materials

Article
Model-Based Adaptive Machine Learning Approach in
Concrete Mix Design
Patryk Ziolkowski 1, * , Maciej Niedostatkiewicz 1 and Shao-Bo Kang 2,3

1 Faculty of Civil and Environmental Engineering, Gdansk University of Technology,

Gabriela Narutowicza 11/12, 80-233 Gdansk, Poland; mniedost@pg.edu.pl
2 School of Civil Engineering, Chongqing University, Chongqing 400045, China; kang0119@cqu.edu.cn
3 Key Laboratory of New Technology for Construction of Cities in Mountain Area, Chongqing University,
Chongqing 400045, China
* Correspondence: patziolk@pg.edu.pl; Tel.: +48-58-347-2385

Abstract: Concrete mix design is one of the most critical issues in concrete technology. This process
aims to create a concrete mix which helps deliver concrete with desired features and quality. Contem-
porary requirements for concrete concern not only its structural properties, but also increasingly its
production process and environmental friendliness, forcing concrete producers to use both chemically
and technologically complex concrete mixtures. The concrete mix design methods currently used in
engineering practice are joint analytical and laboratory procedures derived from the Three Equation
Method and do not perform well enough for the needs of modern concrete technology. This often
causes difficulties in predicting the final properties of the designed mix and leads to precautionary
oversizing of concrete properties for fear of not providing the required parameters. A new approach
that would make it possible to predict the newly designed concrete mix properties is highly desirable.





 The answer to this challenge can be methods based on machine learning, which have been intensively
developed in recent years, especially in predicting concrete compressive strength. Machine learning-
Citation: Ziolkowski, P.;
based methods have been more or less successful in predicting concrete compressive strength, but
Niedostatkiewicz, M.; Kang, S.-B.
Model-Based Adaptive Machine
they do not reflect well the variability that characterises the currently used concrete mixes. A new
Learning Approach in Concrete Mix adaptive solution that allows estimating concrete compressive strength on the basis of the concrete
Design. Materials 2021, 14, 1661. mix main ingredient composition by including two observations for a given batch of concrete is
https://doi.org/10.3390/ma14071661 proposed herein. In presented study, a machine learning model was built with a deep neural network
architecture, trained on an extensive database of concrete recipes, and translated into a mathematical
Academic Editors: Łukasz Sadowski formula. Testing on four concrete mix recipes was performed, which were calculated according
and Qing-feng Liu to contemporary design methods (Bolomey and Fuller method), and a comparative analysis was
conducted. It was found out that the new algorithm performs significantly better than that without
Received: 30 January 2021
adaptive features trained on the same dataset. The presented algorithm can be used as a concrete
Accepted: 24 March 2021
strength checking tool for the concrete mix design process.
Published: 28 March 2021

Keywords: applied machine learning; concrete; concrete mix design; concrete strength prediction;
Publisher’s Note: MDPI stays neutral
data mining
with regard to jurisdictional claims in
published maps and institutional affil-
iations.

1. Introduction
Concrete mix is a mixture of cement, water, and fine-grained and coarse-grained
Copyright: © 2021 by the authors.
aggregate, as well as additives and admixtures. Additives and admixtures are designed to
Licensee MDPI, Basel, Switzerland.
improve the chemical characteristics and performance parameters of concrete, especially
This article is an open access article
compressive strength, durability, and workability. One can distinguish many such supple-
distributed under the terms and ments, including accelerating admixtures, admixtures for improved fresh concrete prop-
conditions of the Creative Commons erties, durability-enhancing admixtures, fibers for concrete strengthening, set-retarding
Attribution (CC BY) license (https:// admixtures, and water-reducing admixtures. The concrete mix’s appropriate design is one
creativecommons.org/licenses/by/ of the most critical issues in the construction process, which is considered on many levels.
4.0/). The concrete mix must be designed economically. This means that it must allow obtaining

Materials 2021, 14, 1661. https://doi.org/10.3390/ma14071661 https://www.mdpi.com/journal/materials

Materials 2021, 14, 1661 2 of 34

the desired properties at the lowest possible cost of raw materials. A concrete mixture
is dedicated to a specific technological process, where properties such as workability or
cement setting speed are vital. Subsequently, there are environmental conditions related
to precipitation, temperature during concreting, the distance from the construction site,
and the traffic volume. The composition of the concrete mix significantly depends on the
assumed construction specification resulting from the construction design, in which it is
assumed, for example, the compressive strength of concrete, or environmental aggression,
such as chloride ingression. The last important factor determining the composition of the
concrete mix is ecological considerations, which have recently gained particular impor-
tance. There are currently many solutions to reduce the carbonation of concrete, such as
admixtures of graphene nanoparticles [1].
In conclusion, it can be said that designing a concrete mix comes down to the appro-
priate selection of the proportions of the primary and secondary components in order to
obtain concrete with the desired properties. During the technological process of concrete
production, the concrete mix is transported to the construction site and placed in the
concrete formwork; then, the process of concrete hardening and gaining strength occurs.
The hydration of the cement initiates the concrete hardening process. Cement hydration
is an exothermic chemical reaction that occurs when cement comes in contact with water.
After starting the hydration process, the cement forms tobermorite gel, hydroxide, and
other ingredients, which enhance adhesion between the fine-grained and coarse-grained
aggregate. Throughout this process, hydration products continuously deposit on the ce-
ment grains and fill the area occupied by the water. The final stage of the hydration process
is when all water molecules are bound, or there is no more unreacted cement. Hardened
concrete obtains partial compressive strength a few days after the hydration process starts,
and most of compressive strength after around 28 days (some types of concrete reach their
full strength later) [2–4]. The amount of water needed to hydrate the cement entirely varies
between 20% and 25% by weight, not including the water trapped in the pores [5,6]. In
keeping with Power’s model, the water required to hydrate cement is 42% by weight [7,8].
Concrete mix design methods, which are currently used in engineering practice, derive
from solutions developed over a dozen years ago and are based on an estimation of con-
crete mortar strength for bending [9–11]. The practical application of these methods is
laborious and ineffective, and it does not consider the chemical complexity and variability
of modern concrete mixtures.
The presented field needs new technological solutions that will face current chal-
lenges, and the answer might lay in machine learning-based methods that are more or
less successful in predicting concrete compressive strength; however, they do not reflect
well the variability of new chemically complex concrete mixes. This paper presents a
novel adaptive approach that allows estimating concrete compressive strength on the
basis of the concrete mix main ingredient composition by including two observations for a
given batch of concrete. The proposed solution is a deep neural network with 46 hidden
neurons, clustered in seven layers. The machine learning model was built, trained on an
extensive database of concrete recipes, and translated into a mathematical formula. Testing
on four concrete mix recipes was performed, which were also calculated according to the
Bolomey and Fuller method, and a comparative analysis was conducted. The presented
algorithm can be used as a concrete strength checking tool for the concrete mix design
process. Machine learning is part of a rapidly developing field of knowledge, within the
broader domain of artificial intelligence. It enables systems to automatically learn and
improve from experience without direct programming and specific tasks. There are many
different approaches to machine learning, among which the most popular are artificial
neural networks (ANNs), with many branches such as deep neural networks (DNNs),
recurrent neural networks (RNNs), and convolutional neural networks (CNNs). There are
also other approaches, such as gene expression programming (GEP) and multi expression
programming (MEP) [12–14]. The basic unit of an ANN is an artificial neuron, grouped into
clusters. A single artificial neuron can be treated as a recitative information carrier, just like
Materials 2021, 14, 1661 3 of 34

a human neuron excitably conducts an electrical signal. Artificial neurons in the network
somewhat mimic the behavior of the human brain. The clusters of neurons are grouped into
layers. The ANN typically includes at least three layers, an input layer, a hidden layer, and
an output layer. The input and output layers consist of input neurons and output neurons,
respectively. These neurons represent the variables that we want to introduce into the
algorithm and the variables resulting from the algorithm’s execution. The neurons process
the input data with the appropriate weight in the hidden layer and provide the output by
the activation function. Depending on the type, specificity, amount of data, and machine
learning architecture, hidden layers may function differently. In some methods, the input
variable’s weight is assigned randomly, and, in others, the weight is meticulously calibrated
by backpropagation. There is also a weight control mechanism, called the “learning rule”.
It can be said that a greater number of hidden layers allows ANNs to solve more difficult
problems, but more computing power and time will be needed for calculations. ANNs are
taught how to solve a problem upon being fed examples. ANNs are proven effective in
finding patterns that would be difficult for human recognition. A deep neural network
(DNN) is an artificial neural network (ANN) with multiple layers of hidden neurons, and
it is characterized by a high level of complexity [15–18].

2. Concrete Mix Design and Machine Learning

2.1. Contemporary Engineering Practice in Concrete Mix Design
Concrete mix design is a complex issue, often requiring extensive knowledge of con-
crete technology and vast experience. The main task in the design process is to select
appropriate material compositions to obtain a concrete mix with desired properties, both
in the form of a fresh mix during transport and concreting and in the form of hardened
concrete. Specific properties can be expected at each stage of the concrete structure manu-
facturing process. Several features characterize concrete performance, such as plasticity,
durability, compressive strength, and modulus of elasticity. The properties mentioned
above have different priorities at different times; for example, adequate compressive
strength is essential from the point of view of designed ultimate limit state, while ade-
quate durability is essential in an aggressive environment [19–22]. Designing a mix with
the improper specification can have many serious consequences; therefore, concrete mix
producers fearing failure to meet the appropriate criteria often deliberately augment its
parameters beyond designed values. This leads to the phenomenon called “concrete su-
perstrength”, which on the surface may seem beneficial, because, for example, in a case of
concrete compressive strength, the strength of concrete is higher; however, the disturbed
stiffness of the structure may cause the structure to behave differently than envisaged by
the designer [23].
Corporate engineering practice varies across the world, while also sharing significant
similarities. In the European Union, the norm governing concrete technology issues is “EN
206 Concrete: Specification, performance, production, and conformity”, while the design
of concrete structures is described in the standard “EN 1992-1-1: Eurocode 2: Design of
concrete structures”. There are equivalents and national appendices for each standard, e.g.,
in Great Britain, the BS EN 206: 2013 + A1: 2016 standard is used, while, in Poland, the PN-
EN 206 + A1: 2016-12 standard is used. Depending on the member state of the European
Union, various methods of designing concrete mix are popular. In Poland, the methods of
Bukowski, Eyman, Klaus, Kopycinski and Paszkowski are most often used, along with the
so-called double coating method [24]. On the other hand, in the United States of America,
Bolomey, Fuller, and the 0.45 power gradation chart methods are the most popular. Most
of these methods are derived from the so-called “Three Equations Method”, a merged
experimental–analytical approach [11,25,26]. The experimental–analytical approach means
that we need to calculate the volume of ingredients needed by an analytical method and
validate it using destructive laboratory tests. This method allows us to determine the
amount of cement, water, and aggregate by weight per unit volume, using three equations
Materials 2021, 14, 1661 4 of 34

of consistency, strength, and water-tightness. Consistency Equation (1) is incorporated into

the water-demand formula, which helps to find the desired consistency.

W = C· w c + K· w k [l], (1)

where W is the amount of water in 1 m3 of concrete, expressed in units of volume (in this
case, L), C represents the weight of cement in 1 m3 of concrete, expressed in kg, wc is a
cement–water demand index which denotes the amount of water in 1 dm3 that should be
added to 1 kg of a given class of cement, expressed in L/kg, K corresponds to the weight of
aggregate in 1 m3 of concrete, expressed in kg, and wk is an aggregate–water demand index
which denotes the amount of water in 1 dm3 that should be added to 1 kg of dry aggregate
of a certain fraction to obtain the desired consistency, expressed in L/kg. The cement–water
demand index and aggregate–water demand index depend on the grain size, shape, surface
roughness, proportion in a given composition, and required consistency of the concrete
mix. The water demand for concrete additives and admixtures is considered by adding
it to aggregate or cement depending on the grain size. The cement–water and aggregate–
water demand indices were developed by Stern and Bolomey [27,28]. The next equation is
called the concrete compressive strength equation, which comes in two versions, Bolomey
and Feret. This equation describes the relationship between the compressive strength of
concrete and parameters such as the water–cement ratio and the grade of cement and
aggregate. Equation (2) is the Feret version of the concrete compressive strength equation.
Equation (3) is the Bolomey version of the concrete compressive strength equation.

C
f cm = A[( + p) − a] [MPa], (2)
W
C
f cm = A1,2 ( ± a) [MPa], (3)
W
where f cm is a medium concrete compressive strength, expressed in MPa, and A, A1,2 are
coefficients that depend on the type and strength class of the aggregate and the strength
class of the cement. The coefficient A1 is taken when C/W < 2.5 and A2 is taken when
C / W > 2.5. C represents the weight of cement in 1 m3 of concrete, expressed in kg, W
represents the amount of water in 1 m3 of concrete, expressed in L; p is the amount of air in
1 m3 of concrete, expressed in dm3 , and a is a numerical value depending on the quality
of cement and aggregate, and it can be taken as a constant equal to 0.5. The a value is
positive when the water–cement ratio is greater than or equal to 2.5 and negative when the
water–cement ratio is less than 2.5. The Feret equation is valid when the aggregate strength
is lower than the grout strength and applies to porous concrete. Lastly, Equation (4) is
called the water-tightness equation, which tells us that the volume sum of the individual
components is equal to the entire concrete mix volume.

C K
+ + W = 1000 [dm3 ], (4)
ρc ρk

where W refers to the amount of water in 1 m3 of concrete, expressed in L, C represents the

weight of cement in 1 m3 of concrete, expressed in kg, ρc is the cement density in kg/dm3 ,
K is the weight of cement in 1 m3 of concrete, expressed in kg, and ρk is the aggregate
density in kg/dm3 . The quantitative composition of the concrete mix, considered as the
amount of cement, water, and aggregate in 1 m3 of mixture, can be calculated using the
equations described above. The Three Equations Method has certain boundary conditions.
The concrete mix porosity should not exceed 0.002 of the mix volume without air-entraining
admixtures or 0.008 of the mix volume using air-entraining admixtures.
The entire process of concrete mix design consists of the following stages: determining
the initial assumptions, determining the required properties of hardened concrete and
fresh concrete mix, selecting and evaluating the components of the concrete mix, designing
the mix composition, checking the technical characteristics of the fresh concrete mix and
Materials 2021, 14, 1661 5 of 34

hardened concrete in a laboratory, and preparing a working recipe. During the formulation
of initial assumptions, several factors need to be considered, such as the concrete mix’s
intended use, which depends on properties of the newly designed structure, including the
location, amount of reinforcement, and geometric characteristics of the cross-section. The
primary technical characteristics of fresh concrete mix are the bulk density, consistency, and
air content, whereas those for hardened concrete are the frost resistance, fire resistance, and
class of concrete compressive strength. It is necessary to analyze the technological process,
as well as assess the conditions of concrete maturation and the method of compacting
fresh concrete mix. The concrete exposure class, which corresponds to the degree and
type of environmental aggression and additional properties, such as concrete tightness,
should also be specified. It is necessary to determine maximum aggregate diameter and
mix workability. The concrete mix components should be selected and evaluated, including
the proper type of cement, appropriate water, and aggregate quality, characterized by the
relevant standards. After designing the concrete mix composition and laboratory tests, the
last part of the process is preparing a working recipe for 1 m3 of concrete mix. It is also
vital to consider the recipe changes that may result from the dampness of the aggregate
and adapt it to individual conditions, such as the capacity of a transport vehicle [29,30].

