Cement and Concrete Research 124 (2019) 105808

Contents lists available at ScienceDirect

Cement and Concrete Research

journal homepage: www.elsevier.com/locate/cemconres

Laboratory investigation and modeling of concrete pavements containing T

AOD steel slag
Tanvi Gupta*, S.N. Sachdeva
Department of Civil Engineering, National Institute of Technology, Kurukshetra - 136119, India

A R T I C LE I N FO A B S T R A C T

Keywords: The present study emphasizes the potential use of Argon Oxygen Decarburization (AOD) steel slag in concrete
AOD steel slag production. Five mixes along with the control mix were made by partially replacing cement with AOD steel slag
Compressive strength varying from 10%, 15%, 20%, and 25%. The different properties of all the concrete mixes were evaluated by
Flexural strength assessing slump, compressive strength, flexural strength and other durability properties. The air permeability
Durability properties
index and sorptivity index showed positive results with an increased percentage of AOD steel slag in the mixes.
Artificial neural network
From the study, it is concluded that AOD steel slag can easily be replaced by cement in the construction of rigid
pavements. Further, we have also constructed a Feed-Forward Neural Network (FFNN) to predict the com-
pressive and the flexural strengths of steel slag mixed cement concrete for pavements.

1. Introduction as a boon to the nation's development. However, shortage of resources

and high construction and material cost remain the fundamental ex-
Increase in industrialization in order to achieve the economic planation behind the construction of flexible pavements in spite of
growth has resulted in environmental degradation globally. Presently in knowing the predominance of rigid pavements over flexible pavements
India, approximately 960 million tons of solid waste is being produced [3, 4]. Further, a large amount of CO2 gas is released in the environ-
annually as by-products of mining, agricultural, municipal and in- ment during the production of concrete ingredients such as cement and
dustrial activities. Out of this around 350 million tons are organic aggregates and thus, leads to accelerating the global warming [5].
wastes from agricultural activities, 290 million tons are inorganic in Generation of 1 ton of cement produces almost equal amount of CO2
nature generated from mining and industrial sectors and about gas [6] which is found to be approximately equal to 7% of the total gas
4.5 million tons of waste are hazardous in nature [1]. Therefore, im- produced globally [7]. Therefore, replacing cement partially with in-
portance of solid waste management is increasing because of increasing dustrial by-product in construction of concrete pavements will be a
amount of waste products generated, scarcity and increased costs of boon for disposal of industrial waste and less usage of cement.
operating landfills.
The thermal power plants are the major source of industrial solid 1.1. Type of steel slags
wastes for generating fly ash, the steel and iron industries producing
ground-granulated blast furnace slag (GGBS) and other steel melting India is known as the third largest manufacturer of steel and also the
slag, non-ferrous industries such as zinc, aluminum, copper and iron second largest manufacturer of stainless steel in the world. Different
tailings and red mud, cement industry producing cement kiln dust, si- manufacturing processes are followed for manufacturing of steel and
lica fume, wood ash [2]. Further, with increased environment aware- stainless steel. The type, property and chemical composition of slag
ness regarding the hazardous effects of the industrial by-products, uti- highly depends upon the manufacturing process followed for the pro-
lization of these products as a substitute in construction industry has duction. Steel slag and steel sludge are the various forms of by-products
become an attractive alternative to disposal. generated from the steel industry [8]. During the steel production,
Concrete is said to be the most used resource after water because of approximately 2–4 tons of wastes are being produced for per ton of the
its high durable nature and almost no maintenance is required if con- steel. The different form of slag and sludge that are produced from steel
crete structures are constructed properly. Due to this, the concrete plants is blast furnace slag, sludge and dust, Linz Donawitz (LD) con-
pavements are considered over flexible pavements and are considered verter slag, sludge and dust, mill sludge, mill scale, acetylene sludge,

*
Corresponding author.
E-mail addresses: tanvigupta0314@gmail.com (T. Gupta), snsachdeva@yahoo.co.in (S.N. Sachdeva).

https://doi.org/10.1016/j.cemconres.2019.105808
Received 20 March 2019; Received in revised form 2 May 2019; Accepted 2 July 2019
0008-8846/ © 2019 Elsevier Ltd. All rights reserved.
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

