»
ceqa.files.wordpress., pp. 504-798, 2014.
[
1
O. Guevara, «Obtención de aceite esencial de citronela
(Cymbopogon winterianus) extraído por arrastre con vapor a escala
Universidad de
Guayaquil
Ingeniería Química y Desarrollo
https://revistas.ug.edu.ec/index.php/iqd
ISSN – p: 1390 –9428 / ISSN – e: 3028-8533 / INQUIDE / Vol. 06 / Nº 01
Facultad de
Ingeniería Química
Ingeniería Química y Desarrollo
Universidad de Guayaquil | Facultad de Ingeniería Química | Telf. +593 4229 2949 | Guayaquil – Ecuador
https://revistas.ug.edu.ec/index.php/iqd
Email: inquide@ug.edu.ec | francisco.duquea@ug.edu.ec
Pag. 47
6
]
piloto: estudio de la influencia de variables en el rendimiento y la
calidad del aceite,» ria.utn.edu.ar, pp. 62-82, Marzo 2014.
[
1
7
]
J. T. Barrera, «Redes Neuronales,» cucei.udg.mx, pp. 3-6, 2012.
[
1
8
]
J. Marshall, «Redes Neuronales,» avellano.fis.usal.es, pp. 1-3, 2013.
[
1
9
]
X. B. Olabe, «Redes Neuronalesa Artificiales y sus Aplicaciones,»
ocw.ehu.eus, pp. 18-20, 2016.
[
2
0
]
E. C. A. Tepán, «Estudio de los Principales tipos de Redes
Neuronales y las Herramientas para su aplicación,»
dspace.ups.edu.ec, pp. 15-16, Febrero 2013.
[
2
1
]
H. G. Asensio, «Inteligencia artificial. Redes neuronales y
aplicaciones,» it.uc3m.es, pp. 2-6, 2011.
[
2
2
]
V. B. Pardo, «Machine Learning con TensorFlow,» oa.upm.es, pp.
1-16, 15 Enero 2017.
[
2
3
]
L. O. Toledo, «Red neuronal artificial para estimar un índice de
calidad de agua,» Scielo, pp. 1-3, Junio 2020.
[
2
4
]
P. Yufera, «Química Agrícola,» de El suelo y los elementos químicos
esenciales para la vida vegetal, Madrid, Mundi-Prensa, 1979, pp.
373-396.