2.2. Machine Learning in Prediction of Concrete Features

Machine learning is used in many science areas, from forecasting real-estate prices to
identifying conditions on the basis of computed tomography images. One of these areas is
civil and structural engineering. In civil and structural engineering, machine learning is
used in structural health monitoring, crack detection, life-cycle cost analysis, prediction of
soil compression coefficient, and many more. In forecasting concrete properties, research
focuses on predicting a concrete’s compressive strength, which is one of the most critical
parameters of concrete and defines its class.
The challenge in predicting concrete strength by machine learning was first described
by Yeh et al. [31] in 1998. Using seven input variables, they performed ANN and linear re-
gression to predict the strength of high-strength concrete. Yeh et al. trained their algorithm
on many concrete samples, but they were not filtered in terms of content. Their analysis
took into account the concrete samples in the maturing phase, even those that were 3 days
old, which may have significantly distorted the results.
In 2003, the topic was developed further by Seung-Chang Lee [32]. He utilized a
modular network architecture, which consisted of five ANNs with unique architectures.
These unique architectures corresponded to concrete in different maturation phases up
until achieving full concrete strength. To estimate the number of neurons in the input
layer, he used the parameter condensation technique. Seung-Chang Lee claimed that the
condensation and weighing techniques he used are useful in finding optimal network
performance. However, since his ANN models the maturation process from the moment
after pouring to reaching full strength, it has no practical application. From an engineering
practice point of view, the focus should be on concrete that has reached its full or most of
its strength.
In 2005, Hola J. and Schabowicz K. [33,34] presented an attractive nondestructive
concrete strength assessment approach. They used an ANN model trained not on the
concrete mix composition, but on the data collected by nondestructive concrete testing
equipment. Their database contained ultrasonic wave velocity, reflection number, hardness,
pull-out strength, concrete age, and bulk density. To obtain the laboratory results, they
tested concrete compressive strength samples with a 28 day strength of 24–105 MPa. They
created the ANN with eight hidden neurons grouped in one layer, using the Levenberg–
Marquardt training method. The authors claimed that the average concrete compressive
strength compared between the ANN and nondestructive tests was similar.
In 2006, Gupta et al. [35] proposed using a neural-expert system to predict the com-
pressive strength of high-performance concrete. In their method, Gupta et al. focused on
training the algorithm using example inferences and used a multilayer ANN trained with
Materials 2021, 14, 1661 6 of 34

generalized backpropagation for interval training patterns. This may lead to algorithm
training based on patterns with insignificant variables. They also used input variables
of completely different metrics not strictly related to the recipe, such as curing time, and
focused on basics such as the concrete mix composition, which may have an unclear effect
on results. The neural-expert system in concrete compressive strength prediction was also
discussed by Dac-Khuong Bui et al. [36], where they focused on the practical application of
this approach.
In 2018, Fangming Deng et al. [37] introduced deep learning to the subject. Fangming
Deng et al. for algorithm training purposes prepared a database of recycled concrete sam-
ples. Their database provided five input variables, such as fly ash replacement, recycled
coarse-grained aggregate replacement ratio, recycled fine-grained aggregate replacement
ratio, and water–cement ratio, used to train the machine learning algorithm. They decided
not to train the algorithm on the concrete mix composition with a direct amount of in-
dividual components, but on the several ratios, which they referred to as deep features.
In this study, a similar approach was used by introducing feature scaling. They used
Softmax regression to look for a suitable prediction model. Fangming Deng et al. claimed
that the introduction of deep learning compared to ANN provided better generalization
capabilities, superior efficiency, and higher precision. However, this was not apparent and
should be the subject of more extensive research. First, convolution neural networks are
computationally expensive, as evidenced by the author’s adoption of a limited database.
They used 74 samples in comparison to the 741 samples in this analysis. The limited num-
ber of samples may result in underfitting, which means that the model does not properly
represent the modeled phenomenon. A similar level of accuracy between artificial neural
networks and deep neural networks was presented by Hosein Naderpour et al. [38] in his
study from 2018.
In 2019, Ziolkowski P. et al. [16] presented an algorithm, which supports designing a
concrete mix by predicting the strength of concrete based on the composition of the concrete
mix. The algorithm gave a quite right prediction of concrete mix strength. However, the
paper’s algorithm gave a weak approximation for the high-strength spectrum of 40 MPa
and above. It was also poorly able to predict the properties of mixtures with concrete
additives and admixtures. Other essential parameters that contribute to proper concrete
performance, such as durability, which is vital to maintain the service quality of structure
in time, were not recognized in the study [39].
In 2020, Adil M. et al. [40] presented a paper in which they studied the effect of the
number of neurons and layers in ANN for generalized concrete mix design. They used
ANN with 17 inputs, such as the specific gravity of concrete mix ingredients, dry density
of aggregates, type of cement and mineral admixtures, water–cement ratio, modulus of
elasticity, and tensile and compressive strength of concrete, as well as five outputs, such as
cement, water, and fine-grained and coarse-grained aggregate content. The authors argued
that this network performed best with one or two hidden layers. It is an entirely different
approach to a large number of previous works, where the technical parameters of concrete
were predicted on the basis of the composition ratio.
In 2020, Nunez I. et al. [41] presented a study in which they built a machine learning
model to predict the recycled aggregate concrete compressive strength and optimize the
concrete mix design process. A reliable optimization method for concrete mix design is
especially significant for recycled aggregate concrete, due to its variability and lack of
proper compressive strength estimation formulas. The authors developed three distinctive
machine learning models, namely, the Gaussian processes model, recurrent neural network
model, and gradient boosted regression trees model, and they claimed to achieve robust
predictive performance. They obtained the best performance using the gradient boosted
regression trees model.
In 2020, Marani A. et al. [42] presented a solution to predict the compressive strength
of ultra-high-performance concrete using a machine learning algorithm. They trained their
algorithm on a database of 810 samples gathered from open-access sources. The database
Materials 2021, 14, 1661 7 of 34

consisted of 15 variables that were taken as input data. The authors used an unusual
technique, whereby, thanks to their database, they generated 6513 plausible synthetic data
samples using tabular generative adversarial networks. Such a large pool of data allowed
for robust training of their machine learning model. The authors found that their model
trained on synthetic data achieved outstanding predictive performance when tested on the
primary dataset.

3. Materials and Methods

3.1. Essentials
Most of the solutions described in the literature that could support the concrete mix
design process consist of predicting the concrete compressive strength as a function of
the concrete mix composition or basic technical parameters. The previously mentioned
approaches could be improved by introducing an algorithm calibrated for a specific batch
of concrete. This is essential for several reasons, the most important of which is the chemical
complexity caused by various concrete admixtures, which is currently a standard. There
are many different concrete admixtures, such as accelerating admixtures, admixtures for
improved fresh concrete properties, durability-enhancing admixtures, fibers for concrete
strengthening, set-retarding admixtures, water-reducing admixtures, and novel additions
of nanomaterials such as graphene. These admixtures can significantly affect the obtained
concrete parameters, especially compressive strength. It is also difficult to predict how a
mixture with many different admixtures or admixtures produced by different manufac-
turers will behave in service. The second reason concerns the variability caused by the
use of raw materials from different manufacturers, from different mining sources, and
characterized by different properties. The variability can manifest itself on many levels,
from the shape of the coarse aggregate to the presence of clay in the fine-grained aggregate.
The algorithm’s primary task is to estimate concrete compressive strength as a function of
the concrete mix composition calibrated by two observations. Observation encompasses
the complete concrete recipe, along with compressive strength for a given type of concrete.
The algorithm uses two recipes for a given concrete and tries to estimate the third recipe’s
concrete compressive strength.
In current considerations, the database from a previous study [16] was used. The
collected database serves as a basis for training the DNN with respect to the dependencies
between individual input variables and the output variable. The data collection is extensive
and includes various concrete mix recipes and laboratory tests results. The concrete mix
recipes included in the database have been designed to be built into concrete structures
of various dimensions, purposes, and functions. Some of them also contain admixtures
of various origins and purposes, such as binding retardants, plasticizers, and workability
boosters. Due to the abovementioned factors, some differences are challenging to predict
between recipes of concrete mixes. Tested samples were standardized concrete cylinders
with 15 cm diameter. Noncylindrical samples were converted into cylindrical following
valid norms [43]. The samples were made from ordinary Portland cement. The aggregate
size in the dataset did not exceed 20 mm. The parameters that were adopted are presented
in Table 1.

Table 1. The parameters adopted in the dataset.

Compressive Strength
Parameter Cement Water Sand 0–2 mm Aggregate above 2 mm Water–Cement Ratio
after 28 days
Codename cs_28 cement water fine_grained_aggregate coarse_aggregate water_cement_ratio
Type Target Input Input Input Input Input
Content of
The 28 day
Content of cement Content of water Content of coarse-grained
compressive strength of
added to the added to the fine-grained aggregate aggregate with a size Water-to-cement
Description concrete that is
mixture, expressed mixture, expressed added to the mixture, more than 2 mm, ratio.
considered to have
in kg. in L. expressed in kg. added to the mixture,
most of its strength.
expressed in kg.
Materials 2021, 14, 1661 8 of 34

The parameters presented in Table 1 are divided into two groups. The inputs refer
to the input variables, such as cement, water, fine-grained and coarse-grained aggregate,
and water–cement ratio. The target refers to the output variable, which is the concrete
compressive strength. In these considerations, there is a general assumption that concrete
achieves its designed compressive strength after 28 days. Such an assumption was made
because, after the concrete mix fabrication, the cement hydration process begins, which
progresses over time and increases the concrete’s strength until it reaches full strength.
According to general knowledge, this process takes around 28 days (for some types of
concrete, this time is longer). Before the indicated time, the concrete has partial strength.
The samples in which the compressive strength test was performed earlier than 28 days
were removed from the database. Table 2 presents the maximal, minimal, mean, median,
and dominant value for each variable.

Table 2. Value ranges of database input variables.

Input Variable Minimum Maximum Mean Median Dominant

Cement 86.00 kg/m3 540.00 kg/m3 278.00 kg/m3 275.00 kg/m3 425.00 kg/m3
Water 121.80 kg/m3 247.00 kg/m3 182.42 kg/m3 185.00 kg/m3 192.00 kg/m3
Fine-grained aggregate
372.00 kg/m3 1329.00 kg/m3 768.55 kg/m3 777.80 kg/m3 594.00 kg/m3
(sand 0–2 mm)
Coarse aggregate
597.00 kg/m3 1490.00 kg/m3 969.08 kg/m3 967.00 kg/m3 932.00 kg/m3
(aggregate above 2 mm)
Water–cement ratio 0.27 1.88 0.76 0.69 0.45

It is recommended to operate only within the limits set by range values for each input
variable. Since the DNN is trained on a specific dataset, going beyond the range values can
lead to the wrong results. The study did not directly analyze the effects of using con-crete
additives and admixtures. Their influence is accounted for indirectly by including two
observations if they affect the target variable. Many other processes influence hard-ened
concrete properties, such as the curing process, but their influence was not consid-ered
in this study. It is assumed that the quality control of the production of concrete mix and
concrete was sufficient. As a result of this research, a trained DNN was obtained, translated
into the source code, and interpreted as an equation, defining the 28 day com-pressive
strength of concrete as a function of the 17 parameters. Of these, 12 parameters relate to the
two recipes (cement, water, fine-grained and coarse-grained aggregate, water–cement ratio,
and concrete compressive strength). The remaining five parameters describe the recipe for
calculating the desired target value. A hyperbolic tangent was used for hid-den layers as
the activation function and a linear tangent was used for the output layer. The practical
application of the presented solution in the concrete mix design process is presented in
Figure 1.
Materials 2021, 14, 1661 9 of 34
Materials 2021, 14, x FOR PEER REVIEW 9 of 33

Figure1.1. Flowchart
Figure Flowchart presenting
presenting the
thepractical
practicalapplication
application of
ofthe
theMAFM21
MAFM21(Machine-learning
(Machine-learningAdaptive
AdaptiveForecasting
ForecastingModel
Model
2021)-based formula in the concrete mix design process.
2021)-based formula in the concrete mix design process.