etc. Also, the chemical composition of all these materials differs de- samples. The study concluded that the samples from the edges showed
pending on the source of generation, raw materials fed and manu- leaching and carbonation traces and the sample from road center was
facturing process adopted but usually consists of some useful metals the same as fresh slag.
and other elements such as iron, calcium, carbon zinc, lead, which can In [18], the authors aimed at verifying the application of two types
be recycled and reused in a prudent manner [9]. of EAF slag as substitute to natural aggregates in base course and
At the time of manufacturing of stainless steel, two main slag i.e. concrete asphalt base for flexible pavements. With the help of this study
Argon Oxygen Decarburization (AOD) slag and Electric Arc Furnace the authors concluded that EAF slag aggregates presented better me-
(EAF) slag are produced. Nowadays more than 80% of the stainless chanical properties as compared to the corresponding asphalt with
steels are manufactured using the AOD refining process. It has been natural aggregates and satisfied the requirements for its use in road
estimated that around 270 kg of slag is generated during the production construction.
of every 1 ton stainless steel in the process of refining. The production In [19], the authors considered and compared two formulations in
of slag was approximately 8.6 million tons in the world in 2011 out of which dredged sediment was treated using the two types of steel slag.
which 3.4 million tons of slag was produced in China alone. Further, Their study also highlighted the stability issue of the steel slag in road
keeping in mind the environmental conservation and protection re- construction and has shown its acceptability with construction guide-
quirements, the pollution caused due to production of AOD slag pre- lines and standards.
vents the development of stainless steel production [10]. Although In [20], the authors analyzed the potential of utilization of Imperial
enormous amount of the solid wastes from the steel and stainless in- Smelting Furnace (ISF) slag in production of concrete. In the study the
dustry has been used in different applications however there is still authors have replaced fine sand by ISF slag in different percentages.
around 35% slag that is unused and dumped as waste [11]. This is a The authors found from their study an improvement in the compressive
huge amount considering the problems faced by the environment due to strength and comparable flexural strength.
dumping of these wastes, which may contain significant amount of In [21], the authors performed detailed analysis of calcium alumi-
heavy metals. Generally, EAF slag and stabilized AOD slag are used as nate rich secondary steel slag in order to know its behavior as a ce-
an aggregate, mostly in road construction. However, use of these slag is mentitious material. In the study the authors investigated hydration,
not preferable as an aggregate as boron in the stabilized slag may leach strength development and porosity of cement composites where the
out posing environmental risks. Compositionally, AOD slag resembles cement was replaced by the secondary steel slag. It was concluded from
Ordinary Portland cement (OPC) therefore, indicating its potential as a the study that steel slag acts as a good cementitious material.
binder than as aggregate. In [22], the authors studied the benefits of a combined admixture of
Nowadays, in order to reduce environmental hazards applications of thermally treated Flue Gas Desulphurization (FGD) gypsum and slag
steel slag have increased manifolds. They have mostly been used in powder on concrete. Further, the authors concluded that these ad-
construction; as a construction material for hydro-engineering struc- mixtures together have synergistic benefits as compared to slag alone.
tures, as a binder agent, as an independent binder, as a material for soil In [23], the authors aimed of producing a high value cementitious
stabilization in road construction, as a material in unbounded mixtures material by adding alkalis to stainless steel refining slag and steam
for road beds or as an aggregate in asphalt mixtures [12]. curing at 80°C. The authors found that silicates in the activation solu-
tion lead to high compressive strength. FT-IR and thermal analysis
1.2. Related works showed the presence of C-S-H and carbonated phases. Further in [24],
the authors carried out life cycle assessment when EAF slag was used in
In [13], the authors have investigated the use of stainless steel slag pavements and its comparison with the traditional raw materials was
in concrete by studying the physiochemical and mineralogical proper- done. The authors concluded that important environmental benefits
ties of AOD and EAF aggregates. The study showed that the use of AOD could be attained by replacing coarse aggregates with black slag in road
and EAF slag aggregates can slightly decrease concrete durability construction.
properties and increase linear expansion. Further, in [14], the authors In [25], the authors incorporated steel slag as coarse aggregates in
have utilized the steel slag, fine sand, compound agent, gravel and alkali activated slag concrete and alkali activated slag fly ash concrete.
admixture to prepare cement-free steel slag cementitious material. The Further, the authors have carried out durability tests on both types of
authors conducted the durability test on the mortar and the result in- concrete and ecological and economic analysis was also done.
dicated low carbonation depth of 0.3 mm and relative dynamic mod- In [26], the authors replaced natural coarse aggregates with steel
ulus of elasticity maintains at 66%. slag and have used domestic waste water for mixing of concrete. The
In [15], the authors have utilized Electric Arc Furnace (EAF) slag as authors found that 50% replacement of basalt aggregate with steel slag
aggregates in cement-treated materials for road construction and stu- aggregate indicated the improvement in compressive strength, flexural
died its durability and mechanical properties. The authors in the study strength and modulus of elasticity of concrete. Rapid chloride perme-
highlighted that the application of these EAF slag aggregates produced ability tests, ultrasonic pulse velocity and SEM analysis indicated dense
a higher compaction difficulty, but also showed very good mechanical micro structure and enhanced durability.
performances. Further they concluded that if the mix is correctly de- In [27], the authors aimed at preparing steel slag–fly ash–phosphor-
signed by properly stabilizing the amount of natural aggregate re- gypsum solidified material, that completely composed of solid wastes to
placement then these mixtures could be suitable and durable solution be utilized as road base material. The study concluded that the 28 days
for sub-base and base pavement layers. and 360 days strength of this material can reach 8 MPa and 12 MPa
In [16], the authors carried out a comprehensive study to evaluate respectively and its resilience modulus reached 1987 MPa.
the properties, mechanical behavior and environmental stability of In [28], the authors carried out a series of strength and durability
concrete designed with 100% of EAF slag as aggregates. The authors tests on concrete consisting of waste glass and steel slag. The authors
also compared EAF slag aggregate concrete mechanical properties with noticed that waste glass was able to improve the workability of slag
the reference concrete consisting of natural aggregates and concluded added concrete and addition of waste glass to steel slag concrete also
that EAF slag concrete is compatible to be used in civil constructions. limited the increase in density. It was also concluded that when both
In [17], the authors proposed a laboratory method for accelerated aging steel slag and waste glass were added to concrete, it highly improved
of steel slag in order to predict technical and environmental properties the fire resistance properties of concrete.
of slag starting from fresh slag. In the study the authors took road In [29], the authors studied the bond between the steel reinforce-
samples from the center and edges from a 10-year-old asphalt pavement ment and the recycled coarse aggregates. In this study, the author re-
consisting of EAF slag as aggregates and compared it with fresh slag placed natural coarse aggregates with EAF slag aggregates completely

2
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

and observed enhanced bond strength at low water-cement ratio. Table 1

Chemical composition of AOD steel slag.

