Drying Technology

An International Journal

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ldrt20

Application of machine learning-based approach

in food drying: opportunities and challenges

Md. Imran H. Khan , Shyam S. Sablani , M. U. H. Joardder & M. A. Karim

To cite this article: Md. Imran H. Khan , Shyam S. Sablani , M. U. H. Joardder & M. A. Karim
(2020): Application of machine learning-based approach in food drying: opportunities and
challenges, Drying Technology, DOI: 10.1080/07373937.2020.1853152

To link to this article: https://doi.org/10.1080/07373937.2020.1853152

Published online: 08 Dec 2020.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ldrt20
DRYING TECHNOLOGY
https://doi.org/10.1080/07373937.2020.1853152

Application of machine learning-based approach in food drying:

opportunities and challenges
Md. Imran H. Khana,b, Shyam S. Sablanic, M. U. H. Joardderd, and M. A. Karima
a
School of Mechanical, Medical and Process Engineering, Queensland University of Technology (QUT), Brisbane, Queensland, Australia;
b
Department of Mechanical Engineering, Dhaka University of Engineering & Technology, Gazipur, Bangladesh; cDepartment of
Biological Systems Engineering, Washington State University, Pullman, Washington, USA; dDepartment of Mechanical Engineering,
Rajshahi University of Engineering & Technology, Rajshahi, Bangladesh

ABSTRACT ARTICLE HISTORY

Application of machine learning (ML)-based algorithms in food drying is an exciting and Received 12 November 2020
innovative approach to advance the drying technology. In order to appropriately develop Accepted 15 November 2020
this novel approach in all aspects of food drying field, significant scientific research is
KEYWORDS
required. The main aspects of food drying research are the determination of material prop-
Food drying; Artificial
erties, microstructural characterization, mathematical modeling, and process optimization. It Neural Network; modeling;
is essential to express this fundamental information through ML-based algorithms to food properties; food
advance the food drying research. This paper aims to present a comprehensive review of microstructure
the application of machine learning-based approaches in food drying modeling, property
prediction, microstructural characterization, and process parameters optimization. Moreover,
this paper discusses the possibilities and challenges to apply ML-based algorithms in multi-
scale modeling and microwave-based hybrid drying. It is expected that this review paper
will be beneficial in advancing the machine learning-based food drying technology.

1. Introduction advantage of drying is that it is an excellent way to

preserve foods that can add variety to meals and pro-
Food security is a major concern in large parts of the
vide delicious, nutritious snacks.
world as about one-third of the global food produc-
Most widely used drying method is convection or
tion, or 1.3 billion tonnes of food is lost annually due
hot air drying, which is the easiest way of drying.
to the lack of proper processing.[1] Food waste means
However, longer drying time, lower energy efficiency
not only loss of food, but it also means a waste of
and products poor-quality attributes are associated
resources used in food production such as land, water,
with convective drying.[4] These problems are mainly
energy, and labor inputs. Moreover, wasted foods con- due to the uncontrolled drying processes and
tribute to massive carbon emissions that are prior improper product development strategies. Currently,
responsible for today’s global warming problem. the agri-based dried food products are developed and
Therefore, proper food processing must be empha- drying systems are designed either by trial and error
sized to reduce this massive loss, promote food secur- or based on limited empirical knowledge, without
ity, reduce global warming, and combat hunger. considering the underlying micro level fundamental
Drying is a method of food preservation that inhibits drying mechanisms. Therefore, the fundamental
the growth of bacteria, yeasts, and mold through the understanding of the drying process and product
removal of water.[2,3] Moreover, dried foods have development is highly required. The development of
gained commercial importance and their growth on a fundamental physics-based models of food drying is
commercial scale has become an important sector of not trivial as it requires physical as well as mathemat-
the agricultural industry.[2] One of the important ical insight into the material and the process.
advantages of dried foods is that they take much less Development of an accurate physics-based model is
storage space than canned or frozen foods and trans- complex and requires significant scientific research
portation cost is significantly reduced. Another efforts. Various fundamental information such as the

CONTACT Imran H. Khan [email protected], [email protected] School of Mechanical, Medical and Process Engineering, Queensland
University of Technology (QUT), 2 George St, Brisbane, QLD 4000, Australia
ß 2020 Taylor & Francis Group, LLC
2 M. I. H. KHAN ET AL.

structural configuration of a diverse range of food Drying research involves mathematical modeling,
materials and their various properties and their rela- properties investigations, microstructural analysis etc.
tionships with moisture content and temperature dis- The current review papers were not focused on these
tributions are required to develop such physics-based various aspects of drying with ML strategies. For
models. Moreover, the optimization of process param- instance, Martynenko and Misra[9] very recently
eters including drying time, level of water activity, developed a comprehensive review of machine learn-
drying air velocity, relative humidity, drying tempera- ing in drying. This papers mostly discussed the con-
ture, and sample geometry. is also required to control cept and mechanism of various machine learning-
the drying operation for ensuring better product qual- based algorithms and general application in drying.
ity.[5] To obtain this various important information Martynenko[18] presented a short review of the possi-
and develop a smart drying technology, world food bilities to apply artificial intelligence-based strategies
researchers have extensively worked for the last few in drying application. Recently, Sun et al.[10] published
decades and significant advancement has been a most useful review paper where a comprehensive
made.[6,7] However, most of the studies have been review of intelligent drying technologies and their
conducted either experimentally or using empirical applications. The paper outlines the basic theoretical
modeling methods.[8] Experimental methods are the knowledge of ANN, fuzzy logic and expert system
most time consuming, costly and complex approach with the possible application of artificial intelligence
and need a high level of sophisticated and expensive in modeling and optimization of heat and
experimental facilities. Moreover, it is quite infeasible mass transfer.
to perform a series of experiments for a large range of Although the aforementioned review paper pro-
food products. To overcome such challenges, machine vided useful information for machine learning appli-
learning (ML) based predictive approach has been cation in drying technology, a comprehensive review
introduced in food processing. ML-based approach paper particularly focusing on machine learning appli-
such as Artificial Neural Network (ANN) is an excel- cations in various aspects of food drying such as
lent platform for biological predictive modeling as this properties prediction, structural characterization, opti-
type of model can perfectly handle the problem which mization and mathematical modeling is not available.
involves massive combinatorial spaces or nonlinear Therefore, this research aims to present a comprehen-
processes and is not easily amenable to traditional sive review of different strategies of machine learning
approaches for a solution.[9] ANN has the unique application in properties prediction, microstructural
intelligence to characterize the fruits and vegetables characterization, and process parameters optimization
(FV) and predict the various properties based on during drying. Moreover, this paper provides valuable
some specific input parameters or training. Due to the information on ML-based approaches for mathemat-
nondestructiveness, simplicity and real-time monitor- ical modeling, and their challenges to address multi-
ing, the ML approaches are increasingly under consid- scale food drying phenomena.
eration by food engineers and scientists for
application as a superior tool to model complex,
2. Machine learning-based strategies
dynamic, highly nonlinear, and ill-defined scientific
and engineering problems in food drying.[10,11] Many A machine learning algorithm, also known as a
researchers attempted to use this novel approach in model, is a kind of mathematical expression of a
many areas of food drying such as energy and exergy problem which characterizes the various linear or
prediction,[12,13] predicting physicochemical proper- nonlinear data. Based on the input data, the ML
ties[14] and drying modeling.[15] To accurately apply model provides predictive values. Machine learning
this novel approach in all aspects of the drying field, algorithm has the superior capacity to handle the large
significant scientific research is still required. Review volumes of data and predict the specific trends and
papers are highly important to clarify the state of patterns. Among the various machine learning strat-
knowledge, explain apparent contradictions, identify egies, the most popular machine learnings are
needed research, and even create a consensus where Regression, Classification, Clustering, Ensemble
none existed before. Therefore, some researchers Methods, Transfer Learning, Reinforcement Learning,
attempted to develop a comprehensive review of a dif- Natural Language Processing and Deep Learning
ferent aspect of ML application in drying.[7,9,10,16,17] algorithms used in the engineering field. The learn-
However, most of the papers are more generic and ing-based methods are divided into three groups:
not focused on a particular application in food drying. supervised, semi-supervised, and unsupervised
 DRYING TECHNOLOGY 3

Figure 1. Various ANN-based algorithms (a) A feed-forward and backpropagation network; (b) recurrent network; (c) mixed
network; and (d) interconnected network (Adapted from Ref.[10]).
learning. In a supervised learning algorithm, the are mathematically designated by the notations Xn,
known data is considered as input or training data where n ¼ 1, 2, 3 … . represents the number of inputs,
and the model is prepared through a training process as shown in Figure 1. This individual input is then
in which it is required to make predictions and it multiplied by its corresponding weights (the strength
allows to correct the output if the predictions are con- of the interconnection amongst neurons inside the
sidered wrong based on the level of error. Logistic ANN) and all the weighted inputs are summed up
Regression and the Back Propagation Neural Network inside the computing unit (neuron). A bias is then
are known as a supervised learning model. On the combined to produce the output non-zero or else to
other hand, a model that investigates for previously scale up to the system’s response if the weighted sum
undetected patterns in a data set with no preexisting equates to zero.[19] Bias has the weight and the input
labels and with a minimum of human supervision is to it is always equal to 1.[19] Here the sum of weighted
called unsupervised ML. The unsupervised ML may inputs can be in the range of 0 to positive infinity. A
be through a mathematical process to systematically certain threshold value is benchmarked to keep the
reduce redundancy, or it may be to organize data by response in the limits of the desired value. Then the
similarity. For example, clustering, dimensionality sum of weighted inputs is passed through an activa-
reduction is considered as the most popular unsuper- tion function which can be defined as the set of trans-
vised ML algorithms, which can evaluate the data in fer functions used to obtain the targeted output.
terms of traits and uses. If the model is trained with a Although there are various categories the activation
partly labeled and partly unlabeled data, then the function, linear or non-linear sets of functions are
model is called a semi-supervised ML. For food drying mostly used for engineering research. The network
research, supervised learning such deep learning connection intensity can be constantly adjusted by the
neural network are widely used due to its unique error between the desired output and the actual out-
advantages to handling the complex food drying prob- put until the satisfactory input-output relationship is
lem. ANN can be defined as a mathematical represen- obtained. Details of ANN modeling can be found in
tation of human brain activities, which can handle this recent publication.[10]
several linear or nonlinear processing elements, also In addition to the ANN application in food drying,
called nodes. The nodes are interconnected through Fuzzy logic and other computer expert system have
weighted connections, as shown in Figure 1.[10] An been also recommended.[10] Fuzzy logic is a modern
ANN framework contains the input layer, the hidden computer-based approach for predicting the data
layer and the output layer and is typically specified by based on the degrees of truth rather than the usual
architecture, learning algorithm and neuron model. true or false (1 or 0) Boolean logic. Fuzzy logic has
The ANN model receives the input signal from the the incredible capacity to characterize and control a
user-defined parameters in the form of a pattern and specific problem whose model is not known or is ill-
image in the form of a vector. Then, the given inputs defined.[20] This model has been designed considering
4 M. I. H. KHAN ET AL.