Later in this
Later thispaper,
paper,the
thedesigned
designedalgorithm
algorithmis referred to astoMAFM21
is referred as MAFM21 (Machine-learn-
(Ma-chine-
ing Adaptive
learning Forecasting
Adaptive Model
Forecasting 2021).
Model In the
2021). In comparative
the comparative analysis presented
analysis presentedbelow, we
below,
used the machine learning algorithm developed in the previous research
we used the machine learning algorithm developed in the previous research [16]. This [16]. This algo-
rithm was was
algorithm built built
usingusing
a different modelmodel
a different but was trained
but on the same
was trained dataset.
on the same Itdataset.
estimatedIt
the compressive
estimated strength ofstrength
the compressive hardened concrete on
of hardened the basis
concrete on of
thethe amount
basis of theand composi-
amount and
tion of main concrete
composition of main mix components.
concrete Later, this algorithm
mix components. Later, thisisalgorithm
referred toisasreferred
MLM19to (Ma-
as
chine Learning
MLM19 (Machine Model 2019).Model 2019).
Learning

3.2.
3.2. Results
Results and
and Discussion
Discussion
The
The dataset
datasetwaswasdivided
dividedinto intothree
threesubsets
subsetstotoperform
performDNNDNN training, namely,
training, training,
namely, train-
selection, and testing datasets. These three datasets have distinct functions.
ing, selection, and testing datasets. These three datasets have distinct functions. The train- The training
dataset is used
ing dataset to feed
is used the DNN,
to feed the DNN,the selection dataset
the selection helpshelps
dataset to adjust the hyperparameters
to adjust the hyperparam-
of DNN, and the testing dataset serves as a tool to assess
eters of DNN, and the testing dataset serves as a tool to assess the DNN’sthe DNN’s effectiveness. The
effectiveness.
whole database had 741 records, whereby the training dataset had 440
The whole database had 741 records, whereby the training dataset had 440 records (59%), records (59%), the
selection dataset had 146 records (20%), and the testing dataset had 146
the selection dataset had 146 records (20%), and the testing dataset had 146 records (20%). records (20%).
There
There were
were nine
nine records
records (1%)(1%) excluded
excluded from
from thethe analysis
analysis as
as univariate
univariate outliers,
outliers, with
with aa
maximum
maximum distance from the center of the dataset defined as three times the median value
distance from the center of the dataset defined as three times the median value
for
for each
each variable.
variable. This
This is
is aa less
less conservative
conservative approach than n
approach than n used
used in in the
the previous
previous study,
study,
where about 11% of the records were excluded. In Figure 2, scatter plots illustrating the
where about 11% of the records were excluded. In Figure 2, scatter plots illustrating the
concrete’s compressive strength related to the individual input variables are presented.
concrete’s compressive strength related to the individual input variables are presented.
Materials 2021, 14, 1661 10 of 34
Materials 2021, 14, x FOR PEER REVIEW 10 of 33

(A) (B)

(C) (D)

(E)
Figure 2.
Figure 2. The
Thescatter
scatterplots
plotsofof
target variable
target vs. vs.
variable input variables.
input The full
variables. Thecompressive strength
full compressive of concrete
strength is on theisvertical
of concrete on the
axis, expressed
vertical in MPa. in
axis, expressed The horizontal
MPa. axis is theaxis
The horizontal material
is thecontent,
materialexpressed in kg for cement,
content, expressed in kg forfine-grained aggregate,
cement, fine-grained
and coarse-grained aggregate and L for water: (A) cement; (B) water; (C) fine-grained aggregate (sand 0–2 mm); (D) coarse-
aggregate, and coarse-grained aggregate and L for water: (A) cement; (B) water; (C) fine-grained aggregate (sand 0–2 mm);
grained aggregate (aggregate above 2 mm); (E) water–cement ratio.
(D) coarse-grained aggregate (aggregate above 2 mm); (E) water–cement ratio.

The initial DNN

The initial DNN architecture
architecturecomprised
comprised1717input
inputvariables
variablesthat
that refer
refer toto
thethe
1717 princi-
principal
pal components and introduce one target output. The initial architecture contained
components and introduce one target output. The initial architecture contained three three
hidden layers, representing the complexity of the model. A DNN includes
hidden layers, representing the complexity of the model. A DNN includes principal principal com-
ponents, perceptron neurons, scaling neurons, and unscaling neurons. In the analysis,
components, perceptron neurons, scaling neurons, and unscaling neurons. In the analysis, the
feature scaling was used, hence the scaling and unscaling neurons. Two
the feature scaling was used, hence the scaling and unscaling neurons. Two recipes recipes were
 Materials
Materials 2021, 14, x 2021, 14, 1661
FOR PEER REVIEW 11 of 33 11 of 34

treated as observations
were treated asbyobservations
introducing into the equationinto
by introducing a setthe
of input variables
equation a set represent-
of input variables
ing cement, water, fine-grained
representing cement, water, and coarse-grained
fine-grained and aggregate, water–cement
coarse-grained aggregate,ratio, and
water–cement
concrete ratio,
compressive strength.
and concrete These two
compressive recipesThese
strength. generated 12 input
two recipes variables.
generated 12 The
input re-variables.
maining The
five input variables
remaining five corresponded
input variables tocorresponded
the composition to of
thethe third recipeof
composition forthe
which
third recipe
it was tried to predict
for which it wasthetried
concrete compressive
to predict strength.
the concrete The input
compressive variables
strength. The(cement,
input variables
(cement, water, fine_grained_aggregate,
water, fine_grained_aggregate, coarse_aggregate, coarse_aggregate,
water_cement_ratio) water_cement_ratio)
presented in Ta- presented
in Table 1 corresponded
ble 1 corresponded with the inputwith the input
neurons. neurons.
The target The target
variable (cs_28)variable (cs_28) was
was associated withassociated
with
the output the output
neuron. To find neuron.
a proper Totraining
find a proper
rate, thetraining rate, the Broyden–Fletcher–Goldfarb–
Broyden–Fletcher–Goldfarb–Shanno
algorithmShanno
[44–49]algorithm
was used.[44–49]
Next, to waslookused.
for aNext, to look fortraining
quasi-Newton a quasi-Newton
direction,training
the Brentdirection,
the Brent
method [50–53] wasmethod
applied. [50–53] was applied.
For analytical For analytical
purposes, purposes,and
a linear correlation a linear correlation and
a correlation
matrix were assessed and calculated. Input contribution calculations were performed, were
a correlation matrix were assessed and calculated. Input contribution calculations
performed,
where training inputswhere
weretraining
selectivelyinputs were selectively
eliminated eliminated
and the output andwere
results the output results were
inspected,
wherebyinspected,
lower and whereby
higher input lower and highervalues
contribution input meancontribution
that the values
variablemean
gavethat the variable
a lower
or highergave a lower or
contribution to higher contribution
the results, to theThe
respectively. results, respectively.
presented analysisThe presented
indicated that analysis
indicated that
the water–cement ratiothe
hadwater–cement
an immenseratio had an immense
contribution contribution
to the results. Sincetotwo
the results.
observa-Since two
tions were used with one actual recipe for predicting concrete compressive strength, everystrength,
observations were used with one actual recipe for predicting concrete compressive
variable every
gave avariable gave a slightly
slightly different different
contribution, contribution,
which followed which followed
the overall patternthefor
overall
each pattern
for each concrete mix ingredient. Feature correlation analysis
concrete mix ingredient. Feature correlation analysis was performed to show the relation- was performed to show the
ship between the searched output variable of the compressive strength and the individual and the
relationship between the searched output variable of the compressive strength
individual
input variables, input variables,
as presented in Figureas presented
3. It can bein seen
Figure 3. It
that thecan be seen that the
water–cement water–cement
ratio and
ratio
the cement and the
amount cement amount
significantly impact significantly
the concrete impact the concrete
strength. strength. of
The significance The significance
the wa- of
the water–cement ratio and the amount of cement for
ter–cement ratio and the amount of cement for concrete compressive strength was also concrete compressive strength was
confirmedalsoin confirmed
literature by invarious
literature by various
researchers researchers
[54]. [54]. In these considerations,
In these considerations, the main focusthe main
focus is on the effect of concrete mix composition on concrete compressive strength.
is on the effect of concrete mix composition on concrete compressive strength.

Figure 3. Figure Feature correlation

Feature3.correlation heatmap. heatmap.

However, However,
it shoulditbe
should
notedbe noted
that the that the strength
strength of concrete
of concrete is influenced
is influenced by several
by several
other factors, primarily related to the environmental conditions and technological process process
other factors, primarily related to the environmental conditions and technological
of concrete
of concrete production.
production. The first The first important
important aspect isaspect
properisconcrete
proper concrete
curationcuration
after theafter the
built-in process. Improper concrete curation can significantly deteriorate
built-in process. Improper concrete curation can significantly deteriorate the properties the properties
of
of concrete, especially its durability. Another vital issue is concreting
concrete, especially its durability. Another vital issue is concreting during unfavorable during unfavorable
weather weather conditions,
conditions, mainly mainly
concreteconcrete
freezing freezing
in the in the stage
early early stage
of theofhydration
the hydration process and
process
too high shrinkage due to drying during high temperatures. Considering environmental
 Materials 2021, 14, x FOR PEER REVIEW 12 of 33

Materials 2021, 14, 1661 12 of 34

and too high shrinkage due to drying during high temperatures. Considering environ-
mental factors, one must refer to adverse environmental aggression, which harms the
quality one
factors, of concrete,
must refer for to
example,
adverseaenvironmental
high risk of carbonation
aggression, [55] and chloride
which harms the aggression
quality of in
coastal regions.
concrete, Concrete
for example, additives
a high risk ofand admixtures
carbonation [55]also
andplay an essential
chloride role in
aggression in facilitat-
coastal
ing the technological
regions. Concrete additives process andofadmixtures
concrete production
also play an and are widely
essential role inused. The origin,
facilitating the
shape, and texture
technological processofofaggregate impact the and
concrete production workability
are widely and durability
used. of concrete,
The origin, shape, and but
their influence
texture is more
of aggregate impactsignificant on fresh and
the workability concrete than hardened
durability of concrete, concrete
but their[56]. Moreo-
influence
isver,
more significant
a grading and onsize
fresh concrete than
distribution hardened concrete
of aggregate determines [56].theMoreover, a grading and
paste requirement for
size distribution
workable of aggregate
concrete [57]. In this determines
analysis, the pastefactors
several requirement
were not for considered,
workable concrete such as[57].
the
In this analysis,
technological severalenvironmental
process, factors were not considered,
factors, and rawsuch as the
material technological
properties, assuming process,
that
environmental factors, and
the quality of obtained raw material
concrete samplesproperties, assuming that
was at an appropriate the quality
level, while allofresults
obtained
de-
concrete samples wasfrom
viating significantly at antheappropriate
mean valueslevel,inwhile all results
the dataset were deviating
removed, significantly
treating them from as
the mean values
univariate outliers.in the dataset were removed, treating them as univariate outliers.
The
The dataset had many
dataset had manyinputinputvariables.
variables.Principal
Principal component
component analysis
analysis (PCA)(PCA)was was
used
used
to keepto keep critical
critical information
information whilewhile
reducingreducing
input input variables
variables in a smaller
in a smaller featurefeature
space.
space.
ThanksThanks to this method,
to this method, among otheramong otherit things,
things, it wastopossible
was possible reduce the to reduce the data
data dimension-
dimensionality
ality by feature by feature extraction.
extraction. The PCA uses Theauxiliary
PCA uses auxiliary
variables tovariables
explain mostto explain most
of the varia-
of the variability in the dataset [58–62]. One of the critical aspects
bility in the dataset [58–62]. One of the critical aspects when building an optimal DNN when building an
optimal DNN model is implementing the order selection algorithm.
model is implementing the order selection algorithm. These algorithms help to minimize These algorithms
help to minimize
the loss of acquired thedata
lossandof acquired data and
find the most findDNN
suitable the mostmodelsuitable
with an DNN model
optimal with
number
an optimal number of neurons which will match the data. In this
of neurons which will match the data. In this study, an incremental order algorithm [63– study, an incremental
order
68] wasalgorithm [63–68]
used. Figure was used.
4 shows the Figure
training4 shows the training
and selection and selection
loss history loss history
for various subsets
for various subsets from the incremental order algorithm
from the incremental order algorithm performance within 10 iterations. performance within 10 iterations.

Figure4.4.The
Figure Theloss
losshistory
history from
from incremental
incremental order
order algorithm
algorithm execution,
execution, within 10 iterations, where
the green line is the training loss and the purple line is the selection loss.Loss
the green line is the training loss and the purple line is the selection loss. Lossisison
onthe
thevertical
verticalaxis
axis and order is on the horizontal
and order is on the horizontal axis. axis.

Thefinal
The finalarchitecture
architectureof
ofthe
theadopted
adoptedDNN DNNisispresented
presentedininFigure
Figure5.5.The
Thefinal
finalDNN
DNN
model had seven hidden layers, 17 inputs, and one output. The model contained
model had seven hidden layers, 17 inputs, and one output. The model contained principal principal
components (blue), perceptron
components perceptronneurons
neurons(red),
(red),scaling neurons
scaling (green),
neurons andand
(green), unscaling neu-
unscaling
neurons (yellow).
rons (yellow). The The
finalfinal
DNN DNN architecture
architecture shouldshould bemost
be the the most optimal
optimal model model for a
for a given
given
task. task.
 Materials 2021, 14, 1661 13 of 34
Materials 2021, 14, x FOR PEER REVIEW 13 of 33

Figure 5. Final
Figure architecture
5. Final of deep
architecture neural
of deep network
neural (DNN)
network that was
(DNN) thatused. The figure
was used. showsshows
The figure the the DNN
DNNarchitecture,
architecture,which
whichincludes
includesprincipal
principalcomponents
components (blue),
(blue), perceptron
perceptron neurons (red), and
neurons (red), and scaling and
scaling and unscaling layers. The scaling neurons are green, and unscaling neurons are yellow.
unscaling layers. The scaling neurons are green, and unscaling neurons are yellow. Abbreviations: C—
Abbreviations: C—cement; W—water; FA—fine-grained aggregate; CA—coarse-grained aggre-
gate;cement; W—water; FA—fine-grained
WC—water–cement ratio; CS—the fullaggregate; CA—coarse-grained
concrete compressive strength;aggregate; WC—water–cement
FO—first observation;
ratio; CS—the
SO—second observation.full concrete compressive strength; FO—first observation; SO—second observation.