1.3. Research significance Composition Percentage (%)

SiO2 33.52
Utilization of AOD steel slag (waste from stainless steel plant) as a CaO 38.64
replacement of cement has not been investigated vastly by many re- FeO 1.63
searchers. Production of cement contributes majorly in global warming MgO 11.98
Al2O3 4.76
by releasing CO2 gas in the environment. Further, the overuse of natural
MnO 6.95
resources for cement manufacturing contributes to the environmental Cr2O3 2.5
degradation. Therefore, usage of AOD steel slag as a partial substitute of
cement has twofold effect, i.e. safe disposal of stainless steel industry
waste by using it as a construction material and less usage of cement Table 2
further helps in maintaining environmental stability and ecological Chemical composition of cement.
balance. Therefore, the present study emphasizes on studying the im-
Chemical composition of cement (%)
pact when up to 25% of cement is replaced with AOD steel slag for the
construction of rigid pavements. Further, prediction model for AOD SiO2 20.05
steel slag mixed cement concrete has not been developed before (to the Al2O3 4.34
Fe2O3 3.21
best of our knowledge). In this paper, we have implemented a feed-
Cao 62.11
forward artificial neural network which has been trained to predict the Mgo 3.53
compressive and flexural strength of AOD slag mixed cement concrete K2O + Na2O 1.82
for pavements. SO3 3.03

2. Experimental investigations

2.1. Materials

Steel slag has been procured from Jindal Stainless Steel Limited,
Hisar, Haryana. Fig. 1 shows the stabilized steel slag used for the study.
AOD steel slag chemical composition is shown in Table 1. Ordinary
Portland Cement (OPC) 43 grade complying with IS: 8112 (1989) [30]
and having specific gravity of 3.15 was used. The consistency and fi-
neness of cement used was 28.5% and 5 mm respectively. The chemical
components of OPC used for the study along with their percentage is
shown in Table 2. Stone dust was used as the fine aggregates in the
study having water absorption of 0.35% and specific gravity of 2.63.
Proportioning of 10 mm and 20 mm aggregates was done in such a way
that the coarse aggregates of 16 mm nominal size having water ab- Fig. 2. Gradation curve of 10 mm aggregates, 20 mm aggregates and stone dust.
sorption and specific gravity of 0.45% and 2.74 respectively was used.
Gradation curve for the 10 mm aggregates, 20 mm aggregates and stone 2.2. Mix design
dust is shown in Fig. 2. In order to achieve a mix of maximum density,
proportioning of aggregates was done. Further, for the desired mix 5 mixes were prepared including the M40 control mix and others by
design 36% by weight of fine aggregates, 32% by weight of 10 mm partially replacing cement by AOD steel slag up to 25%. In the present
aggregates and 32% by weight of 20 mm were used. For the preparation study, 0.38 has been taken as the constant water/cement ratio for all
of the mix and for moist curing of specimens potable water conforming the mixes and varying the percentage replacement of cement from 10%
to IS: 456 (2000) [31] was used. to 25%. Adopting 0.38 value of water cement ratio and following the
specifications of IRC: 44- 2008 [32] and IS: 10262- 2009 [33], the
concrete mix design for M40 is developed and is shown in Table 3. The
dosage of super plasticizer with specific gravity of 1.2 has been adjusted
in order to achieve the desired workability in all the mixes.

2.3. Testing

Fresh concrete undergoes a slump test and to carry out this test a
standard slump test apparatus was used for measuring as per IS: 1199-
1959 [34]. The aim of this test was to know about the workability and
consistency of the different concrete mixes.
Analysis of the compressive strength of the 150 × 150 × 150 mm
cubes as per IS: 516 (BIS 1959) [35] is performed. 45 cubes of the same
size were cast to carry out this test, so that 3 cubes for each mix are
available for testing after 3, 7 and 28 days of moist curing. With the
help of concrete beams as per IS: 516 [35] the flexural strength was
determined. A total of 45 beams were cast (3 beams for each mix) of
size 100 × 100 × 500 mm for evaluating the flexural strength after 3, 7
and 28 days of moist curing.
Fig. 1. Pictorial view of AOD steel slag. The auto clam permeability apparatus was created by Queen's

3
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

Table 3
Mix design of concrete.
S. no Mix grade Cement Steel slag as % of CeMa Steel slag Fine aggregates Coarse aggregates Water S.P.% by weight of CeMa S.P Water/Cem ratio
(kg/m3) (kg/m3) (kg/m3) (kg/m3) (kg/m3) (kg/m3)

1 M40 416 - - 648 1223 158 1 4.16 0.38

2 S10 411.89 10 45.77 627 1218 158 1 4.577 0.345
3 S15 389.01 15 68.65 622 1209 158 1 4.577 0.345
4 S20 366.13 20 91.53 617 1200 158 1 4.577 0.345
5 S25 343.24 25 114.41 613 1191 158 1 4.577 0.345

Fig. 5. Flexural strength of mixes at different curing ages.