the human common-sense reasoning and decision- drying with lower processing time and computational
making ability with composing, fuzzification, fuzzy cost. For example, artificial neural networks-based
inference and defuzzification.[21] In this modeling models can be developed for a rapid calculation of the
strategy, the operator’s technical knowledge or expert drying rates[42] as well as to predict the temperature
experience is translated into forms of fuzzy rules and and moisture content real-time, during the drying pro-
compose a fuzzy rule base.[10] After the fuzzy process- cess.[43] Considering these benefits, recently many
ing, the precise input signal becomes fuzzy input signal researchers attempted to develop food drying models
which is described through fuzzy inference in fuzzy rule using machine learning algorithm.[44]
base to get fuzzy conclusions. A precise specific output Islam, et al.[42] developed a comprehensive diffu-
data can be obtained after the defuzzification of fuzzy sion-based model using Artificial Neural Network for
conclusions. This method can be applied for parameter predicting the drying kinetics of potato. The model
optimisation,[22,23] prediction of drying kinetics,[24] and was developed for rapid prediction of the drying rates
intelligent monitoring of drying processes.[25] using the Page equation fitted to the drying rate
curves. The following two heat and mass balance
3. ML-based strategies for food drying equations were used to address the simultaneous heat
modeling and performance prediction and mass transfer model.
@T
The existing mathematical models for food drying are qn ¼ r:ðkeff TÞ (1)
categorized as the first-generation model, second-gen- @t
eration model, third-generation models, and the @w

þ r: Deff w ¼ 0 (2)
fourth-generation models. The first-generation models @t
are considered as empirical models which are mostly where, T is the sample temperature ( C) at time t, q
process condition dependent and therefore applica- is the density of the sample (kg/m3), n is the specific
tions are limited.[26,27] The second-generation models heat of the material (J/kg/K), keff is the thermal con-
are considered as a semi-physics-based model which ductivity of the material (W/m/K), w is the instantan-
has better prediction capability as compared to the eous moisture concentration (mol/m3), t is time (s)
empirical models.[28] Over the decades, researchers and Deff is the effective moisture diffusivity (m2/s).
significantly worked to develop full physics-based The study was conducted using a multi-layer feed-for-
model, also termed as the third-generation model. ward network structure with input, output, and hidden
The third-generation physics dependent models layer(s). The input layer containing four neurons which
include single phase models (diffusion models), and corresponded to the air temperature, velocity, humid-
multiphase models. The single-phase models of food ity, and product thickness. The output layer had two
drying have been developed based on the concept of neurons containing two parameters and calculated
bulk moisture transport and only consider diffusion as from 256 cases of data. It was demonstrated that ANN
the predominant factor for moisture migration.[29–35] can predict the entire drying rate curves during con-
However, these models have lumped heat and mass vective drying of potato slices under different drying
transport where only liquid (water) phase is consid- conditions. Hernandez-Perez et al.[43] developed a sim-
ered to migrate from sample to the surroundings. The ultaneous heat and mass transfer model using artificial
multiphase model for food drying is the model where neural network for predicting the drying kinetics of
three phases (liquid water, water vapor, and solid) cassava and mango. They considered the shrinkage of
have been considered for coupled heat and mass the product as a function of moisture content in their
transport during drying.[36–39] The fourth-generation modeling. The moisture ratio was calculated using the
models are considered the most effective physics-based sorption isotherm relationship for mango and cassava.
model which considers the transport phenomena at the The Shrinkage was calculated based on the following
different length scale of food materials, know as multi- empirical relationship.
scale model.[40,41] The ultimate goal of these modeling
d ¼ d0 Ddf þ 1  Ddf XX0 (3)
strategies is to obtain the optimal drying condition to
reduce energy consumption and improve product qual- where, d is the thickness of the sample slab, d0 is the
ity based on better understanding. However, the devel- initial thickness, Ddf is the fraction of initial slab
opment of such models is time-consuming and thickness at the end of drying, and X and X0 are the
computationally expensive. In this case, a machine dry basis moisture content at time t, and the initial
learning-based approach can be implemented for food moisture of the product, respectively. It is recom-
drying modeling to enhance the understanding of food mended that neural network modeling can be used to
 DRYING TECHNOLOGY 5

Table 1. Prediction-based studies for different types of food materials.

Drying method Machine learning technique Food products Remarks Source
[46]
Hot air drying multilayer feed-forward neural Pistachio Nuts -Five drying air temperatures (ranging from 40 to
network (MFNN) 80  C) and four input airflow velocities ranging
from 0.5 to 2 m/s were used.
-MFNN model performed better than the
Lewis, Henderson and Pabis, two-term, and
Page models
[53]
Solar drying Feedforward Neural Networks (FNNs) Seedy grapes - Drying rate was measured
-FNN is more accurate and performed more
consistently than alternative approaches
employed in estimating drying rate
[54]
Freeze drying Feedforward Neural Networks (FNNs) apple - moisture content and drying rate was predicted
with R2 ¼ 0.999
Fluidized bed drying Static and recurrent artificial neural carrot -The dying kinetics was predicted with R2 values [55]

networks (ANNs) of greater than 0.94 and 0.96 using static and
recurrent ANNs, receptively.
[56]
Hot air drying Feedforward Neural Networks (FNNs) Carrot -The optimum ANN had minimum MSE when the
number of neurons, step size, momentum
coefficient, number of epochs, and number of
training runs were 23, 0.37, 0.68, 2,482, and 2,
respectively
[57]
Hot air Dryer Feedforward Neural Networks (FNNs) Grain -The neural-network-based model predictive
scheme is able to enhance the control
precision of grain dryers.
[58]
Rotary drum dryer Supervisory Control and Data Sugar beet pulp -It is a first principle model described in partial
Acquisition (SCADA) systems derivatives that include several features that
make it valuable as a simulation engine for a
training simulator
[59]
Low pressure superheated Feedforward Neural Networks (FNNs) Paneer -Page’s model is considered as the best
steam drying predictive model
[60]
Microwave-Vacuum Drying Feedforward Neural Networks (FNNs) Tomato -The ANN models are able to recognize
relationships between process parameters and
product conditions
Combined hot air-infrared ANN and adaptive neuro-fuzzy blackberry - ANFIS method achieved a higher R2 and lower [61]

drying with ultrasound inference system (ANFIS) RMS as compared to ANN method.
[24]
Microwave drying Recurrent self-evolving fuzzy neural Red maple and -The RSEFNNPC can predict the multi-objective
network (RSEFNN) lignite optimization problem accurately

obtain the better result of moisture prediction, how- temperature on drying kinetics of beetroot and potato
ever, less accurate for the temperature prediction as using multilayer perceptron MLP. Furthermore, the
compared to experimental data.[43] researchers used artificial intelligent (AI) aided
Momenzadeh et al.[45] developed a machine learn- nuclear magnetic resonance (NMR) application in
ing-based framework for predicting the drying kinetics food science[48] as NMR is a promising nondestructive
of corn during microwave-assisted fluidized bed dry- technology to investigate the water status in food
ing. Drying temperature, microwave power, and grain material.[49–51] Very recently Sun et al.[52] evaluated
initial moisture content were considered for the model the potential application of artificial intelligent control
input parameters. 170 neurons were used for studying aided by LF-NMR technique to detect in the endpoint
the influence of transfer functions and training algo- of carrots drying. Likewise, many other researchers
rithms. It was found that the trainrp (Resilient back- used the machine learning-based modeling for predic-
propagation) backpropagation algorithm with Tansig tion of drying kinetics in various food materials such
(hyperbolic tangent sigmoid) transfer function made as seedy grape,[53] apple,[54] carrot,[55,56] potato,[42]
the most accurate predictions for the shelled corn dry- grain,[57] and sugar beet pulp.[58] A summary of the
ing system.[45] Omid et al.[46] developed a machine application of machine learning-based framework in
learning framework to predict the drying kinetics of food drying application can be found in Table 1.
pistachio nuts. Various multilayer perceptron (MLP)
topologies, having one and/or two hidden layers of
4. Application of ML-based approaches in
neurons, were studied to find the optimum model and
prediction of food properties
their prediction performances. It was demonstrated
that a network containing eight neurons in the first Knowledge of some fundamental thermophysical
hidden layer and five neurons in the second hidden mechanical and transport properties of food materials
layer can be the best option to predict the moisture such as diffusivity, permeability, thermal conductivity,
content of the pistachio nuts at all drying runs. specific heat capacity, density, modulus of elasticity,
Kaminski et al.[47] investigated the effect of shape and material stiffness, and rigidity is fundamentally
6 M. I. H. KHAN ET AL.