The The developed

developed DNNDNN had onehadtarget
one target variable,
variable, the concrete
the concrete compressive
compressive strength,
strength, with with
the 17 input variables representing two observation recipes and one recipe for the targetedtargeted
the 17 input variables representing two observation recipes and one recipe for the
value.
value. InputInput variables
variables expressed
expressed several several concrete
concrete mix characteristics,
mix characteristics, such such as cement,
as cement, wa- water,
fine-grained and coarse-grained aggregate content, and water–cement
ter, fine-grained and coarse-grained aggregate content, and water–cement ratio. The DNN ratio. The DNN was
translated into a mathematical formula, presented as Equation
was translated into a mathematical formula, presented as Equation (5), along with auxil- (5), along with auxiliary
Equations (A6)–(A49).
iary Equations (A6)–(A49). Equation (5) refers to the 28 day strength of concrete, which can be
which
can considered
be considered as as
thethefull
fullstrength
strengthofofconcrete.
concrete.TheThe equation
equation was optimized
optimizedand andsim-
simplified.
TheThe
plified. principal components
principal components were were drawn
drawn into thethe
into equation.
equation.
f ull cs
𝑓𝑢𝑙𝑙 f c𝑐𝑠 = 40.6152864 − 100.2590155·(tanh0.0630459 − 0.928083· ax − 2.03459· ax + 1.21399· ax53
𝑓𝑐 = 40.6152864 − 100.2590155 ∙ (tanh 0.0630459 − 0.928083 ∙ 𝑎𝑥5151− 2.03459 ∙ 𝑎𝑥52 52 + 1.21399 ∙ (5)
(5)
+1.17276
𝑎𝑥53 + 1.17276 · ax
∙ 𝑎𝑥 −−
54 54 0.56805
0.56805 · ax5555++0.191161
∙ 𝑎𝑥 0.191161∙· ax
𝑎𝑥5656)) [[MPa].
MPa].
The The
variables 𝐶𝐹𝑂 ,C𝑊
variables FO𝐹𝑂, ,W𝐹𝐴 , FA
FO𝐹𝑂 , 𝐶𝐴
FO𝐹𝑂, CA
, 𝐶𝑊
FO𝐹𝑂, CW
, 𝐶𝑆FO𝐹𝑂, ,CS
𝐶𝑆𝑂FO, , 𝑊C𝑆𝑂 WSO
SO,, 𝐹𝐴 𝑆𝑂,, FA
𝐶𝐴SO𝑆𝑂 , CA
𝐶𝑊SO𝑆𝑂 , CWSO ,
𝐶𝑆𝑆𝑂CS, 𝐶,
SO𝑊,
, C,𝐹𝐴,W, 𝐶𝐴, CA,𝑊𝐶
FA,and and WCinused
used in Equations
Equations (A40)–(A49) (A40)–(A49)
denote thedenotecementthe (𝐶),cement
wa- (C),
water
ter (𝑊), (W), fine-grained
fine-grained aggregate aggregate
(𝐹𝐴), coarse(FA), coarse aggregate
aggregate (CA), water–cement
(𝐶𝐴), water–cement ratio (𝑊𝐶), ratioand (WC),
concrete strength (𝐶𝑆).
and concrete The first
strength (CS).sixThevariables
first sixrepresent
variables therepresent
composition of the first obser-
the composition of the first
vation recipe (𝐹𝑂).
observation The subsequent
recipe six variablessix
(FO). The subsequent represent
variables therepresent
composition theofcomposition
the second of the
observation recipe (𝑆𝑂).recipe
second observation The last fiveThe
(SO). variables
last fivedescribe
variables thedescribe
actual concrete
the actual recipe
concretefor recipe
whichforthe compressive
which strengthstrength
the compressive was calculated. Units areUnits
was calculated. expressed in kg/m3 in
are expressed or kg/m
l/m3 for3 or l/m3
variables C, W, FA,
for variables andFA,
C, W, CA. and Variable WC is WC
CA. Variable a numerical ratio that
is a numerical ratiois that
a dimensionless
is a dimensionless
value.
value. There were 36 auxiliary formulas, with Equations (A6) through containing
There were 36 auxiliary formulas, with Equations (A6) through (A39) (A39) containing
variables withwith
variables the the
index 𝑎𝑥, ax,
index which
which areare
thetheauxiliary
auxiliary variables
variables presented
presented in in
five
fivelayers,
layers, each
eachwith
with6–10
6–10variables.
variables. Equations
Equations (A6)–(A49) are presented in
(A6)–(A49) are presented in Appendix A. Appendix A.
The The
comparative
comparative analysis
analysis waswasprepared,
prepared,which which included
included aacomparison
comparisonofof thethe
MAFM21
MAFM21
algorithm with the MLM19 algorithm from 2019 presented in [16], which waswas
algorithm with the MLM19 algorithm from 2019 presented in [16], which trained on
trained
the on
same thedataset
same dataset
but was butbuilt
was built
usingusing a different
a different model.
model. FourFour concrete
concrete mixmix rec- were
recipes
ipestested,
were tested,
designeddesigned
accordingaccording
to the to the Bolomey
Bolomey and Fullerand Fuller
designdesign
methods, methods,
whichwhich are standard
are standard
concrete concrete
mix designmixapproaches.
design approaches.
The firstThe firstwas
recipe recipe was a standard
a standard mix consist-
mix consisting of cement,
ing water,
of cement,and fine-grained and coarse-grained aggregate with a plasticizerplasticizer
water, and fine-grained and coarse-grained aggregate with a (lignosulfonate)
(lignosulfonate) and superplasticizer
and superplasticizer (polycarboxylate (polycarboxylate ether). The
ether). The second recipesecond
was arecipe
standard was mix
a with
plasticizer (lignosulfonate), superplasticizer (polycarboxylate ether), and air entrainer
Materials 2021, 14, 1661 14 of 34

(tensides). The third was a standard mix with superplasticizer (polycarboxylate ether),
retarder (phosphate), and air entrainer (tensides). The fourth was a standard mix with
plasticizer (lignosulfonate) and superplasticizer (naphthalene). The recipes were prepared
for 1 m3 of concrete designed as a concrete slab, with a plastic slump (which gives a
lower value than the usual water–cement ratio), direct pouring with little environmental
aggression, no special desired finishing, no special ambient conditions when casting, air
content at 2.5%, and 20 mm maximum grain diameter. The following materials were
adopted for the design: network water, ordinary Portland cement, natural sand, limestone
gravel 4/10 mm, and limestone gravel 10/20 mm. Combinations of admixtures commonly
used in engineering practice were selected. The exact composition of the designed mixtures
is presented in Table 3 (Bolomey method) and Table 4 (Fuller method). Two similar recipes
designed for a given compressive strength were used as observations to calculate the
concrete compressive strength using the MAFM21 algorithm.

Table 3. Tested concrete mix recipes designed according to Bolomey formula.

Designed
Limestone Limestone
Concrete Cement Natural Sand Water–Cement
Water (L/m3 ) Gravel 4/10 Gravel 10/20
Compressive (kg/m3 ) (kg/m3 ) Ratio (-)
(kg/m3 ) (kg/m3 )
Strength
Recipe #1—standard mix + plasticizer (lignosulfonate) + superplasticizer (polycarboxylate ether)
10 177.09 57.56 1106.54 525.32 681.00 0.325
15 204.57 66.49 1059.83 531.08 674.29 0.325
20 232.05 75.42 1014.45 535.33 667.74 0.325
25 259.53 84.35 968.28 540.54 661.04 0.325
30 287.01 88.17 930.22 547.64 658.12 0.307
35 314.49 91.12 892.82 555.61 656.02 0.290
40 341.97 94.16 855.60 563.45 653.64 0.275
45 369.45 97.26 818.51 570.86 651.40 0.263
50 396.93 100.41 781.08 578.63 649.00 0.253
Recipe #2—standard mix + plasticizer (lignosulfonate) + superplasticizer (polycarboxylate ether) + air entrainer (tensides)
10 178.26 57.93 1105.90 527.60 677.03 0.325
15 205.58 66.81 1060.69 532.11 670.39 0.325
20 233.33 75.83 1015.00 536.35 663.75 0.325
25 261.47 84.98 968.79 540.62 656.93 0.325
30 288.63 88.67 929.92 549.20 654.18 0.307
35 315.36 91.38 894.37 556.25 652.29 0.290
40 343.21 94.50 856.65 564.08 649.89 0.275
45 371.42 97.78 817.79 572.16 647.64 0.263
50 399.12 100.97 780.12 580.04 645.11 0.253
Recipe #3—standard mix + superplasticizer (polycarboxylate ether) + retarder (phosphate) + air entrainer (tensides)
10 176.46 57.35 1107.69 526.38 677.27 0.325
15 203.84 66.25 1062.94 530.80 670.17 0.325
20 231.22 75.15 1016.61 536.10 663.75 0.325
25 258.60 84.05 971.48 540.13 657.40 0.325
30 285.99 87.86 932.98 548.07 654.22 0.307
35 313.37 90.80 896.19 555.60 652.09 0.290
40 340.75 93.82 859.54 562.61 650.09 0.275
45 368.13 96.92 821.81 571.15 647.48 0.263
50 395.51 100.05 784.51 578.91 645.11 0.253
Recipe #4—standard mix + plasticizer (lignosulfonate) + superplasticizer (naphthalene)
10 179.00 58.17 998.39 369.05 947.31 0.325
15 206.77 67.20 1058.77 532.40 675.41 0.325
20 234.55 76.23 1012.64 537.16 668.59 0.325
25 262.32 85.26 966.36 542.02 661.83 0.325
Materials 2021, 14, 1661 15 of 34

Table 3. Cont.

Designed
Limestone Limestone
Concrete Cement Natural Sand Water–Cement
Water (L/m3 ) Gravel 4/10 Gravel 10/20
Compressive (kg/m3 ) (kg/m3 ) Ratio (-)
(kg/m3 ) (kg/m3 )
Strength
30 290.10 89.12 925.90 547.94 652.76 0.307
35 317.88 92.10 890.13 553.89 650.72 0.290
40 345.65 95.17 850.22 564.03 648.42 0.275
45 373.43 98.31 812.49 571.92 646.00 0.263
50 401.20 101.49 774.79 579.60 643.63 0.253

Table 4. Tested concrete mix recipes designed according to Fuller formula.

Designed CC Limestone Limestone

Cement Natural Sand Water–Cement
of Concrete Water (L/m3 ) Gravel 4/10 Gravel 10/20
(kg/m3 ) (kg/m3 ) Ratio (-)
Mix (kg/m3 ) (kg/m3 )
Recipe #1—standard mix + plasticizer (lignosulfonate) + superplasticizer (polycarboxylate ether)
10 177.09 57.56 1056.81 566.37 702.13 0.325
15 204.57 66.49 1035.14 554.76 687.74 0.325
20 232.05 75.42 1013.48 543.14 673.34 0.325
25 259.53 84.35 991.81 531.54 658.95 0.325
30 287.01 88.17 976.42 523.29 648.72 0.307
35 314.49 91.12 962.09 515.61 639.20 0.290
40 341.97 94.16 947.66 507.87 629.61 0.275
45 369.45 97.26 933.15 500.10 619.97 0.263
50 396.93 100.41 918.57 492.29 610.29 0.253
Recipe #2—standard mix + plasticizer (lignosulfonate) + superplasticizer (polycarboxylate ether) + air entrainer (tensides)
10 178.26 57.93 1055.89 565.88 696.32 0.325
15 205.58 66.81 1034.35 554.33 682.12 0.325
20 233.33 75.83 1012.47 542.61 667.69 0.325
25 261.47 84.98 990.29 530.72 653.06 0.325
30 288.63 88.67 975.18 522.62 643.10 0.307
35 315.36 91.38 961.44 515.26 634.04 0.290
40 343.21 94.50 946.76 507.39 624.36 0.275
45 371.42 97.78 931.75 499.34 614.45 0.263
50 399.12 100.97 917.04 491.47 604.76 0.253
Recipe #3—standard mix + superplasticizer (polycarboxylate ether) + retarder (phosphate) + air entrainer (tensides)
10 176.46 57.35 1057.31 566.64 697.26 0.325
15 203.84 66.25 1035.72 555.07 683.02 0.325
20 231.22 75.15 1014.13 543.50 668.79 0.325
25 258.60 84.05 992.55 531.93 654.55 0.325
30 285.99 87.86 977.20 523.71 644.44 0.307
35 313.37 90.80 962.93 516.06 635.02 0.290
40 340.75 93.82 948.55 508.35 625.54 0.275
45 368.13 96.92 934.09 500.60 616.00 0.263
50 395.51 100.05 919.57 492.82 606.43 0.253
10 179.00 58.17 1055.30 565.56 701.13 0.325
15 206.77 67.20 1033.41 553.83 686.59 0.325
20 234.55 76.23 1011.51 542.09 672.04 0.325
25 262.32 85.26 989.61 530.36 657.49 0.325
30 290.10 89.12 974.05 522.20 642.36 0.307
35 317.88 92.10 959.57 514.25 632.80 0.290
40 345.65 95.17 944.98 506.44 623.18 0.275
45 373.43 98.31 930.32 498.58 613.51 0.263
50 401.20 101.49 915.59 490.68 603.80 0.253
Materials 2021, 14, 1661 16 of 34

Gradings and fitting curves for each designed concrete recipe are presented in Ap-
Materials 2021, 14, x FOR PEER REVIEW 16 of 33
pendix B. The comparative analysis of errors is shown in Figure 6.
The comparative analysis of errors is presented in Figure 7.

(A) (B)

(C) (D)

Figure 6. Comparison
Comparisonbetween
betweendesigned
designedconcrete
concretecompressive
compressivestrength
strengthandandcalculated
calculatedconcrete compressive
concrete compressivestrength ac-
strength
cording toto
according MLM19
MLM19(Machine
(MachineLearning
LearningModel
Model2019)
2019)and
andMAFM21
MAFM21formulas
formulasfor forfour
four recipes
recipes calculated
calculated according
according toto the
the
Bolomey and
Bolomey andFuller
Fullermethods.
methods.(A)
(A)Recipe
Recipe#1—standard
#1—standard mix
mix + plasticizer
+ plasticizer (lignosulfonate)
(lignosulfonate) + superplasticizer
+ superplasticizer (polycarbox-
(polycarboxylate
ylate ether); (B) Recipe #2—standard mix + plasticizer (lignosulfonate) + superplasticizer (polycarboxylate ether) + air
ether); (B) Recipe #2—standard mix + plasticizer (lignosulfonate) + superplasticizer (polycarboxylate ether) + air entrainer
entrainer (tensides); (C) Recipe #3—standard mix + superplasticizer (polycarboxylate ether) + retarder (phosphate) + air
(tensides); (C) Recipe #3—standard mix + superplasticizer (polycarboxylate ether) + retarder (phosphate) + air entrainer
entrainer (tensides); (D) Recipe #4—standard mix + plasticizer (lignosulfonate) + superplasticizer (naphthalene).
(tensides); (D) Recipe #4—standard mix + plasticizer (lignosulfonate) + superplasticizer (naphthalene).
The comparative analysis of errors is presented in Figure 7.
 Materials
Materials 14, 14,
2021,
2021, 1661PEER REVIEW
x FOR 1733
17 of of 34

Figure 7. Cont.
Materials2021,
Materials 14,x1661
2021,14, FOR PEER REVIEW 18of
18 of33
34

Figure
Figure7.
7.Errors
Errorsin
inthe
thecomparison
comparisonbetween
betweendesigned
designedconcrete
concretecompressive
compressivestrength
strengthand
andcalculated
calculatedconcrete
concretecompressive
compressive
strength
strength according
according toto MLM19
MLM19 and
and MAFM21
MAFM21 formulas
formulas for
for four
four recipes
recipes designed
designed according
according to
to the
the Bolomey
Bolomey and
and Fuller
Fuller
methods. MAE—mean absolute error; MSE—mean squared error; RMSE—root-mean-squared error.
methods. MAE—mean absolute error; MSE—mean squared error; RMSE—root-mean-squared error.