Fig. 3. Autoclam permeability system.

air permeability index (API) in ln(mbar)/min. In order to conduct this

Table 4
test, 15 plates of dimension 250 × 250 × 75 mm (for each mix) were
Slump value for considered mixes.
cast and moist cured for 28 days.
Concrete mix Slump value (mm) The sorptivity (water absorption) test can be conducted on the same
M40 25.5
apparatus, but at least two hours after the air permeability test. Water
S10 27.5 was admitted into the test area through a pump with the air escaping
S15 29 through the bleed tube. When the test chamber was completely filled
S20 31 with water the pump automatically switches off and the micro pump
S25 31.5
pressurizes the test area to 20 mbar above atmospheric pressure. The
test then starts. The volume of water is measured and recorded every
minute for the duration of 15 min. A plot of the quantity of water ab-
sorbed and the square root of time elapsed was plotted. The slope of this
graph is reported as the sorptivity index with units of m3 × 10−7/ min .
The test was performed by casting 15 plates of dimension
250 × 250 × 75 mm (for each mix) and curing it for 28 days. The auto
clam permeability setup used for carrying out air permeability test and
sorptivity is shown in Fig. 3.
The Cantabro test was carried out to evaluate the property of
abrasion resistance of the concrete mixes. This test is conducted in the
Los Angeles (LA) abrasion machine without the steel ball charges.
Further, in the test the mass of cubes was noted before performing the
test and was made to rotate for around 10–12 min in the Los Angeles
(LA) abrasion machine. The mass of the specimen was noted again after
moving it in the abrasion machine and the difference in mass was used
to characterize the abrasion resistance of cubes [37]. 30 cubes of size
100 × 100 × 100 mm were cast so that 3 cubes from each mix can be
tested after curing for 7 and 28 days.
Fig. 4. Compressive strength of mixes at different curing ages. The acidic environment was generated in the laboratory using 5%
sulfuric acid (H2SO4). 30 concrete specimens of size 150 × 150 × 150
University Belfast to measure the water permeability, sorptivity and air mm were cast and moist cured for 28 days. Further, those specimens
permeability of concrete plates on site [36]. In this study a standard were placed in H2SO4 solution tanks for 28 days and 63 days. The so-
base ring was used to isolate the test area and the instrument was lutions were checked and restored after 14 days to maintain constant
manually pressurized with the help of a syringe attached to the base of acid level throughout the investigation. After 28 and 63 days, the
the testing apparatus. As the pressure reaches 500 mbar, the test started concrete specimens were removed from the acid solution tanks, washed
automatically and pressure decrease was observed every minute for the with tap water, cleaned with cloth, and surface dried mass was noted.
duration of 15 min. The natural logarithm of air pressure was plotted The mass of specimens before and after exposure to acidic environment
against time, with the slope of the last five data points reported as the was noted and their mass difference was used to measure the resistance

4
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

Fig. 6. Pressure vs time graph for different mixes.

mix for water-cement ratio 0.38 should be not more than 25 ± 5 mm as

per IRC- 44: 2008. Therefore, slump test for all the concrete mix having
AOD slag showed good consistency but slightly lower workability.

3.2. Compressive strength

The results of compressive strength are shown in Fig. 4. The results

highlighted that the compressive strength values for each mix was very
high after 28 days of curing as compared to the strength after 3 and
7 days curing. After curing of 3 days, the compressive strength of S10
was 18.78 MPa and was reduced to 12.64 MPa when cement was re-
placed by 25% AOD slag. Further, the compressive strength of M40
concrete mix after 28 days of moist curing was 46.98 MPa but showed
the slight decrease in strength as cement was replaced by AOD slag.
Even after the decrease, the values of compressive strength fulfill the
Fig. 7. Air permeability index for different mixes. design requisites of rigid pavements. The compressive strength of S25
after 28 days curing was just 14.73% less in comparison to that of the
to acidic environment. M40 mix.
Durability of concrete specimens from the sulfate attack was as-
sessed by placing the cube specimens into MgSO4 solution for 28 days.
3.3. Flexural strength
15 concrete specimens of dimensions 150 × 150 × 150 mm were made
so that 3 cubes of different mix can be tested after curing in water for
Flexural strength of all mixes at different curing ages is shown in
28 days. The specimens were weighed before putting them into 5%
Fig. 5. Since there is relation between flexural strength and compressive
MgSO4 solution. Change in compressive strength gave the deterioration
strength, therefore, the flexural strength also followed the same pattern
caused to the concrete specimen due to sulfate attack.
as that of compressive strength. The flexural strength of control mix
M40 was 3.97 MPa after curing of 7 days and reduced to 2.865 MPa
3. Results and discussion when 25% of cement was replaced by AOD steel slag. The values of
flexural strength have not decreased drastically after cement was re-
3.1. Slump test placed with slag in the concrete mixes. The flexural strength of S10
after 28 days of curing was 4.836 MPa and was reduced by approxi-
The slump test value for all the concrete mixes is given in Table 4. It mately 20% when 25% of cement was replaced. The flexural strength
was seen that all the slump values were almost within the range that is values of S15 and S20 after 28 days of curing was 4.257 MPa and
required for the design of rigid pavements. The slump value of concrete 4.065 MPa respectively.

5
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

Fig. 8. Flow quantity versus time graph for all mixes.

Fig. 9. Sorptivity index for all the mixes. Fig. 10. Loss in mass due to abrasion for all the mixes.

3.4. Air permeability 3.5. Sorptivity

Fig. 6 shows the pressure versus time graph for all the 5 mixes. The sorptivity index was observed to decrease with the increase in
Further, the slope of these graphs gave the value of air permeability the percentage of AOD slag in the concrete mixes after curing for
index (ln (Pressure)/min) after 28 days of curing as shown in Fig. 7. All 28 days. Fig. 8 shows the graphs between flow quantity and min for
the mixes had air permeability index less than 0.01 ln(mbar)/min, all the mixes that were used for analyzing the sorptivity index
which depicts very good quality of concrete mixes. As the replacement (m3 × 10−7/ min ) as shown in Fig. 9. The decrease in the sorptivity
percentage of cement increased the air permeability index (API) de- index at higher percentage of AOD slag was due to fine particles of steel
creased, i.e. S25 mix showed excellent result having API of 0.01 ln slag that made the mixes dense and less permeable. The sorptivity index
for control mix M40 was 0.676 m3 × 10−7/ min and reduced to 0.386
(mbar)/min. The results depicted that all the mixes comprises of very
m3 × 10−7/ min for S25.
less pores filled with air and hence they are very much suitable for the
construction of rigid pavements.