important in mathematical modeling studies for the to determine the binary diffusivity of a gas. A set of 336
optimisation of product and process development dur- experimental data points for the binary diffusion coeffi-
ing drying. Currently, the experimental method is the cient of both organic and inorganic gases has been used
only approach to investigate such properties, there- for the network training stage. Eslamloueyan and
fore, developing a predictive smart method to deter- Khademi[71] argues that the ANN model is able to pre-
mine such properties is the huge demand for world dict binary gases diffusion coefficient with a considerably
food processing researcher. Machine learning (ML) lower relative error (4.47%) than that of other alternative
based statistical learning predictive modeling strategies correlations. Trigui et al.[72] developed an ANN model
are the effective approach to apply for the prediction to predict sugar diffusivity of three date varieties
of various food properties during drying. Recently (Manakher, Lemsi, and Alligue) under three different
many researchers used this method, as dis- temperatures (50, 65, and 80  C). The optimal ANN
cussed below. model was developed through a network with two hid-
den layers and seven neurones in both the upper and
lower levels of each hidden layers. Based on their argu-
4.1. Transport properties (Diffusivity)
ments, the model is capable of predicting sugar diffusiv-
Diffusivity is one of the most important required ity from the different date varieties with a mean square
transport properties of food materials to address the error of 0.0037 and an 8.0% error.[72] Then the structure
moisture transport phenomena for mathematical mod- was used to develop the diffusivity prediction model.[73]
eling of food drying. There are two types of diffusion: Jena and Sahoo[62] developed an ANN-based model
binary diffusion and capillary diffusion that occurs for predicting the diffusivity of mushroom, tundli,
during food drying. Binary diffusion is defined as the barbatti, and radis during fluidized bed drying. The
movement of gas (vapor or air) concentrations from feed-forward back-propagation three-layered ANN
higher concentration region to lower concentration was used for their modeling. Kaveh et al.[64] applied
region in multiphase porous media like food materi- an adaptive neuro-fuzzy inference system (ANFIS)
als.[39] Capillary diffusion is defined as the process of and ANN model for predicting the diffusivity of
diffusion that happens due to the capillary action of potato, garlic and cantaloupe during convective hot
the liquid. Capillary action is a result of surface ten- air drying. According to their results, the maximum
sion that occurs when the liquid and the other surface diffusivity value can be 9.76  109 for potato, 0.13 
come in contact with each other. The binary diffusion 109 for garlic, and 9.97  1010 m2/s for cantaloupe.
phenomenon is used for liquid transport in case of All of the aforementioned studies have been con-
multiphase transport modeling. Many researchers ducted at the macroscale. However, food materials,
attempted to use artificial neural network-based pre- particularly plant-based food material are multiscale
dictive approach to determine the diffusivity of vari- in nature and therefore knowledge of microscale
ous materials including plant-based food properties are highly important. However, develop-
[62–64]
materials and building materials such as con- ment of such an ANN-based framework for predicting
crete,[65] foamed polystyrene insulating material,[66] microscale properties is challenging due to the struc-
hydrocarbons and aromatics,[67] and other building tural heterogeneities of food materials. Recently, Wu
materials.[68,69] Limited study has been conducted for et al.[73] attempted to develop a predictive model for
food materials during drying. Mariani et al.[63] devel- determining the micro-level diffusivity of porous
oped a new approach for the estimation of apparent materials using image-based deep (convolutional
diffusivity of banana at different drying temperatures neural network) learning method. In this research, a
using ANN-based inverse method. They found that porous and heterogenous structure has been created
the diffusivity of banana can change from 1.88  107 artificially using reconstruction methods, where
(m2/s) to 2.49  1010 (m2/s). A small change in the 0.28–0.98 porosity variations were considered. This
drying temperature and moisture content of banana approach has been developed considering the general
cause a significant change in this diffusivity value.[63] porous media and their characteristics, which restricts
Mohamed[70] developed a simple and robust method to apply this framework to plant-based food materials.
for the determination of thermal diffusivity of solid
foods using an inverse numerical method. The model
4.2. Thermophysical properties
was validated using computer-generated temperatures
in addition to actual experimental temperatures data. Thermal conductivity, specific heat capacity perme-
Eslamloueyan and Khademi[71] used a neural network ability, viscosity, and density or porosity are the most
 DRYING TECHNOLOGY 7

important thermophysical properties of food materials, for developing a physics-based model to predict the
particularly in food drying. Many researchers deformation as well as textural changes while the food
attempted to apply a machine learning-based frame- products are subjected to the thermal processes such
work to determine such properties in different food as drying processes. Unlike other engineering materi-
materials during different thermal processing. Sablani als, food products, particularly plant-based food mate-
et al.[74] used ANN approach to model the thermal rials are highly structured complex materials and
conductivity of bakery products such as bread, bread therefore investigation of the mechanical properties is
dough, yellow cake, and tortilla chip as a function of challenging, particularly at the fine-scale. In this case,
product moisture content, temperature and apparent machine learning-based predictive approaches can be
density. The optimal model consisted of two hidden an effective method to determine the mechanical
layers, with six neurons in each hidden layer, and able properties of food materials at various length scales.
to predict the thermal conductivity values with a Many researchers attempted to investigate the mech-
mean relative error of about 10% and standard error anical properties using ANN for various types of food
of 0.003 W/m/K. Based on optimal results, the sim- materials, for instance, apple,[84–86] kiwi,[87] date
plest ANN model with one hidden layer and two neu- fruit,[88] pomegranate,[89] chicken nuggets,[90] cumin
rons has been recommended for future research. The seed,[91] canola stem,[92] and Swiss-type cheese.[93] A
ANN approach was also used to predict the thermal multi-layer feed-forward or backward network which
conductivity of various fruit and vegetables such as have one or more hidden layers of sigmoid neurons
apples, pears, corn starch, and potatoes.[75] followed by an output layer of linear neurons are
Sablani and Rahman[76] developed ANN-based mostly used for this type of study. Multiple layers of
model for the prediction of thermal conductivity of neurons with non-linear transfer functions allow the
diverse range of food as a function of moisture con-
network to learn non-linear and linear relationships
tent, temperature and apparent porosity of food mate-
between input and output vectors.[90] The output of
rials. The optimal output of the model has been
the ANN can provide information about various
chosen based on the lowest possible errors of the pre-
mechanical properties based on the input information
dictive model. The model can predict thermal con-
and the training strategies. The input information can
ductivity with a mean relative error of 12.6%, a mean
be based on the structural image processing data or
absolute error of 0.081 W/m K. In comparison with
real experimental values. The input structural image
other statistical approaches, it is found that artificial
information includes image smoothness, contrast,
neural networks have an excellent prediction capabil-
entropy, homogeneity, and image linearity.[90] Based
ity of specific heat, thermal conductivity, and density
on this input structural image information, the mech-
of liquid food such as milk than polynomial model-
anical characteristics such as the maximum load,
ing[77] although a list of the empirical relationship of
energy to breakpoint, and toughness can be obtained
heat capacity prediction models (linear, polynomial,
etc.) has been developed for specific foods and a range as the outputs of the ANN model.[90]
Vasighi-Shojae et al.[84] developed a feed-forward
of moisture contents.[78,79] Rahman et al.[80] developed
a predictive model using the fuzzy logic-based algo- neural network with the backpropagation algorithm-
rithm to investigate the effects of temperature, mois- based ANN model for predicting the firmness, elastic
ture content, and porosity on thermal conductivity of modulus, and stiffness of apples. They applied the
various fruit and vegetables. Recently, the details ther- popular Gradient descent with momentum (GDM)
mophysical properties prediction and investigation algorithm for error minimization in ANNs’ training
method have been discussed by Paluri et al.[81] process. To develop the ANN model, The Weka soft-
ware 3.9.2 was used. A total of 30% of the acquired
data (from the ultrasonic experiment) were randomly
4.3. Mechanical properties selected and used as prediction data sets, whereas the
Modulus of elasticity, material stiffness, rigidity, remaining samples were used for training and valid-
Poison’s ratio, etc. are the most important mechanical ation of ANNs. The various input vectors of the
properties of a material.[82] Therefore, to design and ANNs include ultrasound velocity and attenuation on
develop any types of new materials including dry food D1 (horizontal diameter 1), ultrasound velocity and
products, characterization of these mechanical proper- attenuation on D2 (horizontal diameter 2), and the
ties is essential. These properties indicate the visco- vertical diameter (D3). Based on these input parame-
elastic behaviors of food materials[83] and are required ters, the best firmness value of the golden delicious
8 M. I. H. KHAN ET AL.

food microstructure is not possible. However, classifi-

cation based on similar composition, cellular nature
and air space pattern can be achieved.
Composition of food materials plays remarkable
roles in the forming structure, maintaining physio-
chemical stability, and retention of nutrition. Most of
the food materials encompass polysaccharide and pro-
tein, whereas water act as main plasticizer and solvent.
Loss of water and absorption of excess water affects
the structural features of cell walls and eventually
modify the microstructure of cellular solid foods. The
cellular food microstructure such as plant-based food
Figure 2. Schematic structure of the best ANN for firmness materials contains various cellular compartments such
prediction (Adopted from Ref.[84]). as intracellular, intercellular, cell membrane, and cell
wall. The intracellular components are considered as
apple was predicted by ANN,[84] as shown in
the combination of vacuole with the cytoplasm. The
Figure 2.
middle lamella is considered as intercellular compo-
Most of the studies have been conducted for fresh
nents. The cell membrane is a double layer of protein
food materials at the macroscale. Prediction of micro-
and phospholipids.[49] These cellular environments
scale mechanical properties in line with the drying
contain a different proportion of water with their dif-
processes is challenging as the drying process continu-
ferent water potential.[96] The cells are held together
ously alters the mechanical properties and vice versa.
through the middle lamella, of which the main com-
Due to these complexities, not many pieces of
ponent is pectin.[98] This structural elements and their
research have been attempted to develop an ANN-
interactions are the main drivers for the physical
based model for mechanical properties prediction dur-
behavior and the mechanical stability during food dry-
ing drying. Kaveh et al.[94] developed an ANN-based
ing.[99] Drying processes significantly affect these
model to predict the cell rupture of terebinth fruit
after semi-industrial continuous drying. The feed and structural elements, which significantly impacts the
cascade forward backpropagation networks with learn- mechanical properties such as strength, modulus of
ing algorithms of Levenberg-Marquardt and the elasticity, material stiffness, rigidity, and hardness
Bayesian regulation were used to train the ANN pat- maintain the cell stabilities. The change of these
terns. To predict the optimal rupture force, the feed- mechanical characteristics impacts the food textural
forward network with the Bayesian regulation is the characteristics, for instance, shallowness depends on
best strategy,[94] and therefore can be considered for the force, stress, and strain.[100–102] Therefore, the
further prediction of periodic cell rupture of various characterization of food microstructure is important.
food materials during drying.[95,96] However, the cell However, characterization of food materials is quite
rupture depends on the microstructural features of challenging as it requires a wide variety of image
food materials with the drying conditions, and there- acquisition techniques prevails to get a high-resolution
fore, microstructural characterization is also essential. image of food material including SEM, X-ray micro-
tomography, confocal laser scanning microscopy,
optical Coherence Tomography, magnetic resonance
5. Food microstructure and their imaging. To overcome such problems, machine learn-
characterization using ML-based algorithms ing-based algorithms can be used to artificially charac-
Knowledge of microstructural elements of food mate- terize the microstructure based on the known
rials is crucial for predicting the drying kinetics. Both porosity.[103]
heat and mass transfer phenomena and their associ- Microstructure quantification using machine learn-
ated properties including diffusivities depend on ing requires a system that can be able to execute vis-
microstructure. The microstructure of different food ual data analysis. This system needs to be developed
materials varies due to the diverse structural charac- using hardware’s (sensors) and software containing
teristics prevail for different material.[97] The diversifi- logics of extraction of required quantitative informa-
cation is mainly caused due to the variation of tion. This type of study is complex and challenging
compositions, cellular structure, tissue types and dis- therefore very few studies have been attempted in this
tribution of air spaces. Therefore, a generalization of regard. Castro et al.[104] developed a convolution
 DRYING TECHNOLOGY 9