ItItcan
canbebeseen seeninin Figure
Figure 6 that thethe
6 that MAFM21
MAFM21 algorithm
algorithmpresented
presentedthe best
the fit forfit
best mix-
for
tures designed
mixtures designedaccording
accordingto Bolomey. Good Good
to Bolomey. convergence was especially
convergence visible for
was especially higher
visible for
strength levels above
higher strength levels30aboveMPa. 30 MAFM21 was also was
MPa. MAFM21 characterized as a good as
also characterized fit afor mixtures
good fit for
designed according to Fuller in the lower ranges, below 20 MPa.
mixtures designed according to Fuller in the lower ranges, below 20 MPa. However, there However, there was some
underestimation
was some underestimationin the range in from
theabout
range25fromMPaabout
to 40 MPa,
25 MPa depending
to 40 MPa, on the mixture.on
depending It
should also beItnoted
the mixture. should thatalsoMAFM21
be notedwasthat more
MAFM21volatilewas than MLM19.
more volatileForthan
mixtureMLM19.number For
four,
mixturethere was even
number four,considerable
there was even mismatch for 10mismatch
considerable MPa. MLM19 for 10 hadMPa.low MLM19resilience
had lowfor
high-strength recipes, whereas
resilience for high-strength MAFM21
recipes, whereasperformed
MAFM21 better in that better
performed spectrum.in that Thespectrum.
statisti-
cal
Theanalysis
statistical of analysis
the errors of presented in Figure 7inshowed
the errors presented Figure 7some showed interesting findings.findings.
some interesting Figure
7Figure
shows7that showsthe MAFM21
that the MAFM21 algorithm was characterized
algorithm by a similar
was characterized by alevel of mean
similar levelabsolute
of mean
absolute
error (MAE),error (MAE),
mean squared meanerrorsquared
(MSE), error
and(MSE), and root-mean-squared
root-mean-squared error (RMSE) errorfor(RMSE)
Fuller
for Fullercompared
mixtures mixtures to compared
MLM19 to MLM19
and and significantly
significantly lower MAE, lower
MSE,MAE, and RMSEMSE, and valuesRMSEfor
values for
Bolomey Bolomey
mixtures. mixtures.
Figure Figure
7 presents 7 presents
almost almost
a twofold a twofold
decrease in MAEdecrease
and RMSE in MAE for and
the
RMSE formix
Bolomey theandBolomey
a slightmix and a in
decrease slight
these decrease
errors for in the
these errors
Fuller mix forforthe
theFuller mixsecond
first and for the
first andAsecond
recipes. recipes.
significant A significant
difference can bedifference
noted lookingcan be at noted lookingfor
MSE, where, at MSE, where,ac-
the mixture for
the mixture according to the Fuller method, the error was on
cording to the Fuller method, the error was on a similar level, whereas, for the mixture a similar level, whereas, for
the mixture
according toaccording
the Bolomey to themethod,
Bolomeythe method,
error the
waserrorfivewastimesfivesmaller
times smaller
in favor in favor
of the of
the MAFM21
MAFM21 algorithm.
algorithm. MAFM21 MAFM21 in thein the third
third reciperecipe
gave an gave
evenanlower
even lower
level oflevelMAE, of MSE,
MAE,
MSE,
and and RMSE
RMSE for thefor the Bolomey
Bolomey mixture;
mixture; however,however,
in this incase,
this case,
MLM19 MLM19had hada lowera lower
MSEMSE for
forFuller
the the Fullermix.mix.
Recipe Recipe
four four
had ahad a higher
higher MSE MSEfor thefor the MAFM21
MAFM21 Bolomey Bolomey
mix than mixthethan the
third
third recipe, but it was still significantly lower than the Bolomey
recipe, but it was still significantly lower than the Bolomey mixture for MLM19. The mixture for MLM19. The
MAFM21 algorithm
MAFM21 algorithm for for the
the fourth
fourthrecipe
recipealso
alsoperformed
performed better
better in terms
in terms of MAE,
of MAE, MSE, and
MSE,
RMSE for Fuller mixes. It should be noted that the presented
and RMSE for Fuller mixes. It should be noted that the presented solution is still experi- solution is still experimental,
and it does
mental, and not consider
it does some keysome
not consider issues,
keysuch as the
issues, suchtechnological process and
as the technological durability.
process and
durability.
4. Summary and Conclusions
One of the
4. Summary andmost critical issues in the concrete structure manufacturing technological
Conclusions
process is to obtain predictable properties of both fresh concrete mix and hardened concrete.
One of the most critical issues in the concrete structure manufacturing technological
Manufacturers of concrete mix are obliged to guarantee the appropriate properties of
process is to obtain predictable properties of both fresh concrete mix and hardened con-
concrete delivered to the construction site. However, obtaining the appropriate concrete
crete. Manufacturers of concrete mix are obliged to guarantee the appropriate properties
properties at all stages of the production process is difficult, especially when chemically
of concrete delivered to the construction site. However, obtaining the appropriate con-
complex additives and admixtures for concrete are present. Therefore, the properties of
crete properties
concrete at all
mixes are stages
often of the production
oversized, which canprocess
lead toisvery
difficult, especially
unfavorable when chemi-
phenomena. For
cally complex
example, additives
increasing and admixtures
concrete strength morefor concrete
than the are present.
designed Therefore,
value the properties
may cause changes in
oftheconcrete mixes
structure’s are often
stiffness andoversized, which concrete
lead to so-called can lead superstrength
to very unfavorable phenomena.
[23]. Predicting the
For example, increasing concrete strength more than the
properties of concrete is a complex issue, while most of the solutions used in designed value may cause
engineer-
changes in theare
ing practice structure’s
approximate stiffness and lead
methods that,toespecially
so-called concrete
in recentsuperstrength
years, due to [23]. Pre-
the rapid
dicting
development of materials engineering, have become outdated. Predictive analytics is in
the properties of concrete is a complex issue, while most of the solutions used an
engineering practice are approximate methods that, especially in recent
area of knowledge dealing with predicting all kinds of phenomena, properties, and trends. years, due to the
rapid
The mostdevelopment of materials engineering,
popular applications are economics, have become outdated.
mathematics, PredictiveOne
and medicine. analytics
of the
ismost
an area of knowledge dealing with predicting all kinds of phenomena,
promising branches in predictive analytics is machine learning, which has acquired properties, and
trends.
particular attention in recent years due to the significant development of technology.ofIt
The most popular applications are economics, mathematics, and medicine. One
the most
could be promising branches in
used to significantly predictive
improve analytics
the process of is machine
concrete mixlearning,
design. which has ac-
quired particular
The primaryattention in recent
goal of this years
research wasdue to the significant
to introduce machinedevelopment
learning to theof technol-
concrete
ogy.
mix It couldprocess
design be usedandto significantly
pave the wayimprove
for new the process
quality of concrete
in this mix design.
area of science. The MAFM21
Materials 2021, 14, 1661 19 of 34

algorithm can improve the ability to predict concrete technical parameters by including
two observations. This adaptive approach allows us to predict the concrete mix behavior
depending on its unique characteristics. Concrete mixes used in engineering practice have
a very diverse composition. Designing a separate model for each specific composition of
the concrete mix may turn out to be impractical and, in the long run, could be an obstacle to
its widespread implementation. The core of the presented study was to develop an optimal
deep neural network architecture and train it using an extensive database of concrete
mix recipes with corresponding laboratory destructive tests. The developed algorithm
estimates the concrete mix compressive strength according to its composition. The database
used for training had 741 records, whereby the training dataset had 440 records (59%), the
selection dataset had 146 records (20%), and the testing dataset had 146 records (20%). Nine
records (1%) were excluded from the analysis as univariate outliners, with a maximum
distance from the center of the dataset defined as three times the median value for each
variable. The initially adopted DNN model had 17 input variables, 17 principal components,
four hidden neurons, and one target output. The final DNN architecture had 17 input
variables, 17 principal components, 46 hidden neurons, clustered into seven layers, and
one target output. The applicable training rate and the step for the quasi-Newton training
direction were calculated using the Broyden–Fletcher–Goldfarb–Shanno algorithm and
the Brent method, respectively. The feature correlation analysis revealed that the most
significant impact on concrete strength was the water–cement ratio and cement content.
The dimensionality reduction was performed using principal component analysis. An
incremental order algorithm was used to find the optimal number of neurons in DNN and
minimize loss. The DNN source code was translated into a mathematical equation, which
was then optimized and simplified. The final equation had 44 auxiliary equations. This
method was tested on four concrete mix recipes, calculated according to contemporary
design methods (Bolomey and Fuller method) and a comparative analysis of the new
MAFM21 algorithm with the algorithm developed in 2019 MLM19 was conducted by
analyzing the root-mean-squared error (RMSE), mean absolute error (MAE), and mean
squared error (MSE). The conducted tests allowed us to see how MAFA21 complied with
the methods used in engineering practice, and whether the adaptive approach performed
better than the previous MLM19 algorithm trained on the same dataset.
The comparative analysis brought us a number of findings. The MAFM21 algorithm
was more volatile, but had significantly lower RMSE, MAE, and MSE than MLM19 for
mixtures designed according to the Bolomey method. MAFM21 gave slightly lower RMSE,
MAE, and MSE for mixtures designed according to the Fuller method depending on the
tested mixture. MAFM21 had the best fit for Bolomey mixes, characterized by particu-
larly good convergence for higher strength levels above 30 MPa. MAFM21 performed
well in lower ranges (below 20 MPa) for mixtures designed according to Fuller. MLM19
had low resilience for high-strength recipes, whereas MAFM21 performed better in that
spectrum. The presented approach has boundary conditions and does not fully reflect all
the relationships between the components of concrete mix and concrete properties. This
issue requires further research. However, the results presented in this paper give hope
for broader use of this method in engineering practice. Future research should extend the
method’s use in the concrete mix design process by predicting other properties of fresh
concrete mixtures and hardened concrete, such as durability, consistency, air content, and
service life estimation. It would also be vital to create a more comprehensive approach for
concrete mix optimization.

Author Contributions: Conceptualization, P.Z.; methodology, P.Z.; software, P.Z.; validation, P.Z.;
formal analysis, P.Z.; investigation, P.Z.; resources, P.Z.; data curation, P.Z.; writing—original draft
preparation, P.Z.; writing—review and editing, P.Z. and S.-B.K.; visualization, P.Z.; supervision, P.Z.
and S.-B.K.; project administration, P.Z. and M.N.; funding acquisition, P.Z. All authors read and
agreed to the published version of the manuscript.
Materials 2021, 14, 1661 20 of 34

Funding: This study was financed by grant no. 037628 supporting young researchers at the Faculty
of Civil and Environmental Engineering, Gdansk University of Technology.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: No new data were created or analyzed in this study. Data sharing is
not applicable to this article.
Acknowledgments: The author wishes to acknowledge every institution which provided data and
tools to conduct this study.
Conflicts of Interest: The authors declare no conflict of interest.

Appendix A
Appendix A contains the auxiliary Equations (A6)–(A49) that are part of the main
Equation (5).
ax51 = tanh(−0.112106 + 0.345284· ax41 + 0.312406· ax42 + 0.357058· ax43 − 0.149771· ax44 +
(A6)
0.405909· ax45 − 0.246462· ax46 − 0.307179· ax47 + 0.329967· ax48 ).
ax52 = tanh(−0.10355 + 0.836911· ax41 + 0.902282· ax42 + 0.699127· ax43 + 0.00914586· ax44 +
(A7)
0.853297· ax45 − 0.594561· ax46 − 0.475542· ax47 + 0.687024· ax48 ).
ax53 = tanh(−0.00550484 + 0.0706381· ax41 + 1.37087· ax42 − 0.489375· ax43 + 0.766119· ax44 −
(A8)
0.76727· ax45 + 0.0118196· ax46 + 0.34148· ax47 + 0.0277991· ax48 ).
ax54 = tanh(0.0484762 + 0.207462· ax41 − 1.17439· ax42 + 0.599398· ax43 + 0.884247· ax44 +
(A9)
1.09625· ax45 − 0.181395· ax46 + 0.544369· ax47 − 0.0457279· ax48 ).
ax55 = tanh(0.0185739 + 0.114457· ax41 + 0.193903· ax42 + 0.266349· ax43 + 0.0349446· ax44 +
(A10)
0.360525· ax45 − 0.0275735· ax46 − 0.114269· ax47 − 0.0457279· ax48 ).
ax56 = tanh(0.162456 − 0.200818· ax41 − 0.111272· ax42 − 0.0271401· ax43 + 0.0839329· ax44 +
(A11)
0.0074793· ax45 + 0.0810916· ax46 + 0.0938168· ax47 − 0.0128759· ax48 ).
ax41 = tanh(0.304087 − 0.93554· ax31 − 0.332438· ax32 − 0.735927· ax33 + 0.114776· ax34 −
(A12)
0.210384· ax35 + 0.156608· ax36 − 0.378972· ax37 + 0.203964· ax38 − 0.149654· ax39 + 0.431182· ax310 ).
ax42 = tanh(0.237625 − 0.062685· ax31 + 0.492318· ax32 + 0.0000816928· ax33 + 0.393211· ax34 +
(A13)
0.767934· ax35 + 0.246462· ax36 − 0.421343· ax37 + 0.223478· ax38 − 0.669725· ax39 − 1.02597· ax310 ).
ax43 = tanh(−0.329438 + 0.0174092· ax31 + 0.65158· ax32 + 0.301201· ax33 + 0.675368· ax34 +
(A14)
0.00703462· ax35 + 0.280539· ax36 + 0.502622· ax37 − 0.326401· ax38 + 0.570339· ax39 + 0.682099· ax310 ).
ax44 = tanh(−0.133381 − 0.108074· ax31 − 0.411905· ax32 + 0.0102024· ax33 − 0.0754423· ax34 +
(A15)
0.564025· ax35 − 0.0710046· ax36 − 0.121997· ax37 − 0.674816· ax38 − 0.363496· ax39 + 0.490389· ax310 ).
ax45 = tanh(0.0712822 + 0.369538· ax31 − 0.982137· ax32 + 0.0487879· ax33 − 0.00170483· ax34 −
(A16)
0.389693· ax35 − 0.96266· ax36 + 0.704598· ax37 − 0.555597· ax38 + 0.252501· ax39 + 0.713304· ax310 ).
ax46 = tanh(0.165274 + 0.530625· ax31 + 0.000449785· ax32 − 0.268531· ax33 + 0.365822· ax34 −
(A17)
0.239763· ax35 + 0.220901· ax36 − 0.233759· ax37 − 0.651051· ax38 − 0.329246· ax39 − 0.258471· ax310 ).
ax47 = tanh(0.141672 − 0.0391891· ax31 − 0.0704693· ax32 − 0.175119· ax33 + 0.716787· ax34 −
(A18)
0.146807· ax35 − 0.389877· ax36 + 0.246424· ax37 + 0.377645· ax38 − 0.0800747· ax39 + 0.682193· ax310 ).
ax48 = tanh(0.167999 + 0.880357· ax31 + 0.0174599· ax32 + 0.203427· ax33 − 0.629281· ax34 −
(A19)
0.484972· ax35 − 0.166586· ax36 − 0.303889· ax37 − 0.150484· ax38 − 0.38319· ax39 + 0.18877· ax310 ).
Materials 2021, 14, 1661 21 of 34