6
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

3.7. Acid attack

Figs. 11 and 12 show the compressive strength and weight loss of

concrete mixes after the concrete cubes were exposed to acidic (H2SO4)
environment for 28 and 63 days. The weight loss due to acid attack
increased when the exposure time of concrete specimens in the acid was
increased to 63 days. The weight loss of 6.43% was observed when the
control mix M40 specimen was kept in acid for 28 days. Further, the
loss in mass of the same specimen increased by approximately 40%
when the exposure time was increased to 63 days. The ingress of acidic
solution inside the concrete specimens reduced the compressive
strength of S25 to 22.56 MPa. The compressive strength value for S15
was 24.73 MPa corresponding to 12.45% of weight loss. It was observed
that, J15 and J20 had almost the same amount of strength and weight
loss.

Fig. 11. Compressive strength of specimens exposed to acid solution.

3.8. Sulfate attack

The specimens exposed to 5% MgSO4 solution for 28 and 63 days

depicted that the concrete comprising of AOD slag is less resistant to the
sulfate attack as compared to the control mix. From Fig. 13, it can be
noticed that as the percentage of replacement of AOD slag in concrete
increases, the compressive strength decreases. It was observed that the
compressive strength of control mix M40 after 63 days exposure was
35.45 MPa and was reduced to 28.6 MPa when 25% cement was re-
placed. The compressive strength value for S10, S15 and S20 after
28 days exposure was 37.84 MPa, 36.46 MPa and 33.57 MPa respec-
tively.

4. Prediction of compressive and flexural strength using Feed-

Forward Neural Network

Fig. 12. Loss in mass after being exposed to acid solution. In this section, we will explore the application of FFNN for the
prediction of compressive and flexural strength. It has been proved in
the literature that any discrete function F : DF ⊂ RnI → Rm can be ac-
curately approximated to any degree of accuracy by the F^ on DF, where
DF refers to a compact subset of RnI . The only condition to achieve this
is to have sufficient count of hidden nodes present in the network and
the activation function used must be continuous, bounded and non-
constant [38, 39]. Before going further into the analysis, in the fol-
lowing subsection we have briefly described the structure of FFNN.

4.1. Theoretical foundation of FFNN

The structure of FFNN contains adjustable links that connect the

signal from one neuron of one layer to the neuron of the next layer.
These links are called weights and they are adjusted during the training
process. These weights will capture the hidden relationship existing
between the input-output training data samples. In our paper, we have
Fig. 13. Compressive strength of specimens exposed to sulfate solution. considered only one hidden layer in the FFNN model. Its structure is
shown in Fig. 14.
3.6. Abrasion resistance In FFNN, we have input to hidden layer and hidden to output layer
weights that are to be adjusted during the training procedure. These
Abrasion resistance is an important characteristic of rigid pave- weights actually capture the hidden mathematical relationship existing
ments and this property was studied to know about the wear and tear between the input and output samples. The structure of FFNN is shown
caused on the rigid pavements due to moving vehicles. Abrasion re- in Fig. 14. In FFNN, only feed-forward connections are there. There is
sistance is directly related to the toughness and strength, i.e. the higher no feedback connection. The red dashed square box actually shows the
the strength the lesser will be the mass loss due to abrasion. Fig. 10 working of the hidden layer neurons. Each hidden layer neuron receives
shows the loss in mass of concrete specimens due to abrasion. The loss a weighted sum of input signals. It is called induced field and is denoted
in mass was observed to increase as the amount of AOD slag was in- by V (k). Then a nonlinear activation function operates on this induced
creased. Loss in mass for S25 was 0.56% after curing for 7 days which field and produces the output which is the output of hidden neuron. The
was approximately 32% more as compared to the control mix. The loss output layer neurons receive the weighted sum of the output of hidden
in mass due to abrasion for S10, S15 and S20 was 0.33%, 0.38% and layer neurons. Since linear activation function is considered for the
0.42% respectively. output layer neurons, hence their outputs will be equal to their induced
fields. In the present case, there are two output neurons and their

7
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

Fig. 14. FFNN structure.

expressions are given as follows: and the update equation is given by

1
YNN (k ) = V11 (k ) ∂ET (k )
W o1 (k + 1) = W o1 (k + 1) − η
∂W o1 (k ) (5)
= S1 (k ) w1o1 (k ) + S2 (k ) w2o1 (k ) + S3 (k ) w3o1 (k ). ...+Sq (k ) wqo1 (k )
(1) where η represents the learning rate which governs the speed of the
learning. Its value lies in the interval [−1,1]. Similarly, the update
and
equation for the second output weight vector is given by:
2
YNN (k ) = V12 (k ) ∂ET (k )
W o2 (k + 1) = W o1 (k + 1) − η
= S1 (k ) w1o2 (k ) + S2 (k ) w2o2 (k ) + S3 (k ) w3o2 (k ). ...+Sq (k ) wqo2 (k ) ∂W o2 (k ) (6)
(2) By following the above procedure, the update equations for the
where V11 (k ) and V12 (k ) denote the induced fields of first and second input to hidden layer weight vectors can be found. The sequence of
output neuron respectively. S1(k),S2(k),..Sq(k) denote the output of the training steps is shown in the form of a flowchart as shown in Fig. 15.
hidden neurons (where q denotes the total number of hidden layer
neurons). 4.3. Pre-processing of the experimental data