neural network (CNN) model to generate the vege- information (data) set. Very recently, Sansiribhan
table microstructures. To improve the detection of et al.[106] used fractal dimension algorithm for the
microstructural elements such as cells and intercellular microstructural characterization. The fractal dimen-
spaces, the micrographs of pumpkin were processed sion of each sample image was calculated using the
digitally. A pre-trained network (AlexNet) was used box-counting algorithm. Image processing is the most
for the implementation and cross-validation. One important part of the structural characterization where
hundred repetitions randomizing the information deep learning-based algorithm can be used.[107]
delivered to the training and validation processes were Although several studies are available in connection
considered. The AlexNet has five convolutional layers with image segmentation, there is still limited study
in which three are fully connected. Each convolved regarding the analysis of the image of different con-
layer contains multiple filters 3D (cores), connected to trast modes, different textures, and colored images.
the output of the previous layer. It is suggested that Advanced image segmentation model using machine
CNN can be used for the discrimination of structures learning can overcome this challenge of microstruc-
in food materials.[104] Wu et al.[73] developed a two- tural image analysis. Successful implementation of
dimensional pore structure using reconstruction Fuzzy logic, fractal analysis, genetic algorithms, ANN,
method and represented as image by applying CNN and hybrid model can provide great insight of micro-
algorithm. Based on the structural characteristics, the structural image quantification and optimization.
effective diffusivity of potato parenchyma has been
investigated. The porosity of porous structures was
6. Optimization of quality attributes and
encoded directly into the neural network to avid the
drying process parameters using ML
error for properties prediction.
Oblitas-Cruz et al.[105] developed an excellent prob- Optimization of drying process parameters and quality
abilistic neural network called newppn based method attributes are important for increasing the energy effi-
to generate the microstructure of pumpkin. The pro- ciency of food drying system and improve the quality
cedure flowed as interface restoration and classifica- of the products. Drying process parameters include
tion of the image. The restoration procedure is drying temperature, air velocity, air pressure, ambient
followed by segmentation such as the conversion of air quality, product’s size and shape etc. Three key
grayscale image to a binary image. In this interface, quality attributes include such as textural attributes,
automatic classification of elements as cells, intercellu- physical appearance (colour) and nutritional health
lar spaces and unrecognizable elements is also per- promising components (vitamin C, ascorbic acid, total
formed, generating a labeled image. The individual phenol, antioxidant, etc). The texture is a sensory fea-
size and shape parameter such as area equivalent ture and is composed of several textural properties
diameter, major axis length, minor axis length, perim- including mechanical (hardness, elastic modulus, and
eter, roundness, elongations, compactness of the struc- viscosity), geometrical (particle size and shape) and
tural elements was used as the input parameters of compositional (moisture and fat content) characteris-
the model. Firstly, the parameters were normalized tics.[108] These quality components can be measured
through dividing by the corresponding normalization by fundamental, imitative, and empirical ways, and
constant from a matrix for normalization. Each nor- needs to be optimized. Machine learning-based opti-
malization constant was the value that achieves 0.95 mization approaches such as deep learning optimiza-
in cumulative frequency distribution for the respective tion, optimization in reinforcement learning,
parameter. The study proposed the combinations of optimization in meta-learning, optimization in vari-
the various structural parameters that generated the ational inference, Markov chain Monte Carlo, deriva-
lowest error in the classification of the microstructural tive free optimization, etc. Among the various popular
elements by comparison with manual classifica- methods, heuristic algorithms in derivative free opti-
tion.[105] Furthermore, the fractal analysis can be used mization are the most practical application-based opti-
in characterizing microstructural images and eventu- mization method. It is characterized by empirical
ally offers great potential in food engineering applica- rules and chooses methods that have already worked
tions in quantitative image analysis. The fractal well, rather than derives solutions systematically.[109]
dimension is a geometrical concept and refer to the The heuristic optimization methods include genetic
shape of an object that appears the same in different algorithms, classical simulated annealing arithmetic,
scale of observation. Fractal dimension in data ana- ant colony algorithms, and particle swarm optimiza-
lysis refers to the degrees of freedom of the tion. Among them, genetic algorithm (GA) based
10 M. I. H. KHAN ET AL.

optimization is the most attractive method for food drying time 120 min to 130 min. The energy require-
drying researcher due to its significant benefits as ments decrease exponentially with an increment of
compared to the traditional optimization methods. moisture content within a specific drying temperature
The GA can easily deal with the multi-objective range (30  C–50  C and 12.5%–20% MC).[113] A com-
complex problems, whether the objective (fitness) prehensive review has been conducted by Devahastin
function is stationary or nonstationary (changes with and Niamnuy[114] to show the optimization of quality
time), linear or nonlinear, continuous or discontinu- attributes using machine learning-based strategies. To
ous, or with random noise. The population (or any predict the shrinkage of kiwi fruit, Fathi, et al.[115] a
subgroup) can explore the search space simultaneously neural network platform was developed using fractal
in multi-directions as multiple offspring in a popula- dimension and a genetic algorithm was applied for
tion act like independent agents.[110] These features optimization of the neural network’s parameters. Very
make it ideal to parallelize the algorithms for imple- recently, Sun, et al.[116] monitored the garlic flavor
mentation. Different parameters and even different LF-NMR combined with partial least squares (PLS)
groups of encoded strings can be manipulated at the and back-propagation artificial neural network (BP-
same time. Due to this unique characteristic of GA, ANN). Besides, few researchers attempted to optimize
more than 90% of food drying researcher used GA the quality attributes parameters using ML-based
based optimization method for optimizing the quality strategies for various food materials such as blueber-
attributes and process parameters. Morimoto et al.[111] ries,[117] apple,[118] green peas,[119] pomegranate
developed a new control technique, involving neural
juice,[120] sweet potato[121] bread,[122] kiwi,[123] lac-
networks and genetic algorithms to control and opti-
tose,[124] banana,[125] and yoghurt.[126]
mize the process parameters. Firstly, relative humidity
as the most influencing parameter for dehydration
was identified using neural network and then was 7. Challenges to developing ML-based
optimized using GA. The GA with high crossover and algorithms in food drying application
mutation rates allowed the optimal value to be quickly
Development of an accurate machine learning-based
sought and control. Curvelo Santana, et al.[112] devel-
food drying model is a challenging and rigorous work
oped a GA based optimization model to estimate the
as a drying process itself is a complex simultaneous
optimal temperature (T) and time (t) parameters to
heat and mass transfer process with continuous phase
maintain the maximum activity of a- and b-amylases
change. The main challenges are discussed below.
enzymes in the obtained corn product. Various
experimental data were used to fit the models. The
chromosome structure was considered as a binary 7.1. ANN modeling challenges
string with 20 bits divided into two genes g1 and g2
Although ANN is considered as an excellent platform
with the variables t and T. It was considered that at
each generation the best one-sixth of the population to handle the complex problem with nonlinear data
could be selected and among them, 10% was pre- set, understanding the comprehensibility of ANN
served intact for the next generation. Two crossover modeling processes is complicated and challenging.
points were used for maximizing the objective func- This is because the ANN model seems like a “Black
tion. The various remaining parameters of the GA box” as the user cannot see exactly what is happened
configuration was population size: 400 individuals during the computation of the model. Therefore, there
with 10% mutation rate. Their optimizations results is a great possibility to provide misleading output. To
showed that at the optimal temperature with 54  C overcome this problem, Decision Tree (DT) algorithm
and drying time 5.6 and 6.4 h of the drying process, can be used, as the DT is very straight forward to
values of the specific activity of 5.26 ± 0.06 SKB/mg at show the comprehensibility of the processes without
15.69 ± 0.10% of moisture in corn malt. much complexity. Moreover, understanding the study
The same strategy was followed by Rahman of the nonlinear problem including the nonlinear
et al.[113] to optimize the energy consumption during dynamic system, adaptation, self-organizing, chaotic
drying of rice. The study clearly presents the input of neural network and neural network mathematical the-
the crossover operation called guardians and the ory,[10] with the basic characteristics of ANN such as
reproduced output chromosome known as children. It stability, robustness, convergence, fault and tolerance
was recommended that the best operation zone for is one of the biggest challenges of the neural network-
rice drying could be at drying temperature 47  C with based modeling for food drying.
 DRYING TECHNOLOGY 11