ax31 = tanh(0.583476 + 0.257305· ax21 + 0.39606· ax22 + 0.625066· ax23 − 0.265241· ax24 −
(A20)
0.59034· ax25 − 0.521324· ax26 − 0.278803· ax27 + 0.393518· ax28 − 0.715381· ax29 − 0.134194· ax210 ).
ax32 = tanh(−0.4261 + 0.634804· ax21 − 0.121212· ax22 − 0.289991· ax23 + 0.609914· ax24 +
(A21)
0.0439993· ax25 − 0.451075· ax26 + 0.334219· ax27 − 0.0891551· ax28 − 0.584922· ax29 + 0.306643· ax210 ).
ax33 = tanh(0.10659 − 0.615507· ax21 − 0.0443891· ax22 − 0.0848592· ax23 + 0.456915· ax24 +
(A22)
0.142345· ax25 − 0.439859· ax26 + 0.12741· ax27 + 0.202291· ax28 − 0.184254· ax29 − 0.466343· ax210 ).
ax34 = tanh(−0.150688 + 0.0475739· ax21 + 0.411486· ax22 − 0.748264· ax23 + 0.0760448· ax24 −
(A23)
0.00419205· ax25 − 0.269192· ax26 − 0.139757· ax27 − 0.481452· ax28 − 0.61357· ax29 − 0.51427· ax210 ).
ax35 = tanh(0.257212 + 0.25639· ax21 + 0.277878· ax22 + 0.102443· ax23 + 0.666715· ax24 −
(A24)
0.401726· ax25 + 0.281977· ax26 − 0.215321· ax27 + 0.303246· ax28 − 0.405421· ax29 + 0.000360402· ax210 ).
ax36 = tanh(0.212833 + 0.403199· ax21 + 0.341386· ax22 − 0.293033· ax23 + 0.149875· ax24 −
(A25)
0.0625897· ax25 − 0.104747· ax26 + 0.747322· ax27 − 0.0410873· ax28 − 0.381452· ax29 + 0.557865· ax210 ).
ax37 = tanh(0.13612 + 0.070754· ax21 − 0.141931· ax22 + 0.317007· ax23 + 0.473061· ax24 +
(A26)
0.971685· ax25 + 0.674186· ax26 − 0.117296· ax27 + 0.165819· ax28 + 0.0580616· ax29 − 0.549804· ax210 ).
ax38 = tanh(−0.241159 − 0.150599· ax21 − 0.31649· ax22 − 1.04636· ax23 + 0.242558· ax24 +
(A27)
0.241968· ax25 − 0.110343· ax26 + 0.158529· ax27 − 0.0196748· ax28 + 0.296961· ax29 + 0.184· ax210 ).
ax39 = tanh(0.110104 − 0.820008· ax21 − 0.540616· ax22 − 0.263387· ax23 + 0.00700437· ax24 −
(A28)
0.592159· ax25 + 0.27605· ax26 − 0.0406511· ax27 − 0.484756· ax28 − 0.426393· ax29 − 0.31372· ax210 ).
ax310 = tanh(−0.127382 − 0.615599· ax21 − 0.281532· ax22 + 0.488731· ax23 − 0.730324· ax24 +
(A29)
0.859829· ax25 + 0.45729· ax26 − 0.441791· ax27 − 0.616392· ax28 − 0.371053· ax29 − 0.603444· ax210 ).
ax21 = tanh(−0.03503 + 1.17792· ax11 + 0.536391· ax12 − 0.694888· ax13 + 0.488167· ax14 +
(A30)
0.724072· ax15 − 0.0434757· ax16 + 0.190797· ax17 + 0.316145· ax18 + 0.33833· ax19 + 0.224688· ax110 ).
ax22 = tanh(0.219872 − 0.139292· ax11 + 0.498359· ax12 + 0.184706· ax13 + 0.506286· ax14 −
(A31)
0.179573· ax15 + 0.177916· ax16 − 0.105241· ax17 − 0.461557· ax18 + 0.365369· ax19 + 0.341493· ax110 ).
ax23 = tanh(0.441113 + 0.585713· ax11 + 0.710594· ax12 + 0.000187878· ax13 + 0.694946· ax14 −
(A32)
0.0253969· ax15 + 0.0666956· ax16 + 0.78373· ax17 − 0.150309· ax18 + 0.0361135· ax19 + 0.529901· ax110 ).
ax24 = tanh(0.277789 + 0.00996534· ax11 + 0.311853· ax12 − 0.515361· ax13 − 1.21524· ax14 −
(A33)
0.483273· ax15 + 0.0458816· ax16 + 0.680251· ax17 − 0.47257· ax18 − 0.377953· ax19 − 0.0153472· ax110 ).
ax25 = tanh(−0.0365425 + 0.000522626· ax11 + 0.186409· ax12 + 0.625137· ax13 − 1.29349· ax14 +
(A34)
0.441413· ax15 − 0.654638· ax16 + 0.203844· ax17 + 0.186017· ax18 + 0.162006· ax19 + 0.335084· ax110 ).
ax26 = tanh(−0.117201 + 0.131572· ax11 − 0.359282· ax12 + 0.389462· ax13 + 0.0835228· ax14 +
(A35)
0.513271· ax15 + 0.0913045· ax16 + 1.07162· ax17 − 0.0790713· ax18 − 0.423777· ax19 − 0.258301· ax110 ).
ax27 = tanh(−0.187318 − 0.593468· ax11 − 0.457257· ax12 − 0.218044· ax13 + 0.170376· ax14 +
(A36)
0.230136· ax15 − 0.384798· ax16 + 0.0577948· ax17 + 0.11429· ax18 + 0.533959· ax19 + 0.244539· ax110 ).
ax28 = tanh(−0.0126921 − 0.0590893· ax11 − 0.0584762· ax12 + 0.428384· ax13 − 0.162718· ax14 −
(A37)
0.28943· ax15 − 0.165741· ax16 + 0.223617· ax17 − 0.6501· ax18 + 0.546167· ax19 − 0.57052· ax110 ).
ax29 = tanh(0.04489 − 0.605727· ax11 − 0.00105592· ax12 − 0.507067· ax13 − 0.354351· ax14 +
(A38)
0.727824· ax15 + 0.0357734· ax16 + 0.208872· ax17 − 0.277591· ax18 + 0.0822228· ax19 − 0.251631· ax110 ).
Materials 2021, 14, 1661 22 of 34

ax210 = tanh(0.0274462 − 0.464155· ax11 + 0.335326· ax12 − 0.189354· ax13 + 0.00273104· ax14 +
(A39)
0.138291· ax15 + 0.699764· ax16 + 0.378692· ax17 + 0.363206· ax18 + 0.698019· ax19 − 0.267254· ax110 ).
ax11 = tanh(4.767351553463 + 0.00010018415828968·CFO + 0.0041259755901707·WFO +
−1
0.000209184787522328· FAFO + 0.00147687126084511·CAFO − 0.170582379127354·WCFO −
0.00739795485330284·CSFO + 0.0014713281157207·CSO + 0.00573830926470374·WSO −
−1
(A40)
0.00348658190095467· FASO − 0.00375199793921971·CASO − 0.271415999837499·WCSO +
0.00269179297167401·CSSO + 0.000645907802623046·C − 0.00511651789903316·W −
0.00119068504591733· FA + 0.00121507017924426·CA − 0.194768583102371·WC −1 ).
ax12 = tanh(−2.90937028957849 + 0.000513987509738448·CFO + 0.0104205287263435·WFO −
−1
0.0000300236136823883· FAFO + 0.00193442295465733·CAFO + 0.0304831451994348·WCFO −
0.00348102557161981·CSFO − 0.00117363708739863·CSO + 0.00426298901301358·WSO −
−1
(A41)
0.00060247532418286· FASO + 0.000619551260777826·CASO + 0.202375734386783·WCSO −
0.000233978962466558·CSSO − 0.00038450344939621·C + 0.00544033538809217·W −
0.000295834802908707· FA − 0.001576885731436·CA − 0.437227409538789·WC −1 ).
ax13 = tanh(−1.28397136623645 − 0.00265425528655059·CFO + 0.00134392831181661·WFO +
−1
0.00187950192048241· FAFO − 0.000256323950062182·CAFO + 0.012594583425795·WCFO +
0.0137268911005708·CSFO − 0.00153196251341365·CSO − 0.0065857765027207·WSO +
−1
(A42)
0.00208609505432554 − FASO − 0.00000592967658050422·CASO − 0.11595281342328·WCSO −
0.0104040315572809·CSSO + 0.00196054736213827·C + 0.0102110804430631·W +
0.000620781117815345· FA − 0.00187136972051576·CA − 0.255163361753211·WC −1 ).
ax14 = tanh(4.21395574600285 − 0.00105127459631726·CFO − 0.0112148941682978·WFO +
−1
0.00163727698209709· FAFO + 0.000638406176270943·CAFO + 0.245800411366534·WCFO +
0.0074057996976356·CSFO − 0.00233892425018069·CSO + 0.00377676692330639·WSO −
−1
(A43)
0.000934383939230747· FASO − 0.00383296357061587·CASO + 0.599324030373907·WCSO −
0.00246137349085538·CSSO − 0.000522500768390411·C − 0.00430098689691885·W −
0.00200997561736293· FA + 0.00176580319743452·CA − 0.242934179812162·WC −1 ).
ax15 = tanh(−1.6446570342455 + 0.000533587828047032·CFO + 0.000952303651589457·WFO +
−1
0.000871116154702258· FAFO − 0.00000492904456363928·CAFO + 0.0360531074572857·WCFO −
0.00992286147701841·CSFO + 0.001469808247159·CSO + 0.00642860151461661·WSO −
−1
(A44)
0.00271657540072071· FASO + 0.000246865509647522·CASO + 0.0279125858257541·WCSO +
0.0133855714042725·CSSO − 0.000541184824066393·C − 0.00824577692013099·W −
0.00000397881537241383· FA + 0.00235002971670078·CA − 0.180135513126205·WC −1 ).
ax16 = tanh(−3.84884812355591 + 0.00115407871985479·CFO − 0.00375105915008307·WFO +
−1
0.000390252688951681· FAFO + 0.00221305190140043·CAFO + 0.327755674417453·WCFO +
0.00166910211893054·CSFO − 0.000117013083312986·CSO − 0.010706630159067·WSO +
−1
(A45)
0.000537751740713538· FASO − 0.000386509173984826·CASO + 0.038443112623923·WCSO +
0.00789947089570246·CSSO − 0.00163089480403489·C + 0.00485445960094249·W +
0.00174073824194078· FA + 0.00107642389772748·CA − 0.0903836711929193·WC −1 ).
Materials 2021, 14, 1661 23 of 34

ax17 = tanh(−1.87654029183636 + 0.00140014344181078·CFO − 0.00531433362617444·WFO −

−1
0.00413045250962745· FAFO + 0.00228016416194765·CAFO − 0.312440638205596·WCFO +
0.00244581570673698·CSFO − 0.00140933248958429·CSO + 0.00161013791429313·WSO +
−1
(A46)
0.00171544141339727· FASO − 0.000329656644298609·CASO − 0.448421258516497·WCSO −
0.00723447006861429·CSSO + 0.000353775031899132·C − 0.0017893761303401·W +
0.0018920172469581· FA + 0.00306888686044287·CA − 0.0131009807144161·WC −1 ).
ax18 = tanh(7.30775875969612 + 0.0025578242040305·CFO − 0.00303262947215815·WFO −
−1
0.00165574273119679· FAFO − 0.00306199296034884·CAFO − 0.156923718980124·WCFO +
0.00137102269966526·CSFO − 0.000419742348687123·CSO + 0.0061626385341524·WSO −
−1
(A47)
0.00237717598480315· FASO − 0.000972453336543304·CASO − 0.190327552460012·WCSO −
0.00783522799632837·CSSO − 0.000892188050113014·C − 0.0015154846950607·W +
0.000314135115650862· FA + 0.000748914790764869·CA − 0.184269602615267·WC −1 ).
ax19 = tanh(2.06633738444408 − 0.00195515963988164·CFO + 0.00138958650164713·WFO +
−1
0.00262801203222595· FAFO − 0.000536163722811296·CAFO − 0.188122517295472·WCFO +
0.00407903424912999·CSFO − 0.000183834803792603·CSO − 0.00183731758403244·WSO −
−1
(A48)
0.000774757922991741· FASO + 0.000574911771861783·CASO − 0.100967264097832·WCSO −
0.0083462680924114·CSSO + 0.000798708705572374·C − 0.00233339826228642·W −
0.00163577879701815· FA − 0.00020955799115713·CA − 0.28849090239718·WC −1 ).
ax110 = tanh(−5.09651077352556 + 0.000206496906491233·CFO + 0.00444536475350176·WFO +
−1
0.00109290014495785· FAFO + 0.00228266395774151·CAFO + 0.191531435179031·WCFO +
0.000987346948074925·CSFO + 0.000289024996218767·CSO + 0.00254484992089329·WSO +
−1
(A49)
0.00133254035553042· FASO + 0.0018234072550613·CASO + 0.332154625352893·WCSO −
0.00258354089455455·CSSO + 0.000785116534796457·C − 0.00462902690413978·W −
0.00163107840072543· FA − 0.000691117800517391·CA − 0.134078811558199·WC −1 ).

Appendix B
Appendix B contains the gradings and fitting curves for the designed recipes. Recipe
#1 designed according to the Bolomey method is presented in Figure A1, and that designed
according to the Fuller method is presented in Figure A2. Recipe #2 designed according to
the Bolomey method is presented in Figure A3, and that designed according to the Fuller
method is presented in Figure A4. Recipe #3 designed according to the Bolomey method is
presented in Figure A5, and that designed according to the Fuller method is presented in
Figure A6. Recipe #4 designed according to the Bolomey method is presented in Figure A7,
and that designed according to the Fuller method is presented in Figure A8.
 #1 designed according to the Bolomey method is presented in Figure A1, and that de-
signed according to the Fuller method is presented in Figure A2. Recipe #2 designed ac-
cording to the Bolomey method is presented in Figure A3, and that designed according to
the Fuller method is presented in Figure A4. Recipe #3 designed according to the Bolomey
Materials 2021, 14, 1661 24 of 34 is
method is presented in Figure A5, and that designed according to the Fuller method
presented in Figure A6. Recipe #4 designed according to the Bolomey method is presented
in Figure A7, and that designed according to the Fuller method is presented in Figure A8.