4.1.1. Learning algorithm for adjusting the weights of the FFNN Since each input and output signal in the training data can have any
To adjust the weights of the FFNN we need some kind of me- individual range and to have effective prediction results from the FFNN
chanism/procedure. In this paper, we have used the Back-propagation normalization procedure can be used. In this procedure, we make the
method (which is based on a Gradient descent principle) to derive the signal values to lie in a desired range, for example [0,1] then the
weights update equation. First, an objective function is defined which is learning speed of the FFNN will be improved as the FFNN will be more
to be minimized. The most popular one is the Mean Square Error (MSE) sensitive to the input values falling in this range. In our paper we have
function. It is defined as follows: performed a linear normalization using the following normalization
2 formula:
ET (k ) = ∑ Ei = E1 (k ) + E2 (k ) X − Xmin
i=1 (3) N=
1
Xmax − Xmin (7)
1
where ET(k) is the TMSE and E1 (k ) = 2 (Y1 (k ) − YNN (k ))2 and
1 2 where N represents the normalized value of the variable X and Xmax and
E2 (k ) = 2 (Y2 (k ) − YNN (k ))2 . The algorithm aims at minimizing the
Xmin represents the maximum and minimum value respectively of the
ET(k) by adjusting the values of the weights of the FFNN model.
input variable X.
4.2. Update equations for weight vectors
4.4. Preparation of training and testing/validation data set
The update equations for the first output weight vectors are given
below: In our study we have experimentally obtained a total of 45 input-
output samples out of which we have used 30 data pairs for the training
∂ET (k ) ∂ET (k ) ∂Y NN1 (k ) and the rest of the 15 for the testing/validation purpose. Total 8 inputs
= ×
∂W o1 (k ) ∂Y NN1 (k ) ∂W o1 (k ) (4) are there on which the two outputs viz compressive and flexural

8
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

0.9

0.8

0.7

0.6

Response
0.5

0.4

0.3

0.2

0.1
Y (k)
1
0
Y1 (k)
NN
−0.1
5 10 15 20 25 30 35 40 45
Samples

1
Fig. 16. Response of YNN (k ) of FFNN at the end of the training.

1
Y (k)
2
0.9 2
YNN(k)
0.8

0.7

0.6

Response
0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30 35 40 45
Samples

2
Fig. 17. Response of YNN (k ) of FFNN at the end of the training.

0.035
1
MSE for YNN(k)

0.03

0.025

0.02
MSE

0.015

0.01

0.005

0
0 500 1000 1500 2000 2500
Epochs

1
Fig. 18. MSE obtained for YNN (k ) of FFNN during the training.
Fig. 15. Sequence of steps involved in the training of FFNN model.
0.045
2
MSE for YNN
Table 5 0.04

Range of input parameters in the experimental data. 0.035

Input parameters Minimum value Maximum value 0.03

0.025
W/C ratio 0.3450 0.3800
MSE

Water (kg/m3) 158.1 158.1 0.02

Cement (kg/m3) 343.2400 416.0500
0.015
Steel slag (kg/m3) 0 114.41
Coarse aggregates (kg/m3) 1191.54 1247.31 0.01
Fine aggregates (kg/m3) 613.619 661.595
0.005
Super plasticizer (kg/m3) 4.16 4.5770
Curing time (days) 3 28 0
0 500 1000 1500 2000 2500
Epochs

2
strengths depends. Table 5 shows the minimum and maximum range of Fig. 19. MSE obtained for YNN (k ) of FFNN during the training.
the 8 input parameters that were used for modeling of results.
following 2 performance indices: Mean Average Error (MAE) and Mean
Square Error (MSE). These indices values will show how accurately the
4.5. Performance evaluation indices
identification model has understood the unknown relationship among
the training samples. These indices are defined below.
To test the performance of the FFNN model we will be using the

9
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

0.25
MAE for Y1 (k)
where Ti(k) and Pi(k) denote the desired and predicted value.
NN

0.2 5. Prediction results

0.15 We have used a single hidden layer FFNN containing only 6 hidden
neurons. Since, two quantities are to be predicted so we have two
MAE

0.1
neurons in the output layer. Further, the total experiment data consist
of 45 input-output samples. Out of which 30 samples are used during
the training stage and the rest of the 15 are used during the validation
0.05
stage. In the hidden and output layers of FFNN we have used tangent
hyperbolic function and linear activation function respectively. The
0
0 500 1000 1500 2000 2500 learning rate η value is set equal to 0.051. A total of 2500 epochs are
Epochs
used during the training. The responses obtained from the FFNN model
1
Fig. 20. MAE obtained for YNN (k ) of FFNN during the training. outputs are shown in Figs. 16 and 17. It can be seen from the figures
that the outputs of the FFNN are able to track the desired outputs.
0.2 Further, the plots of MSE obtained for both the FFNN model outputs
2
MAE for YNN(k)
during the training are shown in Figs. 18 and 19 respectively and the
0.18
MAE plots obtained are shown in Figs. 20 and 21. It can be seen that as
0.16 the training progressed the weights of the FFNN tuned in the right di-
rection and the value of MSE and MAE keep getting reduced with the
0.14
training.
MAE

0.12 Now after the training we test the generalization ability of the FFNN
by applying those input-output data samples that are not used during
0.1
the training. The results are shown in Figs. 22 and 23. It can be seen
0.08 from the figures that the prediction given by the FFNN outputs are very
close to the actual outputs. This shows the ability of FFNN in approx-
0.06
0 500 1000 1500 2000 2500
Epochs
imating the hidden relationship exiting between the input-output ex-
perimental data.
2
Fig. 21. MAE obtained for YNN (k ) of FFNN during the training.
6. Conclusion
1.1
Y (k)
1