7.2. Challenges to applying ML-based algorithms fundamental structure-property relationship is

in microwave-based hybrid drying required to address the structural heterogeneity in
food drying. Moreover, understanding of microstruc-
Microwave based drying has proved itself as an
tural features and microscale properties are also
emerging technology to drastically reduce the drying
important to improve the product’s physical as well as
time (up to 80% of drying time).[127] However, con-
textural attributes, which are considered as the most
tinuous application of microwave energy deteriorates
important quality parameters of dry food prod-
the product’s quality due to the uneven heating (cre-
ation of hot spots and cold spots).[128] To overcome ucts,[132] in terms of customer satisfaction. Therefore,
such problems, microwave energy can be used in a the development of a machine learning-based predic-
pulsed or intermittent manner with convective hot air tion model to estimate the microscale properties con-
drying that is defended as intermittent microwave sidering the microstructural changes continuous
convective drying (IMCD). IMCD has been proven to moisture loss during drying is highly important to
improve product quality and energy efficiency by advance the food drying technology. However, devel-
allowing redistribution of temperature and moisture opment of such an ML-based predictive model is a
within the product during tempering (MW off times) very complex problem as it requires two-step model-
period.[129,130] Development of ML-based IMCD ing to address two different scale data. Besides, the
model is challenging as the formulation of microwave development of a multiscale food drying model is
heating is a complex process. Usually, Lambert’s law incredibly demandable for accurately predicting food
and Maxwell’s equation are a well-known approach drying performance. This type of model is essentially
for mathematical modeling of microwave heating. The a hierarchy of sub-models which describe the material
Lambert’s law is quite simple and therefore formula- behavior at different spatial scales in such a way that
tion of the model using ML-based algorithms is easier. the sub-models are interconnected. However, develop-
However, Maxwell’s equations are very complex to ment of such a physics-based model is much complex
develop mathematical modeling, particularly in and computationally expensive. Therefore, the devel-
Multiphysics applications such as food drying. opment of a multiscale model for food drying using a
Maxwell equations consist of four vectory equations machine learning-based framework is a prerequisite
such as Faraday’s law, Amp
ere-Maxwell law, Gauss’s although it is highly challenging research as it requires
law for the electric field, and Gauss’s law for the mag- much fundamental information to formulate the
netic field. Combination of these four different laws is model. For instance, various cellular water transport
used to solve the electric and magnetic fields. Once mechanism such as microscale diffusion and periodic
the electric and magnetic field distributions are cell rupture needs to be considered to formulate an
observed, volumetric unit power generation can be ML-based multiscale model.[95] Although the develop-
calculated based on the Poynting theorem.[128] Due to ment of ML-based multiscale mode challenging, but
extreme complexity to express these combine once that have huge potential to advance the food
Multiphysics problems by ML-based algorithms, study drying technology.
to develop an ML-based IMCD would be challenging,
but once that has huge potential to advance the 8. Conclusions
microwave-based hybrid drying technology.
Like other engineering fields, application of machine
learning-based predictive modeling approaches in
7.3. Challenges to address multiscale features in food drying could be highly potential in order to
food drying determine the drying kinetics information without
Food materials, particularly plant-based food materials developing computationally intensive physics-based
are structurally heterogeneous multiscale cellular numerical model. Moreover, the ML-based predictive
materials. As the size and shape of individual micro- model is also required for predicting the food proper-
structural elements of food materials such as cells, the ties and microstructural features, without performing
intercellular spaces, and cell walls vary, the unique cel- extensive experimentation. However, the significant
lular properties such as diffusivity, permeability, ther- scientific effort is required to advance this field. This
mal conductivity, and elastic modulus of these paper presents a comprehensive review of the current
elements are also different.[131] All of these properties status of machine learning-based modeling for food
depend on the microstructural features of food mate- drying and the associated strategies for ML implemen-
rials, and therefore, the development of the tation. Food properties play an important role to
12 M. I. H. KHAN ET AL.

develop food drying model. However, currently, most [5] Freire, F. B.; Vieira, G. N.; Freire, J. T.; Mujumdar,
of the required properties are unavailable due to the A. S. Trends in Modeling and Sensing Approaches
complexity to investigate such properties through for Drying Control. Drying Technol. 2014, 32,
1524–1532. DOI: 10.1080/07373937.2014.925471.
experimental investigation. This paper presents the [6] Mujumdar, A. S. Research and Development in
attempts undertaken to measure important food prop- Drying: Recent Trends and Future Prospects.
erties using machine learning-based predictive models. Drying Technol. 2004, 22, 1–26. DOI: 10.1081/DRT-
Furthermore, optimization of the drying process 120028201.
parameters such as drying time, drying temperature, [7] Su, Y.; Zhang, M.; Mujumdar, A. S. Recent
air velocity, humidity etc. is required to reduce the Developments in Smart Drying Technology. Drying
Technol. 2015, 33, 260–276. DOI: 10.1080/
energy consumption and improve the quality of 07373937.2014.985382.
the product. This paper also presents an overview of [8] Mujumdar, A. S.; Zhonghua, W. Thermal Drying
the application of machine learning-based framework Technologies—Cost-Effective Innovation Aided by
for the optimization of drying process parameters and Mathematical Modeling Approach. Drying Technol.
quality attributes. Moreover, the challenges of apply- 2007, 26, 145–153. DOI: 10.1080/
ing machine learning-based approaches for multiscale 07373930701812606.
[9] Martynenko, A.; Misra, N. Machine Learning in
modeling is also discussed. It is necessary to advance Drying. Drying Technol. 2020, 38, 596–609. DOI:
this field by applying Ml-based modeling strategies in 10.1080/07373937.2019.1690502.
multiscale modeling and properties prediction. [10] Sun, Q.; Zhang, M.; Mujumdar, A. S. Recent
Moreover, the existing researchers mostly used ANN Developments of Artificial Intelligence in Drying of
for food drying researcher. As ANN is much complex Fresh Food: A Review. Crit. Rev. Food Sci. Nutr.
2019, 59, 2258–2275. DOI: 10.1080/10408398.2018.
and hard to understand the process comprehensibility,
1446900.
other models like Decision Tree (DT) or other regres- [11] Huang, B.; Mujumdar, A. Use of Neural Network
sion type model can be used to exactly understand the to Predict Industrial Dryer Performance. Drying
model comprehensibility. Moreover, hybrid machine Technol. 1993, 11, 525–541. DOI: 10.1080/
learning-based algorithm combining ANN, fuzzy logic, 07373939308916842.
DT, or their combinations with classical mathematical [12] Liu, Z.-L.; Bai, J.-W.; Wang, S.-X.; Meng, J.-S.;
Wang, H.; Yu, X.-L.; Gao, Z.-J.; Xiao, H.-W.
models can be used to overcome the problem of single
Prediction of Energy and Exergy of Mushroom
ANN application. It is expected that the crucial infor- Slices Drying in Hot Air Impingement Dryer by
mation of this paper may help to further advance the Artificial Neural Network. Drying Technol. 2020,
application of machine learning-based approaches for 38, 1959–1970. DOI: 10.1080/07373937.2019.
food drying. 1607873.
[13] Nazghelichi, T.; Aghbashlo, M.; Kianmehr, M. H.;
Omid, M. Prediction of Energy and Exergy of
ORCID Carrot Cubes in a Fluidized Bed Dryer by Artificial
Neural Networks. Drying Technol. 2011, 29,
M. A. Karim http://orcid.org/0000-0001-9074-0384
295–307. DOI: 10.1080/07373937.2010.494237.
[14] Yousefi, G.; Emam-Djomeh, Z.; Omid, M.; Askari,
G. R. Prediction of Physicochemical Properties of
References
Raspberry Dried by Microwave-Assisted Fluidized
[1] Gustavsson, J.; Cederberg, C.; Sonesson, U. L. F. Bed Dryer Using Artificial Neural Network. Drying
Global Food Losses and Food Waste-Extent, Causes Technol. 2014, 32, 4–12. DOI: 10.1080/07373937.
and Prevention. Food and Agriculture 2013.801849.
Organization (FAO): Rome, 2011. [15] Farkas, I. Use of Artificial Intelligence for the
[2] Karim, M. A.; Hawlader, M. Mathematical Modelling of Drying Processes. Drying Technol.
Modelling and Experimental Investigation of 2013, 31, 848–855. DOI: 10.1080/07373937.2013.
Tropical Fruits Drying. Int. J. Heat Mass Transf. 769002.
2005, 48, 4914–4925. DOI: 10.1016/j.ijheatmas- [16] Aghbashlo, M.; Hosseinpour, S.; Mujumdar, A. S.
stransfer.2005.04.035. Application of Artificial Neural Networks (ANNs)
[3] Karim, M. A.; Hawlader, M. Drying Characteristics in Drying Technology: A Comprehensive Review.
of Banana: Theoretical Modelling and Experimental Drying Technol. 2015, 33, 1397–1462. DOI: 10.
Validation. J. Food Eng. 2005, 70, 35–45. DOI: 10. 1080/07373937.2015.1036288.
1016/j.jfoodeng.2004.09.010. [17] Ttayagarajan, T.; Ponnavaikko, M.; Shanmugam, J.;
[4] Chou, S. K. C.; Chua, K. J. New Hybrid Drying Panda, R.; Rao, P. Artificial Neural Networks:
Technologies for Heat Sensitive Foodstuffs. Trends Principle and Application to Model Based Control
Food Sci. Technol. 2001, 12, 359–369. DOI: 10. of Drying Systems-A Review. Drying Technol. 1998,
1016/S0924-2244(01)00102-9. 16, 931–966. DOI: 10.1080/07373939808917449.
 DRYING TECHNOLOGY 13