(A) (B) (C)

(D) (E) (F)

(G) (H) (I)

LEGEND:
—Natural sand
—Limestone gravel 4/10
—Limestone gravel 10/20
—Reference
—Result (combined)
Figure
FigureA1.
A1.Gradings
Gradingsand
and fitting
fitting curves for Recipe
curves for Recipe#1—standard
#1—standardmix
mix with
with plasticizer
plasticizer (lignosulfonate)
(lignosulfonate) andand superplasticizer
superplasticizer
(polycarboxylate ether) designed according to the Bolomey Method: (A) 10 MPa; (B) 15 MPa; (C) 20 MPa;
(polycarboxylate ether) designed according to the Bolomey Method: (A) 10 MPa; (B) 15 MPa; (C) 20 MPa; (D) (D) 25
25 MPa;
MPa; (E)
30 MPa; (F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.
(E) 30 MPa; (F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.
 Materials 2021, 14, 1661 25 of 34
Materials 2021, 14, x FOR PEER REVIEW 25 of 33

(A) (B) (C)

(D) (E) (F)

(G) (H) (I)

LEGEND:
—Natural sand
—Limestone gravel 4/10
—Limestone gravel 10/20
—Reference
—Result (combined)
Figure A2.A2.
Figure Gradings
Gradingsand fitting
and curves
fitting curvesforforRecipe
Recipe#1—standard
#1—standardmix
mix with
with plasticizer (lignosulfonate)and
plasticizer (lignosulfonate) andsuperplasticizer
superplasticizer
(polycarboxylate ether)
(polycarboxylate designed
ether) designedaccording
according to the Fuller
to the Method:
Fuller (A)(A)
Method: 10 MPa; (B) 15
10 MPa; (B)MPa; (C) 20
15 MPa; (C)MPa; (D) 25
20 MPa; (D)MPa; (E) 30
25 MPa;
MPa; (F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.
(E) 30 MPa; (F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.

(A) (B) (C)

 —Limestone gravel 10/20
—Reference
—Result (combined)
Figure2021,
Materials A2. 14,
Gradings
1661 and fitting curves for Recipe #1—standard mix with plasticizer (lignosulfonate) and superplasticizer
26 of 34
(polycarboxylate ether) designed according to the Fuller Method: (A) 10 MPa; (B) 15 MPa; (C) 20 MPa; (D) 25 MPa; (E) 30
MPa; (F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.

Materials 2021, 14, x FOR PEER REVIEW 26 of 33

(A) (B) (C)

(D) (E) (F)

(G) (H) (I)

LEGEND:
—Natural sand
—Limestone gravel 4/10
—Limestone gravel 10/20
—Reference
—Result (combined)
Figure
Figure A3.
A3. Gradings
Gradings andfitting
and fittingcurves
curvesfor
for Recipe
Recipe #2—standard
#2—standard mixmixwith
withplasticizer
plasticizer(lignosulfonate),
(lignosulfonate), superplasticizer
superplasticizer
(polycarboxylate
(polycarboxylate ether)
ether) andair
and airentrainer
entrainer(tensides)
(tensides) designed
designed according
according to
tothe
theBolomey
BolomeyMethod:
Method:(A) (A)1010MPa; (B)(B)
MPa; 15 15
MPa;
MPa;
(C)(C)
20 20 MPa;
MPa; (D)(D)
2525 MPa;(E)
MPa; (E)3030MPa;
MPa;(F)
(F)35
35MPa;
MPa; (G) 40 MPa;
MPa; (H)
(H)45
45MPa;
MPa;(I)
(I)5050MPa.
MPa.Legend
Legend in in
thethe
leftleft
bottom corner.
bottom corner.

(A) (B) (C)

 —Reference
—Result (combined)

Figure2021,
Materials A3.14,Gradings
1661 and fitting curves for Recipe #2—standard mix with plasticizer (lignosulfonate), superplasticizer
27 of 34
(polycarboxylate ether) and air entrainer (tensides) designed according to the Bolomey Method: (A) 10 MPa; (B) 15 MPa;
(C) 20 MPa; (D) 25 MPa; (E) 30 MPa; (F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.

(A) (B) (C)

Materials 2021, 14, x FOR PEER REVIEW 27 of 33

(D) (E) (F)

(G) (H) (I)

LEGEND:
—Natural sand
—Limestone gravel 4/10
—Limestone gravel 10/20
—Reference
—Result (combined)
Figure
Figure A4.A4. Gradingsand
Gradings andfitting
fittingcurves
curves for
for Recipe
Recipe #2—standard
#2—standard mixmixwith
withplasticizer
plasticizer(lignosulfonate),
(lignosulfonate), superplasticizer
superplasticizer
(polycarboxylate
(polycarboxylate ether)
ether) and
and airair entrainer
entrainer (tensides)designed
(tensides) designedaccording
accordingtotothetheFuller
FullerMethod:
Method:(A) (A)1010MPa;
MPa;(B)
(B)15
15MPa;
MPa;(C)
20 (C)
MPa;20 (D)
MPa;25(D) 25 MPa;
MPa; (E) 30(E) 30 MPa;
MPa; (F)MPa;
(F) 35 35 MPa;
(G) (G) 40 MPa;
40 MPa; (H)(H) 45 MPa;
45 MPa; (I) (I)
5050 MPa.
MPa. Legend
Legend ininthe
theleft
leftbottom
bottomcorner.
corner.

(A) (B) (C)

 —Reference
—Result (combined)

Figure2021,
Materials A4.14,Gradings
1661 and fitting curves for Recipe #2—standard mix with plasticizer (lignosulfonate), superplasticizer
28 of 34
(polycarboxylate ether) and air entrainer (tensides) designed according to the Fuller Method: (A) 10 MPa; (B) 15 MPa; (C)
20 MPa; (D) 25 MPa; (E) 30 MPa; (F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.

(A) (B) (C)

(D) (E) (F)

Materials 2021, 14, x FOR PEER REVIEW

(G) (H) (I)

LEGEND:
—Natural sand
—Limestone gravel 4/10
—Limestone gravel 10/20
—Reference
—Result (combined)

Figure
Figure A5. Gradings and fitting curves A5. Gradings
for Recipe and fitting
#3—standard curves
mix with for Recipe #3—standard
superplasticizer mixether),
(polycarboxylate with superplasticizer
retarder (poly
(phosphate)
(phosphate) and air entrainer (tensides) and air entrainer
designed according (tensides)
to the Bolomey designed
Method: according
(A) 10 MPa; (B)to
15the Bolomey
MPa; Method: (A) 10 MP
(C) 20 MPa;
(D)40
(D) 25 MPa; (E) 30 MPa; (F) 35 MPa; (G) 25MPa;
MPa;(H)
(E)45
30MPa;
MPa;(I)(F)
5035 MPa;
MPa. (G) 40inMPa;
Legend (H)bottom
the left 45 MPa; (I) 50 MPa. Legend in the left bo
corner.

(A) (B)
 —Limestone gravel 10/20
—Reference
—Result (combined)

Figure
Materials A5.14,
2021, Gradings
1661 and fitting curves for Recipe #3—standard mix with superplasticizer (polycarboxylate ether), retarder
29 of 34
(phosphate) and air entrainer (tensides) designed according to the Bolomey Method: (A) 10 MPa; (B) 15 MPa; (C) 20 MPa;
(D) 25 MPa; (E) 30 MPa; (F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.

(A) (B) (C)

(D) (E) (F)

(G) (H) (I)

LEGEND:
—Natural sand
—Limestone gravel 4/10
—Limestone gravel 10/20
—Reference
—Result (combined)
Figure
Figure A6.
A6. Gradingsand
Gradings andfitting
fittingcurves
curvesfor
forRecipe
Recipe #3—standard
#3—standard mixmixwith
withsuperplasticizer
superplasticizer (polycarboxylate
(polycarboxylate ether), retarder
ether), retarder
(phosphate)
(phosphate) andair
and airentrainer
entrainer(tensides)
(tensides) designed
designed according
according to
to the
theFuller
FullerMethod:
Method:(A)(A)1010MPa;MPa;(B)(B)
15 15
MPa;
MPa;(C)(C)
20 MPa;
20 MPa;
(D)
(D) 2525 MPa;(E)
MPa; (E)3030MPa;
MPa;(F)(F)35
35MPa;
MPa; (G)
(G) 40 MPa;
MPa; (H)
(H)45
45MPa;
MPa;(I)(I)5050MPa.
MPa.Legend
Legendinin
thethe
leftleft
bottom corner.
bottom corner.
 Materials
Materials 14,14,
2021,
2021, 1661PEER REVIEW
x FOR 30 of
2934
of 33

(A) (B) (C)

(D) (E) (F)

(G) (H) (I)

LEGEND:
—Natural sand
—Limestone gravel 4/10
—Limestone gravel 10/20
—Reference
—Result (combined)

Figure A7.A7.
Figure Gradings and
Gradings fitting
and fittingcurves
curvesfor
forRecipe
Recipe#4—standard with plasticizer
#4—standard mix with plasticizer(lignosulfonate)
(lignosulfonate) and
and superplasticizer
superplasticizer
(naphthalene)
(naphthalene)designed according
designed accordingtotothe
theBolomey
BolomeyMethod:
Method: (A)
(A) 10 MPa; (B)
(B) 15
15MPa;
MPa;(C)(C)2020MPa;
MPa;(D)
(D)2525 MPa;
MPa; (E)(E)
30 30 MPa;
MPa;
(F) (F)
35 35
MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend
MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.
left bottom corner.

(A) (B) (C)

 —Limestone gravel 10/20
—Reference
—Result (combined)

Figure2021,
Materials A7. 14,
Gradings
1661 and fitting curves for Recipe #4—standard mix with plasticizer (lignosulfonate) and superplasticizer
31 of 34
(naphthalene) designed according to the Bolomey Method: (A) 10 MPa; (B) 15 MPa; (C) 20 MPa; (D) 25 MPa; (E) 30 MPa;
(F) 35 MPa; (G) 40 MPa; (H) 45 MPa; (I) 50 MPa. Legend in the left bottom corner.

Materials 2021, 14, x FOR PEER REVIEW 30 of 33

(A) (B) (C)

(D) (E) (F)

(G) (H) (I)

LEGEND:
—Natural sand
—Limestone gravel 4/10
—Limestone gravel 10/20
—Reference
—Result (combined)
Figure
Figure A8.A8. Gradings
Gradings andfitting
and fittingcurves
curvesfor
forRecipe
Recipe #4—standard
#4—standard mix mixwith
withplasticizer
plasticizer(lignosulfonate) and
(lignosulfonate) superplasticizer
and superplasticizer
(naphthalene)
(naphthalene) designed
designed accordingtotothe
according theFuller
FullerMethod:
Method:(A)(A)1010MPa;
MPa;(B)
(B)15
15MPa;
MPa;(C)(C)2020MPa;
MPa;(D)
(D)2525MPa;
MPa;(E)
(E)30
30 MPa;
MPa; (F)
35 (F)
MPa; (G) 40(G)
35 MPa; MPa; (H) 45
40 MPa; (H)MPa; (I) 50
45 MPa; (I)MPa. Legend
50 MPa. in the
Legend leftleft
in the bottom corner.
bottom corner.

References
References
Dimov,
1. 1. Dimov, D.;D.; Amit,I.;I.;Gorrie,
Amit, Gorrie,O.;
O.;Barnes,
Barnes,M.D.;
M.D.; Townsend,
Townsend, N.J.;
N.J.; Neves,
Neves, A.I.S.;
A.I.S.;Withers,
Withers,F.;F.;Russo,
Russo,S.;S.;
Craciun, M.F.
Craciun, Ultrahigh
M.F. Ultrahigh
Performance
Performance NanoengineeredGraphene-Concrete
Nanoengineered Graphene-Concrete Composites
Composites for
forMultifunctional
MultifunctionalApplications.
Applications. Adv.
Adv.Funct. Mater.
Funct. 2018,
Mater. 28, 28,
2018, 28. 28,
[CrossRef]
doi:10.1002/adfm.201705183.
2. 2. Marchon,
Marchon, D.;D.; Cementand
Cement andConcrete
ConcreteResearch. Mechanisms of
Research. Mechanisms of Cement
CementHydration;
Hydration;Elsevier:
Elsevier: Amsterdam,
Amsterdam, The Netherlands,
The 2016.
Netherlands, 2016.
3. 3. Scrivener,
Scrivener,
K.;K.; Snellings,R.;
Snellings, R.;Lothenbach,
Lothenbach,B. APractical
B.A Practical Guide
Guide to
to Microstructural
MicrostructuralAnalysis
AnalysisofofCementitious
CementitiousMaterials; CRC
Materials; Press:
CRC Boca
Press: Boca
Raton,
Raton, FL,FL, USA,
USA, 2018.
2018.
4. Kurdowski, W. Cement and Concrete Chemistry; Springer: Heidelberg, Germany, 2014.
4. Kurdowski, W. Cement and Concrete Chemistry; Springer: Heidelberg, Germany, 2014.
5. Young, J.F.; Mindess, S.; Darwin, D. Concrete; Prentice Hall: Upper Saddle River, NJ, USA, 2002; ISBN 0130646326.
5. Young, J.F.; Mindess, S.; Darwin, D. Concrete; Prentice Hall: Upper Saddle River, NJ, USA, 2002; ISBN 0130646326.
6. Hover, K.C. The influence of water on the performance of concrete. Constr. Build. Mater. 2011, 25, 3003–3013. [CrossRef]
6. Hover, K.C. The influence of water on the performance of concrete. Constr. Build. Mater. 2011, 25, 3003–3013,
doi:10.1016/j.conbuildmat.2011.01.010.
7. Jensen, O.; Cement and Concrete Research. Water-Entrained Cement-Based Materials: I. Principles and Theoretical Background; Else-
vier: Amsterdam, The Netherlands, 2001.
8. Jensen, O.; Cement and Concrete Research. Water-Entrained Cement-Based Materials: II. Experimental Observations; Elsevier: Am-
sterdam, The Netherlands, 2002.
Materials 2021, 14, 1661 32 of 34