1.05
1
YNN(k) The present study emphasized on analyzing the consequences of use
of AOD steel slag on the various properties of concrete mixes. Moreover,
1
to know the impact of AOD steel slag on fresh, mechanical and dur-
0.95
ability properties when slag is partially replaced by cement was in-
Response

vestigated. A total of 5 mixes including the control mix consisting of

0.9 different proportions of AOD steel slag were prepared and tested. On
0.85
the basis of the results obtained, the following conclusions can be
drawn:
0.8

1. With the same amount of water in all the mixes, replacement of AOD
0.75
32 34 36 38 40 42 44
Samples
slag by cement did not reduce the slump values by a higher degree.
The slump values for all the concrete mixes were in the range of
1
Fig. 22. Response of YNN (k ) of FFNN during the validation stage. 25–50 mm, which is suitable for rigid pavements construction.
2. Addition of higher percentage of AOD steel slag as cement re-
1
Y2(k)
placement reduced the mechanical properties like compressive
Y2NN(k) strength and flexural strength. Still the compressive strength value
0.9
of concrete mix incorporated with AOD slag was more than 40 MPa
0.8 for all mix. Hence, 25% replacement of cement by slag is proved to
be favorable for construction of concrete pavements.
Response

0.7 3. The low values of air permeability index and sorptivity index as the
amount of AOD steel slag in the mix increases are highly favorable
0.6
for the concrete pavements. The lower indices value shows that
0.5
incorporation of slag into mix makes it more dense and durable.
4. Some other durability properties like abrasion resistance, acid attack
0.4 and sulfate attack showed a bit negative results at the higher re-
32 34 36 38 40 42 44
Samples placement levels but this can be managed by quality control at the
2
Fig. 23. Response of YNN (k ) of FFNN during the validation stage. time of construction and by adding chemical admixtures to mini-
mize their impact.
N
5. The prediction model based on FFNN has been successfully devel-
1 oped which can predict the compressive and flexural strength of
MAE =
N
∑ |Ti (k ) − Pi (k )|
i=1 (8) AOD slag mixed cement concrete pavements given the values of the
8 parameters on which these quantities are dependent.
N 6. As large amount of AOD steel slag is generated every year and is left
1
MSE =
N
∑ (Ti (k ) − Pi (k ))2 unused or dumped in landfills, its replacement in concrete as a ce-
i=1 (9) mentitious material will help in its disposal. Higher amounts of AOD