[18] Martynenko, A. Artificial Intelligence: Is It a Good Shrinkage and Variable Diffusion Coefficient.
Fit for Drying. Dry. Technol, 2017, 1–3. DOI: 10. Drying Technol. 2013, 31, 40–51. DOI: 10.1080/
1080/07373937.2017.1362153. 07373937.2012.714826.
[19] Savaram, R. What Is Artificial Neural Network and [33] Kumar, C.; Millar, G. J.; Karim, M. A. Effective
How It Works? MindMajix, 2017. Diffusivity and Evaporative Cooling in Convective
[20] Togai, M.; Chiu, S. A Fuzzy Logic Chip and a Drying of Food Material. Drying Technol. 2015, 33,
Fuzzy Inference Accelerator for Real-Time 227–237. DOI: 10.1080/07373937.2014.947512.
Approximate Reasoning. Proceedings of the 17th [34] Fowler, A. J.; Bejan, A. The Effect of Shrinkage on
International Symposium of Multiple Valued Logic, the Cooking of Meat. Int. J. Heat Fluid Flow 1991,
1987, pp 25–29. 12, 375–383. DOI: 10.1016/0142-727X(91)90027-S.
[21] Adnan, M. M.; Sarkheyli, A.; Zain, A. M.; Haron, [35] Khan, M.; Kumar, C.; Joardder, M.; Karim, M.
H. Fuzzy Logic for Modeling Machining Process: A Determination of Appropriate Effective Diffusivity
Review. Artif. Intell. Rev. 2015, 43, 345–379. DOI: for Different Food Materials. Drying Technol. 2017,
10.1007/s10462-012-9381-8. 35, 335–346. DOI: 10.1080/07373937.2016.1170700.
[22] Stawczyk, J.; Comaposada, J.; Gou, P.; Arnau, J. [36] Datta, A. K. Porous Media Approaches to Studying
Fuzzy Control System for a Meat Drying Process. Simultaneous Heat and Mass Transfer in Food
Drying Technol. 2004, 22, 259–267. DOI: 10.1081/ Processes. I: Problem Formulations. J. Food Eng.
DRT-120028232. 2007, 80, 80–95. DOI: 10.1016/j.jfoodeng.2006.05.
[23] Taprantzis, A. V.; Siettos, C. I.; Bafas, G. V. Fuzzy 013.
Control of a Fluidized Bed Dryer. Drying Technol. [37] Gulati, T.; Datta, A. K. Mechanistic Understanding
1997, 15, 511–537. DOI: 10.1080/ of Case-Hardening and Texture Development dur-
07373939708917244. ing Drying of Food Materials. J. Food Eng. 2015,
[24] Li, J.; Xiong, Q.; Wang, K.; Shi, X.; Liang, S. A 166, 119–138. DOI: 10.1016/j.jfoodeng.2015.05.031.
Recurrent Self-Evolving Fuzzy Neural Network [38] Mercier, S.; Marcos, B.; Moresoli, C.; Mondor, M.;
Predictive Control for Microwave Drying Process. Villeneuve, S. Modeling of Internal Moisture
Drying Technol. 2016, 34, 1434–1444. DOI: 10. Transport during Durum Wheat Pasta Drying.
1080/07373937.2015.1122612. J. Food Eng. 2014, 124, 19–27. DOI: 10.1016/j.jfoo-
[25] Tavakolipour, H.; Mokhtarian, M.; Kalbasi-Ashtari, deng.2013.09.028.
A. Intelligent Monitoring of Zucchini Drying [39] Khan, M. I. H.; Joardder, M. U. H.; Kumar, C.;
Process Based on Fuzzy Expert Engine and ANN. Karim, M. A. Multiphase Porous Media Modelling:
J. Food Process Eng. 2014, 37, 474–481. DOI: 10. A Novel Approach to Predicting Food Processing
1111/jfpe.12101. Performance. Crit. Rev. Food Sci. Nutr. 2018, 58,
[26] Jomaa, W.; Puiggali, J. R. Drying of Shrinking 528–546. DOI: 10.1080/10408398.2016.1197881.
Materials: Modellings with Shrinkage Velocity. [40] Welsh, Z.; Simpson, M. J.; Khan, M. I. H.; Karim,
Drying Technol. 1991, 9, 1271–1293. DOI: 10.1080/ M. A. Multiscale Modeling for Food Drying: State
07373939108916750. of the Art. Compr. Rev. Food Sci. Food Saf. 2018,
[27] Kiranoudis, C. T.; Tsami, E.; Maroulis, Z. B.; 17, 1293–1308. DOI: 10.1111/1541-4337.12380.
Marinos-Kouris, D. Drying Kinetics of Some Fruits. [41] Rahman, M.; Joardder, M. U.; Khan, M.; Pham,
Drying Technol. 1997, 15, 1399–1418. DOI: 10. N. D.; Karim, M. Multi-Scale Model of Food
1080/07373939708917300. Drying: Current Status and Challenges. Crit. Rev.
[28] Putranto, A.; Chen, X. D.; Webley, P. A. Modeling Food Sci. Nutr. 2018, 58, 858–876. DOI: 10.1080/
of Drying of Food Materials with Thickness of 10408398.2016.1227299.
Several Centimeters by the Reaction Engineering [42] Islam, M. R.; Sablani, S.; Mujumdar, A. An
Approach (REA). Drying Technol. 2011, 29, Artificial Neural Network Model for Prediction of
961–973. DOI: 10.1080/07373937.2011.557793. Drying Rates. Drying Technol. 2003, 21, 1867–1884.
[29] Ben Mabrouk, S.; Benali, E.; Oueslati, H. DOI: 10.1081/DRT-120025512.
Experimental Study and Numerical Modelling of [43] Hernandez-Perez, J. A.; Garcıa-Alvarado, M. A.;
Drying Characteristics of Apple Slices. Food Trystram, G.; Heyd, B. Neural Networks for the
Bioprod. Process. 2012, 90, 719–728. DOI: 10.1016/j. Heat and Mass Transfer Prediction during Drying
fbp.2012.02.001. of Cassava and Mango. Innovative Food Sci. Emerg.
[30] Kaya, A.; Aydın, O.; Dincer, I. Experimental and Technol. 2004, 5, 57–64. DOI: 10.1016/j.ifset.2003.
Numerical Investigation of Heat and Mass Transfer 10.004.
during Drying of Hayward Kiwi Fruits (Actinidia [44] Jumah, R.; Mujumdar, A. S. Modeling Intermittent
Deliciosa Planch). J. Food Eng. 2008, 88, 323–330. Drying Using an Adaptive Neuro-Fuzzy Inference
DOI: 10.1016/j.jfoodeng.2008.02.017. System. Drying Technol. 2005, 23, 1075–1092. DOI:
[31] Batista, L. M.; da Rosa, C. A.; Pinto, L. A. A. 10.1081/DRT-200059138.
Diffusive Model with Variable Effective Diffusivity [45] Momenzadeh, L.; Zomorodian, A.; Mowla, D.
considering Shrinkage in Thin Layer Drying of Experimental and Theoretical Investigation of
Chitosan. J. Food Eng. 2007, 81, 127–132. DOI: 10. Shelled Corn Drying in a Microwave-Assisted
1016/j.jfoodeng.2006.10.014. Fluidized Bed Dryer Using Artificial Neural
[32] Golestani, R.; Raisi, A.; Aroujalian, A. Mathematical Network. Food Bioprod. Process. 2011, 89, 15–21.
Modeling on Air Drying of Apples considering DOI: 10.1016/j.fbp.2010.03.007.
14 M. I. H. KHAN ET AL.

[46] Omid, M.; Baharlooei, A.; Ahmadi, H. Modeling a Training Simulator. Drying Technol. 2017, 35,
Drying Kinetics of Pistachio Nuts with Multilayer 1765–1780. DOI: 10.1080/07373937.2016.1275674.
Feed-Forward Neural Network. Drying Technol. [59] Shrivastav, S.; Kumbhar, B. K. Drying Kinetics and
2009, 27, 1069–1077. DOI: 10.1080/ ANN Modeling of Paneer at Low Pressure
07373930903218602. Superheated Steam. J. Food Sci. Technol. 2011, 48,
[47] Kaminski, W.; Tomczak, E.; Skorupska, E. 577–583. DOI: 10.1007/s13197-010-0167-1.
Estimation of the Effect of Shape and Temperature [60] Poonnoy, P.; Tansakul, A.; Chinnan, M. Artificial
on Drying Kinetics Using MLP. Drying Technol. Neural Network Modeling for Temperature and
2004, 22, 191–200. DOI: 10.1081/DRT-120028227. Moisture Content Prediction in Tomato Slices
[48] Khan, M. I. H.; Wellard, R. M.; Nagy, S. A.; Undergoing Microwave-Vacuum Drying. J. Food
Joardder, M. U. H.; Karim, M. A. Investigation of Sci. 2007, 72, E042–E047. DOI: 10.1111/j.1750-
Bound and Free Water in Plant-Based Food 3841.2006.00220.x.
Material Using NMR T2 Relaxometry. Innovative [61] Taghinezhad, E.; Kaveh, M.; Khalife, E.; Chen, G.
Food Sci. Emerg. Technol. 2016, 38, 252–261. DOI: Drying of Organic Blackberry in Combined Hot
10.1016/j.ifset.2016.10.015. Air-Infrared Dryer with Ultrasound Pretreatment.
[49] Khan, M. I. H.; Karim, M. Cellular Water Drying Technol. 2020, 1–17. DOI: 10.1080/
Distribution, Transport, and Its Investigation 07373937.2020.1753066
Methods for Plant-Based Food Material. Food Res. [62] Jena, S.; Sahoo, A. ANN Modeling for Diffusivity of
Int. 2017, 99, 1–14. DOI: 10.1016/j.foodres.2017.06. Mushroom and Vegetables Using a Fluidized Bed
037. Dryer. Particuology 2013, 11, 607–613. DOI: 10.
[50] Khan, M. I. H.; Rahman, M.; Karim, M. Recent 1016/j.partic.2012.07.015.
Advances in Micro-Level Experimental [63] Mariani, V. C.; Barbosa de Lima, A. G.; dos Santos
Investigation in Food Drying Technology. Drying Coelho, L. Apparent Thermal Diffusivity Estimation
Technol. 2020, 38, 557–576. DOI: 10.1080/ of the Banana during Drying Using Inverse
07373937.2019.1657145. Method. J. Food Eng. 2008, 85, 569–579. DOI: 10.
[51] Khan, M. I. H.; Wellard, R. M.; Nagy, S. A.; 1016/j.jfoodeng.2007.08.018.
Joardder, M.; Karim, M. Experimental Investigation [64] Kaveh, M.; Rasooli Sharabiani, V.; Amiri Chayjan,
of Bound and Free Water Transport Process during R.; Taghinezhad, E.; Abbaspour-Gilandeh, Y.;
Drying of Hygroscopic Food Material. Int. J.
Golpour, I. ANFIS and ANNs Model for Prediction
Therm. Sci. 2017, 117, 266–273. DOI: 10.1016/j.
of Moisture Diffusivity and Specific Energy
ijthermalsci.2017.04.006.
Consumption Potato, Garlic and Cantaloupe
[52] Sun, Q.; Zhang, M.; Mujumdar, A. S. Evaluation of
Drying under Convective Hot Air Dryer. Inf.
Potential Application of Artificial Intelligent
Process. Agric. 2018, 5, 372–387. DOI: 10.1016/j.
Control Aided by LF-NMR in Drying of Carrot as
inpa.2018.05.003.
Model Material. Drying Technol. 2020, 1–9. DOI:
[65] Hodhod, O. A.; Ahmed, H. I. Developing an
10.1080/07373937.2020.1743999
Artificial Neural Network Model to Evaluate
[53] Çakmak, G.; Yıldız, C. The Prediction of Seedy
Grape Drying Rate Using a Neural Network Chloride Diffusivity in High Performance Concrete.
Method. Comput. Electron. Agric. 2011, 75, HBRC J. 2013, 9, 15–21. DOI: 10.1016/j.hbrcj.2013.
132–138. DOI: 10.1016/j.compag.2010.10.008. 04.001.
[54] €
Menlik, T.; Ozdemir, M. B.; Kirmaci, V. [66] Chudzik, S. Measurement of Thermal Diffusivity of
Determination of Freeze-Drying Behaviors of Insulating Material Using an Artificial Neural
Apples by Artificial Neural Network. Expert Syst. Network. Meas. Sci. Technol. 2012, 23, 065602.
Appl. 2010, 37, 7669–7677. DOI: 10.1016/j.eswa. DOI: 10.1088/0957-0233/23/6/065602.
2010.04.075. [67] Lashkarbolooki, M.; Hezave, A. Z.; Bayat, M.
[55] Nazghelichi, T.; Kianmehr, M. H.; Aghbashlo, M. Thermal Diffusivity of Hydrocarbons and
Prediction of Carrot Cubes Drying Kinetics during Aromatics: Artificial Neural Network Predicting
Fluidized Bed Drying by Artificial Neural Network. Model. J. Thermophys. Heat Transfer 2017, 31,
J. Food Sci. Technol. 2011, 48, 542–550. DOI: 10. 621–627. DOI: 10.2514/1.T5041.
1007/s13197-010-0166-2. [68] Grieu, S.; Faugeroux, O.; Traore, A.; Claudet, B.;
[56] Aghbashlo, M.; Kianmehr, M. H.; Nazghelichi, T.; Bodnar, J.-L. Artificial Intelligence Tools and
Rafiee, S. Optimization of an Artificial Neural Inverse Methods for Estimating the Thermal
Network Topology for Predicting Drying Kinetics Diffusivity of Building Materials. Energy Build.
of Carrot Cubes Using Combined Response Surface 2011, 43, 543–554. DOI: 10.1016/j.enbuild.2010.10.
and Genetic Algorithm. Drying Technol. 2011, 29, 020.
770–779. DOI: 10.1080/07373937.2010.538819. [69] Grieu, S.; Traore, A.; Faugeroux, O.; Claudet, B.;
[57] Li, H.; Chen, S. A Neural-Network-Based Model Thiery, F.; Bodnar, J.-L.; Perpignan, F. Recurrent
Predictive Control Scheme for Grain Dryers. Drying Neural Networks and Neuro-Inverse Approach for
Technol. 2020, 38, 1079–1091. DOI: 10.1080/ Estimating the Thermal Diffusivity of Building
07373937.2019.1611598. Materials. 11th International Building Performance
[58] Merino, A.; Alves, R.; Acebes, L.; Prada, C. Simulation Association Conference and Exhibition,
Modeling and Simulation of a Beet Pulp Dryer for Glasgow, Scotland, 2009.
 DRYING TECHNOLOGY 15