7. Jensen, O.; Cement and Concrete Research. Water-Entrained Cement-Based Materials: I. Principles and Theoretical Background; Elsevier:
Amsterdam, The Netherlands, 2001.
8. Jensen, O.; Cement and Concrete Research. Water-Entrained Cement-Based Materials: II. Experimental Observations; Elsevier:
Amsterdam, The Netherlands, 2002.
9. Jiménez, C.; Barra, M.; Valls, S.; Aponte, D.; Vázquez, E. Durabilidad de Hormigones Con Áridos Reciclados Diseñados Con El
Método de Volumen de Mortero Equivalente (EMV): Validación Bajo El Contexto Español y Adaptación a La Metodología de
Bolomey; Durability of Recycled Aggregate Concrete Designed with the Equivalent Mortar Volume (EMV) Method: Validation
under the Spanish Context and Its Adaptation to Bolomey Methodology. Mater. Constr. 2014, 64, 6. [CrossRef]
10. Abdelgader, H.S. How to design concrete produced by a two-stage concreting method. Cem. Concr. Res. 1999, 29, 331–337.
[CrossRef]
11. Abdelgader, H.S.; Suleiman, R.E.; El-Baden, A.S.; Fahema, A.H.; Angelescu, N. Concrete mix proportioning using three equations
method (Laboratory Study). In Proceedings of the UKIERI Concrete Congress Innovations in Concrete Construction, Jalandhar,
Punjab, India, 5–8 March 2013.
12. Iqbal, M.F.; Liu, Q.-F.; Azim, I.; Zhu, X.; Yang, J.; Javed, M.F.; Rauf, M. Prediction of mechanical properties of green concrete
incorporating waste foundry sand based on gene expression programming. J. Hazard. Mater. 2020, 384, 121322. [CrossRef]
[PubMed]
13. Iqbal, M.F.; Javed, M.F.; Rauf, M.; Azim, I.; Ashraf, M.; Yang, J.; Liu, Q.-F. Sustainable utilization of foundry waste: Forecasting
mechanical properties of foundry sand based concrete using multi-expression programming. Sci. Total. Environ. 2021, 146524,
146524. [CrossRef]
14. Renigier-Biłozor, M.; Chmielewska, A.; Walacik, M.; Janowski, A.; Lepkova, N. Genetic algorithm application for real estate
market analysis in the uncertainty conditions. Neth. J. Hous. Environ. Res. 2021, 1–42. [CrossRef]
15. Liu, Q.-F.; Iqbal, M.F.; Yang, J.; Lu, X.-Y.; Zhang, P.; Rauf, M. Prediction of chloride diffusivity in concrete using artificial neural
network: Modelling and performance evaluation. Constr. Build. Mater. 2021, 268, 121082. [CrossRef]
16. Ziolkowski, P.; Niedostatkiewicz, M. Machine Learning Techniques in Concrete Mix Design. Materials 2019, 12, 1256. [CrossRef]
17. Kurpinska, M.; Kułak, L. Predicting Performance of Lightweight Concrete with Granulated Expanded Glass and Ash Aggregate
by Means of Using Artificial Neural Networks. Materials 2019, 12, 2002. [CrossRef]
18. Czarnecki, S.; Sadowski, Ł.; Hoła, J. Artificial neural networks for non-destructive identification of the interlayer bonding between
repair overlay and concrete substrate. Adv. Eng. Softw. 2020, 141, 102769. [CrossRef]
19. Aitcin, P. The durability characteristics of high performance concrete: A review. Cem. Concr. Compos. 2003, 25, 409–420. [CrossRef]
20. Al-Obaidi, S.; Bamonte, P.; Luchini, M.; Mazzantini, I.; Ferrara, L. Durability-Based Design of Structures Made with Ultra-High-
Performance/Ultra-High-Durability Concrete in Extremely Aggressive Scenarios: Application to a Geothermal Water Basin Case
Study. Infrastructures 2020, 5, 102. [CrossRef]
21. Cheng, Y.; Zhang, Y.; Tan, G.; Jiao, Y. Effect of Crack on Durability of RC Material under the Chloride Aggressive Environment.
Sustainability 2018, 10, 430. [CrossRef]
22. K˛epniak, M.; Woyciechowski, P.; Łukowski, P.; Kuziak, J.; Kobyłka, R. The Durability of Concrete Modified by Waste Limestone
Powder in the Chemically Aggressive Environment. Materials 2019, 12, 1693. [CrossRef] [PubMed]
23. Ambroziak, A.; Ziolkowski, P. Concrete Compressive Strength Under Changing Environmental Conditions During Placement
Processes. Materials 2020, 13, 4577. [CrossRef]
24. Abdelgader, H.S.; Saud, A.F.; Othman, A.M.; Fahema, A.H.; El-Baden, A.S. Concrete Mix Design Using the Double-Coating
Method. Plant Precast. Technol. 2014, 80, 66–74.
25. Rajamane, N.P.; Ambily, P.S.; Nataraja, M.C.; Das, L. Discussion: Modified Bolomey equation for strength of lightweight concretes
containing fly ash aggregates. Mag. Concr. Res. 2014, 66, 1286–1288. [CrossRef]
26. Zhang, X.-B.; Deng, S.-C.; Deng, X.-H.; Qin, Y.-H. Experimental research on regression coefficients in recycled concrete Bolomey
formula. J. Central South Univ. Technol. 2007, 14, 314–317. [CrossRef]
27. Gołaszewski, J.; Kostrzanowska-Siedlarz, A.; Cygan, G.; Drewniok, M. Mortar as a model to predict self-compacting concrete
rheological properties as a function of time and temperature. Constr. Build. Mater. 2016, 124, 1100–1108. [CrossRef]
28. Abdelgader, H.S.; El-Baden, A.S.; Shilstone, J.M. Bolomeya model for normal concrete mix design. J. Concr. Plant Int. 2012, 2,
68–74.
29. Kaplan, G.; Yaprak, H.; Memiş, S.; Alnkaa, A. Artificial Neural Network Estimation of the Effect of Varying Curing Conditions
and Cement Type on Hardened Concrete Properties. Buildings 2019, 9, 10. [CrossRef]
30. Tang, C.-W.; Cheng, C.-K.; Tsai, C.-Y. Mix Design and Mechanical Properties of High-Performance Pervious Concrete. Materials
2019, 12, 2577. [CrossRef] [PubMed]
31. Yeh, I.-C. Modeling of strength of high-performance concrete using artificial neural networks. Cem. Concr. Res. 1998, 28, 1797–1808.
[CrossRef]
32. Lee, S.-C. Prediction of concrete strength using artificial neural networks. Eng. Struct. 2003, 25, 849–857. [CrossRef]
33. Hoła, J.; Schabowicz, K. New technique of nondestructive assessment of concrete strength using artificial intelligence. NDT E Int.
2005, 38, 251–259. [CrossRef]
34. Hoła, J.; Schabowicz, K.; Hoùa, J. Application of Artificial Neural Networks to Determine Concrete Compressive Strength Based
on Non-Destructive Tests. J. Civ. Eng. Manag. 2005, 6, 23–32. [CrossRef]
Materials 2021, 14, 1661 33 of 34

35. Gupta, R.; Kewalramani, M.A.; Goel, A. Prediction of Concrete Strength Using Neural-Expert System. J. Mater. Civ. Eng. 2006, 18,
462–466. [CrossRef]
36. Bui, D.-K.; Nguyen, T.; Chou, J.-S.; Nguyen-Xuan, H.; Ngo, T.D. A modified firefly algorithm-artificial neural network expert
system for predicting compressive and tensile strength of high-performance concrete. Constr. Build. Mater. 2018, 180, 320–333.
[CrossRef]
37. Deng, F.; He, Y.; Zhou, S.; Yu, Y.; Cheng, H.; Wu, X. Compressive strength prediction of recycled concrete based on deep learning.
Constr. Build. Mater. 2018, 175, 562–569. [CrossRef]
38. Naderpour, H.; Rafiean, A.H.; Fakharian, P. Compressive strength prediction of environmentally friendly concrete using artificial
neural networks. J. Build. Eng. 2018, 16, 213–219. [CrossRef]
39. McCormac, J.; Brown, R. Design of Reinforced Concrete; John Wiley & Sons Inc.: Hoboken, NJ, USA, 2015.
40. Adil, M.; Ullah, R.; Noor, S.; Gohar, N. Effect of number of neurons and layers in an artificial neural network for generalized
concrete mix design. Neural Comput. Appl. 2020, 1–9. [CrossRef]
41. Nunez, I.; Marani, A.; Nehdi, M.L. Mixture Optimization of Recycled Aggregate Concrete Using Hybrid Machine Learning
Model. Materials 2020, 13, 4331. [CrossRef] [PubMed]
42. Marani, A.; Jamali, A.; Nehdi, M.L. Predicting Ultra-High-Performance Concrete Compressive Strength Using Tabular Generative
Adversarial Networks. Materials 2020, 13, 4757. [CrossRef]
43. Toniolo, G.; Prisco, M. Reinforced Concrete Design to Eurocode 2; Springer: Heidelberg, Germany, 2017.
44. Andrei, N. An adaptive scaled BFGS method for unconstrained optimization. Numer. Algorithms 2018, 77, 413–432. [CrossRef]
45. Battiti, R.; Masulli, F. BFGS Optimisation for Faster and Automated Supervised Learning. In International Neural Network
Conference; Springer: Dordrecht, The Netherlands, 1990; pp. 757–760.
46. Berahas, A.S.; Takáč, M. A robust multi-batch L-BFGS method for machine learning. Optim. Methods Softw. 2019, 35, 191–219.
[CrossRef]
47. Hagan, M.; Menhaj, M. Training feedforward networks with the Marquardt algorithm. IEEE Trans. Neural Netw. 1994, 5, 989–993.
[CrossRef] [PubMed]
48. Li, D.-H.; Fukushima, M. A modified BFGS method and its global convergence in nonconvex minimization. J. Comput. Appl.
Math. 2001, 129, 15–35. [CrossRef]
49. Abdi, F.; Shakeri, F. A globally convergent BFGS method for pseudo-monotone variational inequality problems. Optim. Methods
Softw. 2017, 34, 25–36. [CrossRef]
50. Grabowska, K.; Szczuko, P. Ship Resistance Prediction with Artificial Neural Networks. In Proceedings of the IEEE 2015 Signal
Processing: Algorithms, Architectures, Arrangements, and Applications (SPA), Poznan, Poland, 23–25 September 2015.
51. Le, D.; Huang, W.; Johnson, E. Neural network modeling of monthly salinity variations in oyster reef in Apalachicola Bay in
response to freshwater inflow and winds. Neural Comput. Appl. 2019, 31, 6249–6259. [CrossRef]
52. Luenberger, D.G.; Ye, Y. Linear and Nonlinear Programming; Springer: New York, NY, USA, 2008.
53. Yildizel, S.A.; Arslan, Y. Flexural Strength Estimation of Basalt Fiber Reinforced Fly-Ash Added Gypsum Based Composites. J.
Eng. Res. Appl. Sci. 2018, 7, 829–834.
54. Sanjuán, M.A.; Argiz, C.; Menéndez, E. Efecto de la adición de mezclas de ceniza volante y ceniza de fondo procedentes del
carbón en la resistencia mecánica y porosidad de cementos Portland. Mater. Construcción 2013, 63, 49–64. [CrossRef]
55. Sanjuán, M.Á.; Estévez, E.; Argiz, C.; del Barrio, D. Effect of curing time on granulated blast-furnace slag cement mortars
carbonation. Cem. Concr. Compos. 2018, 90, 257–265. [CrossRef]
56. Poloju, K.K.; Anil, V.; Manchiryal, R.K. Properties of Concrete as Influenced by Shape and Texture of Fine Aggregate. Am. J. Appl.
Sci. Res. 2017, 3, 28. [CrossRef]
57. Chinchillas-Chinchillas, M.J.; Corral-Higuera, R.; Gómez-Soberón, J.M.; Arredondo-Rea, S.P.; Alamaral-Sánchez, J.L.; Acuña-
Aguero, O.H.; Rosas-Casarez, C.A. Influence of the Shape of the Natural Aggregates, Recycled and Silica Fume on the Mechanical
Properties of Pervious Concrete. Int. J. Adv. Comput. Sci. Appl. 2014, 4, 216–220.
58. Jo, H.-S.; Park, C.; Lee, E.; Choi, H.K.; Park, J. Path Loss Prediction Based on Machine Learning Techniques: Principal Component
Analysis, Artificial Neural Network, and Gaussian Process. Sensors 2020, 20, 1927. [CrossRef]
59. Choi, Y.-Y.; Shon, H.; Byon, Y.-J.; Kim, D.-K.; Kang, S. Enhanced Application of Principal Component Analysis in Machine
Learning for Imputation of Missing Traffic Data. Appl. Sci. 2019, 9, 2149. [CrossRef]
60. Gadekallu, T.R.; Khare, N.; Bhattacharya, S.; Singh, S.; Maddikunta, P.K.R.; Ra, I.-H.; Alazab, M. Early Detection of Diabetic
Retinopathy Using PCA-Firefly Based Deep Learning Model. Electronics 2020, 9, 274. [CrossRef]
61. Abdi, H.; Williams, L.J. Principal Component Analysis. Wiley Interdiscip. Rev. Comput. Stat. 2010, 2, 433–459. [CrossRef]
62. Song, F.; Guo, Z.; Mei, D. Feature Selection Using Principal Component Analysis. In 2010 International Conference on System Science,
Engineering Design and Manufacturing Informatization, Proceedings of the 2010 International Conference on System Science, Engineering
Design and Manufacturing, Yichang, China, 18 November 2010; Institute of Electrical and Electronics Engineers (IEEE): New York,
NY, USA, 2010; Volume 1, pp. 27–30.
63. Tabakhi, S.; Moradi, P.; Akhlaghian, F. An unsupervised feature selection algorithm based on ant colony optimization. Eng. Appl.
Artif. Intell. 2014, 32, 112–123. [CrossRef]
64. Song, Y.; Liang, J.; Lu, J.; Zhao, X. An efficient instance selection algorithm for k nearest neighbor regression. Neurocomputing
2017, 251, 26–34. [CrossRef]
Materials 2021, 14, 1661 34 of 34

65. Yu, H.; Reiner, P.D.; Xie, T.; Bartczak, T.; Wilamowski, B.M. An Incremental Design of Radial Basis Function Networks. IEEE
Trans. Neural Netw. Learn. Syst. 2014, 25, 1793–1803. [CrossRef] [PubMed]
66. Iqbal, S.Z.; Gull, H.; Ahmed, J. Incremental Sorting Algorithm. In 2009 Second International Conference on Computer and Electrical
Engineering, Proceedings of the 2009 Second International Conference on Computer and Electrical Engineering, Dubai, United Arab Emirates,
15 January 2010; Institute of Electrical and Electronics Engineers (IEEE): New York, NY, USA, 2009; Volume 2, pp. 378–381.
67. Gürbüzbalaban, M.; Ozdaglar, A.; Parrilo, P.A. On the Convergence Rate of Incremental Aggregated Gradient Algorithms. SIAM
J. Optim. 2017, 27, 1035–1048. [CrossRef]
68. Shobha, G.; Rangaswamy, S. Machine Learning. In Handbook of Statistics; Elsevier B.V.: Amsterdam, The Netherlands, 2018;
Volume 38, pp. 197–228, ISBN 9780444640420.