10
T. Gupta and S.N. Sachdeva Cement and Concrete Research 124 (2019) 105808

slag i.e. more than 25% can also be used for the construction of low [19] M. Miraoui, R. Zentar, N.-E. Abriak, Road material basis in dredged sediment and
traffic volume rigid pavements and village roads. basic oxygen furnace steel slag, Construct. Build. Mater. 30 (2012) 309–319.
[20] B. Tripathi, A. Misra, S. Chaudhary, Strength and abrasion characteristics of ISF slag
concrete, J. Mater. Civ. Eng. 25 (11) (2012) 1611–1618.
References [21] A. Mladenovič, B. Mirtič, A. Meden, V.Z. Serjun, Calcium aluminate rich secondary
stainless steel slag as a supplementary cementitious material, Construct. Build.
Mater. 116 (2016) 216–225.
[1] A. Pappu, M. Saxena, S.R. Asolekar, Solid wastes generation in India and their re-
[22] X.-L. Guo, H.-S. Shi, H.-Y. Liu, Effects of a combined admixture of slag powder and
cycling potential in building materials, Build. Environ. 42 (6) (2007) 2311–2320.
thermally treated flue gas desulphurization (FGD) gypsum on the compressive
[2] R. Siddique, Utilization of industrial by-products in concrete, Proc. Eng. 95 (2014)
strength and durability of concrete, Mater. Struct. 42 (2) (2009) 263–270.
335–347.
[23] M. Salman, Ö. Cizer, Y. Pontikes, R. Snellings, L. Vandewalle, B. Blanpain, K. Van
[3] S. Singh, G. Ransinchung, S. Debbarma, P. Kumar, Utilization of reclaimed asphalt
Balen, Cementitious binders from activated stainless steel refining slag and the ef-
pavement aggregates containing waste from sugarcane mill for production of con-
fect of alkali solutions, J. Hazard. Mater. 286 (2015) 211–219.
crete mixes, J. Clean. Prod. 174 (2018) 42–52.
[24] V.J. Ferreira, A.S.-D.-G. Vilaplana, T. García-Armingol, A. Aranda-Usón, C. Lausín-
[4] S. Singh, G. Ransinchung, P. Kumar, Feasibility study of RAP aggregates in cement
González, A.M. López-Sabirón, G. Ferreira, Evaluation of the steel slag incorpora-
concrete pavements, Road Mater. Pavement Des. 20 (1) (2019) 151–170.
tion as coarse aggregate for road construction: technical requirements and en-
[5] Z. Cao, L. Shen, J. Zhao, L. Liu, S. Zhong, Y. Sun, Y. Yang, Toward a better practice
vironmental impact assessment, J. Clean. Prod. 130 (2016) 175–186.
for estimating the CO2 emission factors of cement production: an experience from
[25] N. Palankar, A.R. Shankar, B. Mithun, Durability studies on eco-friendly concrete
China, J. Clean. Prod. 139 (2016) 527–539.
mixes incorporating steel slag as coarse aggregates, J. Clean. Prod. 129 (2016)
[6] M. Faisal, K. Muhammad, Synthesis and characterization of geopolymer from ba-
437–448.
gasse bottom ash, waste of sugar industries and naturally available China clay, J.
[26] S. Saxena, A. Tembhurkar, Impact of use of steel slag as coarse aggregate and
Clean. Prod. 129 (2016) 491–495.
wastewater on fresh and hardened properties of concrete, Construct. Build Mater.
[7] A. Rerkpiboon, W. Tangchirapat, C. Jaturapitakkul, Strength, chloride resistance,
165 (2018) 126–137.
and expansion of concretes containing ground bagasse ash, Construct. Build. Mater.
[27] W. Shen, M. Zhou, W. Ma, J. Hu, Z. Cai, Investigation on the application of steel
101 (2015) 983–989.
slag-fly ash-phosphogypsum solidified material as road base material, J. Hazard.
[8] N.H. Roslan, M. Ismail, Z. Abdul-Majid, S. Ghoreishiamiri, B. Muhammad,
Mater. 164 (1) (2009) 99–104.
Performance of steel slag and steel sludge in concrete, Construct. Build. Mater. 104
[28] X. Yu, Z. Tao, T.-Y. Song, Z. Pan, Performance of concrete made with steel slag and
(2016) 16–24.
waste glass, Construct. Build. Mater. 114 (2016) 737–746.
[9] B. Das, S. Prakash, P. Reddy, V. Misra, An overview of utilization of slag and sludge
[29] F. Faleschini, A. Santamaria, M.A. Zanini, J.-T. San José, C. Pellegrino, Bond be-
from steel industries, Resour. Conserv. Recycl. 50 (1) (2007) 40–57.
tween steel reinforcement bars and electric arc furnace slag concrete, Mater. Struct.
[10] H.-q. Zhao, Y.-h. Qi, Y.-l. Shi, X.-z. Na, H.-l. Feng, Mechanism and prevention of
50 (3) (2017) 170.
disintegration of AOD stainless steel slag, J. Iron Steel Res. Int. 20 (4) (2013) 26–30.
[30] Indian Standard, Grade ordinary Portland cement-specification, Tech. rep., Tech.
[11] H. Motz, J. Geiseler, Products of steel slags an opportunity to save natural resources,
Rep. IS: 8112, Bureau of Indian Standards, New Delhi, India, 2005View at (43).
Waste Manag. 21 (3) (2001) 285–293.
[31] Indian Standard Plain, Reinforced Concrete-Code of Practice (4th Revision), (2000).
[12] I. Netinger, D. Bjegović, G. Vrhovac, Utilisation of steel slag as an aggregate in
[32] IRC 44, Tentative Guidelines for Cement Concrete Mix Design for Pavements, New
concrete, Mater. Struct. 44 (9) (2011) 1565–1575.
Delhi Indian Road Congress, 2008, pp. 1–16.
[13] G. Adegoloye, A.-L. Beaucour, S. Ortola, A. Noumowé, Concretes made of EAF slag
[33] IS 10262-2009, Concrete mix proportioning – guidelines.
and AOD slag aggregates from stainless steel process: mechanical properties and
[34] BIS (Bureau of Indian Standards), Indian Standard Methods of Sampling and
durability, Construct. Build. Mater. 76 (2015) 313–321.
Analysis of Concrete, Indian Standards, 1959.
[14] X.-d. Xiang, J.-c. Xi, C.-h. Li, X.-w. Jiang, Preparation and application of the cement-
[35] IS: 516, Indian Standard Methods of Tests for Strength of Concrete, (1959).
free steel slag cementitious material, Construct. Build. Mater. 114 (2016) 874–879.
[36] K. Yang, P. Basheer, B. Magee, Y. Bai, Investigation of moisture condition and
[15] F. Autelitano, F. Giuliani, Electric arc furnace slags in cement-treated materials for
Autoclam sensitivity on air permeability measurements for both normal concrete
road construction: mechanical and durability properties, Construct. Build. Mater.
and high performance concrete, Construct. Build Mater. 48 (2013) 306–314.
113 (2016) 280–289.
[37] Q. Dong, H. Wu, B. Huang, X. Shu, K. Wang, Investigation into laboratory abrasion
[16] L. Rondi, G. Bregoli, S. Sorlini, L. Cominoli, C. Collivignarelli, G. Plizzari, Concrete
test methods for pervious concrete, J. Mater. Civ. Eng. 25 (7) (2012) 886–892.
with EAF steel slag as aggregate: a comprehensive technical and environmental
[38] R. Kumar, S. Srivastava, J. Gupta, Modeling and adaptive control of nonlinear dy-
characterisation, Compos. Part B 90 (2016) 195–202.
namical systems using radial basis function network, Soft Comput. 21 (15) (2017)
[17] P. Suer, J.-E. Lindqvist, M. Arm, P. Frogner-Kockum, Reproducing ten years of road
4447–4463.
ageing–accelerated carbonation and leaching of EAF steel slag, Sci. Total Environ.
[39] Rajesh Kumar, Smriti Srivastava, J.R.P. Gupta, Amit Mohindru, Temporally local
407 (18) (2009) 5110–5118.
recurrent radial basis function network for modeling and adaptive control of non-
[18] M. Pasetto, N. Baldo, Experimental evaluation of high performance base course and
linear systems, ISA transactions 87 (2019) 88–115.
road base asphalt concrete with electric arc furnace steel slags, J. Hazard. Mater.
181 (1-3) (2010) 938–948.

11