[70] Mohamed, I. O. Development of a Simple and Food Material during Drying. Food Biosci. 2018, 23,
Robust Inverse Method for Determination of 45–53. DOI: 10.1016/j.fbio.2018.03.002.
Thermal Diffusivity of Solid Foods. J. Food Eng. [84] Vasighi-Shojae, H.; Gholami-Parashkouhi, M.;
2010, 101, 1–7. DOI: 10.1016/j.jfoodeng.2010.05. Mohammadzamani, D.; Soheili, A. Predicting
002. Mechanical Properties of Golden Delicious Apple
[71] Eslamloueyan, R.; Khademi, M. H. A Neural Using Ultrasound Technique and Artificial Neural
Network-Based Method for Estimation of Binary Network. Food Anal. Methods 2020, 13, 699–705.
Gas Diffusivity. Chemom. Intell. Lab. Syst. 2010, DOI: 10.1007/s12161-019-01689-z.
104, 195–204. DOI: 10.1016/j.chemolab.2010.08.009. [85] Xiaoyu, L.; Xiaofang, W.; Wei, W.; Jun, Z.
[72] Trigui, M.; Gabsi, K.; Amri, I. E.; Helal, A. N.; Estimation of Apple Storage Quality Properties
Barrington, S. Modular Feed Forward Networks to Based on the Mechanical Properties with BP Neural
Predict Sugar Diffusivity from Date Pulp Part I. Network. Trans. Chin. Soc. Agric. Eng. 2007, 26.
Model Validation. Int. J. Food Prop. 2011, 14, [86] Pan, H.; Yang, J.; Shi, Y.; Li, T. BP Neural Network
356–370. DOI: 10.1080/10942910903191609. Application Model of Predicting the Apple
[73] Wu, H.; Fang, W.-Z.; Kang, Q.; Tao, W.-Q.; Qiao, Hardness. J. Comput. Theor. Nanosci. 2015, 12,
R. Predicting Effective Diffusivity of Porous Media 2802–2807. DOI: 10.1166/jctn.2015.4180.
from Images by Deep Learning. Sci. Rep. 2019, 9, [87] Vahedi Torshizi, M.; Khojastehpour, M.; Tabarsa,
1–12. DOI: 10.1038/s41598-019-56309-x. F.; Ghorbanzadeh, A.; Akbarzadeh, A. Investigation
[74] Sablani, S. S.; Baik, O.-D.; Marcotte, M. Neural of Physical Properties Changes of Kiwi Fruit during
Networks for Predicting Thermal Conductivity of Different Loadings, Storage, and Modeling with
Bakery Products. J. Food Eng. 2002, 52, 299–304. Artificial Neural Network. Int. J. Fruit Sci. 2020,
DOI: 10.1016/S0260-8774(01)00119-4. 1–19. DOI: 10.1080/15538362.2020.1796889
[75] Hussain, M. A.; Rahman, M. S. Thermal [88] Hobani, A. I.; Thottam, A. M.; Ahmed, K. A.
Conductivity Prediction of Fruits and Vegetables Development of a Neural Network Classifier for
Using Neural Networks. Int. J. Food Prop. 1999, 2, Date Fruit Varieties Using Some Physical
121–137. DOI: 10.1080/10942919909524596. Attributes. King Saud Univ. Agric. Res. Center.
[76] Sablani, S. S.; Rahman, M. S. Using Neural 2003, 126, 5–18.
Networks to Predict Thermal Conductivity of Food [89] Saeidirad, M. H.; Rohani, A.; Zarifneshat, S.
as a Function of Moisture Content, Temperature Predictions of Viscoelastic Behavior of Pomegranate
and Apparent Porosity. Food Res. Int. 2003, 36, Using Artificial Neural Network and Maxwell
617–623. DOI: 10.1016/S0963-9969(03)00012-7. Model. Comput. Electron. Agric. 2013, 98, 1–7.
[77] Mattar, H. L.; Minim, L. A.; Coimbra, J. S. R.; DOI: 10.1016/j.compag.2013.07.009.
Minim, V. P. R.; Saraiva, S. H.; Telis-Romero, J. [90] Qiao, J.; Wang, N.; Ngadi, M.; Kazemi, S.
Modeling Thermal Conductivity, Specific Heat, and Predicting Mechanical Properties of Fried Chicken
Density of Milk: A Neural Network Approach. Int. Nuggets Using Image Processing and Neural
J. Food Prop. 2004, 7, 531–539. DOI: 10.1081/JFP- Network Techniques. J. Food Eng. 2007, 79,
200032964. 1065–1070. DOI: 10.1016/j.jfoodeng.2006.03.026.
[78] Delgado, A. E.; Sun, D.-W.; Rubiolo, A. C. 1 [91] Saiedirad, M. H.; Mirsalehi, M. Prediction of
Thermal Physical Properties of Foods. In Thermal Mechanical Properties of Cumin Seed Using
Food Processing: New Technologies and Quality Artificial Neural Networks. J. Texture Stud. 2010,
Issues, Da-Wen Sun, Ed.; CRC Press: Boca Raton, 41, 34–48. DOI: 10.1111/j.1745-4603.2009.00211.x.
2005, pp 18–49. [92] Azadbakht, M.; Torshizi, M. V.; Ziaratban, A.;
[79] Heldman, D. R.; Lund, D. B.; Sabliov, C. Handbook Ghajarjazi, E. Application of Artificial Neural
of Food Engineering. CRC Press: Boca Raton, 2018. Network (ANN) in Predicting Mechanical
[80] Rahman, M. S.; Rashid, M. M.; Hussain, M. A. Properties of Canola Stem under Shear Loading.
Thermal Conductivity Prediction of Foods by Agric. Eng. Int. CIGR J. 2016, 18, 413–425.
Neural Network and Fuzzy (ANFIS) Modeling [93] Vasquez, N.; Magan, C.; Oblitas, J.; Chuquizuta, T.;
Techniques. Food Bioprod. Process. 2012, 90, Avila-George, H.; Castro, W. Comparison between
333–340. DOI: 10.1016/j.fbp.2011.07.001. Artificial Neural Network and Partial Least Squares
[81] Paluri, S.; Phinney, D. M.; Heldman, D. R. Recent Regression Models for Hardness Modeling during
Advances in Thermophysical Properties— the Ripening Process of Swiss-Type Cheese Using
Measurements, Prediction, and Importance. Curr. Spectral Profiles. J. Food Eng. 2018, 219, 8–15.
Opin. Food Sci. 2018, 23, 142–148. DOI: 10.1016/j. DOI: 10.1016/j.jfoodeng.2017.09.008.
cofs.2018.10.005. [94] Kaveh, M.; Abbaspour-Gilandeh, Y.; Chayjan, R. A.;
[82] Khan, M. I. H.; Patel, N.; Mahiuddin, M.; Karim, Taghinezhad, E.; Mohammadigol, R. Mass Transfer,
M. A. Characterisation of Mechanical Properties of Physical, and Mechanical Characteristics of
Food Materials during Drying Using Terebinth Fruit (Pistacia atlantica L.) under
Nanoindentation. J. Food Eng. 2021, 291, 110306. Convective Infrared Microwave Drying. Heat Mass
DOI: 10.1016/j.jfoodeng.2020.110306. Transfer 2018, 54, 1879–1899. DOI: 10.1007/
[83] Mahiuddin, M.; Khan, M. I. H.; Pham, N. D.; s00231-018-2287-5.
Karim, M. Development of Fractional Viscoelastic [95] Khan, M. I. H.; Farrell, T.; Nagy, S.; Karim, M.
Model for Characterizing Viscoelastic Properties of Fundamental Understanding of Cellular Water
16 M. I. H. KHAN ET AL.

Transport Process in Bio-Food Material during [107] Egmont-Petersen, M.; de Ridder, D.; Handels, H.
Drying. Sci. Rep. 2018, 8, 15191. DOI: 10.1038/ Image Processing with Neural Networks—A
s41598-018-33159-7. Review. Pattern Recognit. 2002, 35, 2279–2301.
[96] Khan, M. I. H.; Nagy, S. A.; Karim, M. Transport DOI: 10.1016/S0031-3203(01)00178-9.
of Cellular Water during Drying: An [108] Ansari, S.; Maftoon-Azad, N.; Farahnaky, A.;
Understanding of Cell Rupturing Mechanism in Hosseini, E.; Badii, F. Effect of Moisture Content
Apple tissue. Food Res. Int. 2018, 105, 772–781. on Textural Attributes of Dried Figs. Int. Agrophys.
DOI: 10.1016/j.foodres.2017.12.010. 2014, 28, 403–412. DOI: 10.2478/intag-2014-0031.
[97] Joardder, M. U.; Kumar, C.; Karim, M. Food [109] Sun, S.; Cao, Z.; Zhu, H.; Zhao, J. A Survey of
Structure: Its Formation and Relationships with Optimization Methods from a Machine Learning
Other Properties. Crit. Rev. Food Sci. Nutr. 2017, Perspective. IEEE Trans. Cybern. 2020, 50,
57, 1190–1205. DOI: 10.1080/10408398.2014. 3668–3681.
971354. [110] Yang, X.-S. Chapter 5—Genetic Algorithms. In
[98] Lopez-Sanchez, P.; Svelander, C.; Bialek, L.; Nature-Inspired Optimization Algorithms, Yang, X.-
Schumm, S.; Langton, M. Rheology and S., Ed. Elsevier: Oxford, 2014; pp 77–87.
Microstructure of Carrot and Tomato Emulsions as [111] Morimoto, T.; De Baerdemaeker, J.; Hashimoto, Y.
a Result of High-Pressure Homogenization Optimization of Storage System of Fruits Using
Conditions. J. Food Sci. 2011, 76, E130–E140. DOI: Neural Networks and Genetic Algorithms.
10.1111/j.1750-3841.2010.01894.x. Proceedings of 1995 IEEE International Conference
[99] Devahastin, S. Food Microstructure and Its on Fuzzy Systems, IEEE, Yokohama, Japan, 1995;
Relationship with Quality and Stability. Sawston, Vol. 1, pp 289–294.
United Kingdom: Woodhead Publishing, 2017. [112] Curvelo Santana, J.; Ara ujo, S.; Librantz, A.;
[100] Pascua, Y.; Koç, H.; Foegeding, E. A. Food Tambourgi, E. Optimization of Corn Malt Drying
Structure: Roles of Mechanical Properties and Oral by Use of a Genetic Algorithm. Drying Technol.
Processing in Determining Sensory Texture of Soft 2010, 28, 1236–1244. DOI: 10.1080/07373937.2010.
Materials. Curr. Opin. Colloid Interface Sci. 2013, 500439.
18, 324–333. DOI: 10.1016/j.cocis.2013.03.009. [113] Rahman, M. M.; Mustayen, A. G. M. B.; Mekhilef,
[101] Martynenko, A.; Janaszek, M. A. Texture Changes S.; Saidur, R. The Optimization of Solar Drying of
during Drying of Apple Slices. Drying Technol. Grain by Using a Genetic Algorithm. Int. J. Green
2014, 32, 567–577. DOI: 10.1080/07373937.2013. Energy 2015, 12, 1222–1231. DOI: 10.1080/
845573. 15435075.2014.890106.
[102] Thussu, S.; Datta, A. K. Texture Prediction during [114] Devahastin, S.; Niamnuy, C. Invited Review:
Deep Frying: A Mechanistic Approach. J. Food Eng. Modelling Quality Changes of Fruits and
2012, 108, 111–121. DOI: 10.1016/j.jfoodeng.2011. Vegetables during Drying: A Review. Int. J. Food
07.017. Sci. Technol. 2010, 45, 1755–1767. DOI: 10.1111/j.
[103] Kerdpiboon, S.; Kerr, W. L.; Devahastin, S. Neural 1365-2621.2010.02352.x.
Network Prediction of Physical Property Changes [115] Fathi, M.; Mohebbi, M.; Razavi, S. M. A.
of Dried Carrot as a Function of Fractal Dimension Application of Fractal Theory for Prediction of
and Moisture Content. Food Res. Int. 2006, 39, Shrinkage of Dried Kiwifruit Using Artificial Neural
1110–1118. DOI: 10.1016/j.foodres.2006.07.019. Network and Genetic Algorithm. Drying Technol.
[104] Castro, W.; Yoshida, H.; Gil, L. S.; Lopez, L. M.; 2011, 29, 918–925. DOI: 10.1080/07373937.2011.
Oblitas, J.; De-la-Torre, M.; Avila-George, H. 553755.
Microstructural Analysis in Foods of Vegetal [116] Sun, Y.; Zhang, M.; Ju, R.; Mujumdar, A. Novel
Origin: An Approach with Convolutional Neural Nondestructive NMR Method Aided by Artificial
Networks. 2019 8th International Conference on Neural Network for Monitoring the Flavor Changes
Software Process Improvement (CIMPS), IEEE, of Garlic by Drying. Drying Technol. 2020, 1–12.
2019; pp 1–5. DOI: 10.1109/CIMPS49236.2019. DOI: 10.1080/07373937.2020.1821211.
9082421. [117] Chen, C.; Ramaswamy, H.; Alli, I. Prediction of
[105] Oblitas-Cruz, J. F.; Castro-Silupu, W. M.; Mayor- Quality Changes during Osmo-Convective Drying
L
opez, L, Effect of Different Combinations of Size of Blueberries Using Neural Network Models for
and Shape Parameters in the Percentage Error of Process Optimization. Drying Technol. 2001, 19,
Classification of Structural Elements in Vegetal 507–523. DOI: 10.1081/DRT-100103931.
Tissue of the Pumpkin Cucurbita pepo L. Using [118] Scala, K. D.; Meschino, G.; Vega-Galvez, A.;
Probabilistic Neural Networks. Rev. Fac. Ing. Lemus-Mondaca, R.; Roura, S.; Mascheroni, R. An
Antioquia 2015, 78, 30–37. DOI: 10.17533/udea. Artificial Neural Network Model for Prediction of
redin.n78a04. Quality Characteristics of Apples during Convective
[106] Sansiribhan, S.; Devahastin, S.; Soponronnarit, S. Dehydration. Food Sci. Technol. 2013, 33, 411–416.
Quantitative Evaluation of Microstructural Changes DOI: 10.1590/S0101-20612013005000064.
and Their Relations with Some Physical [119] Barzegar, M.; Zare, D.; Stroshine, R. L. An
Characteristics of Food during Drying. J. Food Sci. Integrated Energy and Quality Approach to
2010, 75, E453–E461. DOI: 10.1111/j.1750-3841. Optimization of Green Peas Drying in a Hot Air
2010.01739.x. Infrared-Assisted Vibratory Bed Dryer. J. Food Eng.
 DRYING TECHNOLOGY 17

2015, 166, 302–315. DOI: 10.1016/j.jfoodeng.2015. [126] Koc, B.; Yilmazer, M. S.; Balkır, P.; Ertekin, F. K.
06.026. Spray Drying of Yogurt: Optimization of Process
[120] Youssefi, S.; Emam-Djomeh, Z.; Mousavi, S. Conditions for Improving Viability and Other
Comparison of Artificial Neural Network (ANN) Quality Attributes. Drying Technol. 2010, 28,
and Response Surface Methodology (RSM) in the 495–507. DOI: 10.1080/07373931003613809.
Prediction of Quality Parameters of Spray-Dried [127] Khan, M. I. H.; Welsh, Z.; Gu, Y.; Karim, M.;
Pomegranate Juice. Drying Technol. 2009, 27, Bhandari, B. Modelling of Simultaneous Heat and
910–917. DOI: 10.1080/07373930902988247. Mass Transfer Considering the Spatial Distribution
[121] Onwude, D. I.; Hashim, N.; Abdan, K.; Janius, R.; of Air Velocity during Intermittent Microwave
Chen, G. The Potential of Computer Vision, Convective Drying. Int. J. Heat Mass Transf. 2020,
Optical Backscattering Parameters and Artificial 153, 119668. DOI: 10.1016/j.ijheatmasstransfer.2020.
Neural Network Modelling in Monitoring the 119668.
Shrinkage of Sweet Potato (Ipomoea batatas L.) [128] Apicella, R.; Romano, V. Comparison between
during Drying. J Sci Food Agric 2018, 98, Lambert and Maxwell Approaches in the Modelling
1310–1324. DOI: 10.1002/jsfa.8595. of Microwave Heating of Liquid Foods. ENG. 2019,
[122] Banooni, S.; Hosseinalipour, S.; Mujumdar, A.;
11, 1–13. DOI: 10.4236/eng.2019.111001.
Taherkhani, P.; Bahiraei, M. Baking of Flat Bread
[129] Pham, N. D.; Khan, M.; Karim, M. A Mathematical
in an Impingement Oven: Modeling and
Model for Predicting the Transport Process and
Optimization. Drying Technol. 2009, 27, 103–112.
Quality Changes during Intermittent Microwave
DOI: 10.1080/07373930802565954.
Convective Drying. Food Chem. 2020, 325, 126932.
[123] Liu, Z.-L.; Wei, Z.-Y.; Vidyarthi, S. K.; Pan, Z.;
DOI: 10.1016/j.foodchem.2020.126932.
Zielinska, M.; Deng, L.-Z.; Wang, Q.-H.; Wei, Q.;
[130] Duc Pham, N.; Khan, M. I. H.; Joardder, M.;
Xiao, H.-W. Pulsed Vacuum Drying of Kiwifruit
Slices and Drying Process Optimization Based on Rahman, M.; Mahiuddin, M.; Abesinghe, A. N.;
Artificial Neural Network. Drying Technol. 2020, Karim, M. Quality of Plant-Based Food Materials
1–13. DOI: 10.1080/07373937.2020.1817063. and Its Prediction during Intermittent Drying. Crit.
[124] Mihajlovic, T.; Ibric, S.; Mladenovic, A. Application Rev. Food Sci. Nutr. 2019, 59, 1115–1197. DOI: 10.
of Design of Experiments and Multilayer 1080/10408398.2017.1399103.
Perceptron Neural Network in Optimization of the [131] Mebatsion, H. K.; Verboven, P.; Ho, Q. T.;
Spray-Drying Process. Drying Technol. 2011, 29, Verlinden, B.; Nicolaï, B. Modelling Fruit (Micro)
1638–1647. DOI: 10.1080/07373937.2011.592960. Structures, Why and How? Trends Food Sci.
[125] Khawas, P.; Dash, K. K.; Das, A. J.; Deka, S. C. Technol. 2008, 19, 59–66. DOI: 10.1016/j.tifs.2007.
Modeling and Optimization of the Process 10.003.
Parameters in Vacuum Drying of Culinary Banana [132] Mahiuddin, M.; Khan, M. I. H.; Kumar, C.;
(Musa ABB) Slices by Application of Artificial Rahman, M.; Karim, M. Shrinkage of Food
Neural Network and Genetic Algorithm. Drying Materials during Drying: Current Status and
Technol. 2016, 34, 491–503. DOI: 10.1080/ Challenges. Compr. Rev. Food Sci. Food Saf. 2018,
07373937.2015.1060605. 17, 1113–1126. DOI: 10.1111/1541-4337.12375.