Académique Documents
Professionnel Documents
Culture Documents
PREDICTION OF COMPRESSIVE
STRENGTH OF HIGH PERFORMANCE
CONCRETE CONTAINING INDUSTRIAL BY
PRODUCTS USING ARTIFICIAL NEURAL
NETWORKS
Dr. B. Vidivelli
Professor, Department of Civil & Structural Engineering,
A. Jayaranjini
Research Scholar,
Department of Civil & Structural Engineering,
Annamalai University, Tamilnadu, India
ABSTRACT
This paper presents artificial neural network (ANN) based model to
predict the compressive strength of concrete containing Industrial Byproducts
at the age of 28, 56, 90 and 120 days. A total of 71 specimens were casted with
twelve different concrete mix proportions. The experimental results are
training data to construct the artificial neural network model. The data used
in the multilayer feed forward neural network models are arranged in a
format of ten input parameters that cover the age of specimen, cement, Fly
ash, Silica fume, Metakaolin, bottom ash, sand, Coarse aggregate, water and
Superplasticizer. According to these parameter in the neural network models
are predicted the compressive strength values of concrete containing
Industrial Byproducts. This study leads to the conclusion that the artificial
neural network (ANN) performed well to predict the compressive strength of
high performance concrete for various curing period.
Key word: Compressive Strength, High Performance Concrete, Industrial by
Products, Neurons, Neural Network.
Cite this Article: Dr. B.Vidivelli and A. Jayaranjini. Prediction of
Compressive Strength of High Performance Concrete Containing Industrial by
products Using Artificial Neural Networks, International Journal of Civil
Engineering and Technology, 7(2), 2016, pp. 302314.
http://www.iaeme.com/IJCIET/issues.asp?JType=IJCIET&VType=7&IType=2
http://www.iaeme.com/IJCIET/index.asp 302
editor@iaeme.com
INTRODUCTION
In view of the global sustainable development, it is imperative that supplementary
cementing materials be used in replace of cement in the concrete industry. The most
worldwide available supplementary cementing materials are silica fume (SF), a byproduct of silicon metal and fly ash (FA), a by-product of thermal power stations, and
blast-furnace slag (BS), a byproduct of steel mill. It is estimated that approximately
600 million tons of FA are available worldwide now, but at present, the current
worldwide utilization rate of FA in concrete is about 10%. However, the recent
development of green high performance concrete (GHPC) brings the abundant
utilization of these mineral mixtures. When these different reactive mineral admixtures are added into concrete at the same time, they develop their own characteristics
with the development. SF can increase the strength of the concrete significantly;
however, it affects the workability of the fresh concrete greatly, while adding large
amount of FA to the concrete contributes the workability of the concrete but not to the
strength. In addition, those mineral admixtures show different effects on the strength
of the concrete within different ages due to their different pozzolan reactions. The aim
of this study is to build models which have two different architectures in ANN system
to evaluate the effect of FA, MK, SF and BA on compressive strength of concrete. For
purpose of constructing this models, 12 different mixtures with 36 specimens of the
28 days compressive strength results of concrete containing FA, SF, MK and BA used
in training for ANN system were collected for the Experimental work. In training of
models constituted with different architectures. The age of specimen(AS), Cement(C),
Fly ash (FA), Silica fume(SF), Metakaolin(MK), Sand(S), Bottom ash (BA), Coarse
aggregate (CA), Water(W) and Superplasticizer(SP) were entered as input; while
compressive strength(fc) values were used as output. The models were trained with 71
data of experimental results were obtained.
LITERATURE REVIEW
Noorzaei et al. (2007) focused on development of artificial neural networks (ANNs)
for prediction of compressive strength of concrete after 28 days. To predict the
compressive strength of concrete six input parameters cement, water, silica fume,
super plasticizer, fine aggregate and coarse aggregate were identified considering
two hidden layers for the architecture of neural network. The results of the study
indicated that ANNs have strong potential as a feasible tool for predicting the
compressive strength of concrete. Atici et al., (2009) applies multiple regression
analysis and an artificial neural network in estimating the compressive strength of
concrete that contains varying amounts of blast furnace slag and fly ash. The results
reveal that the artificial neural network models performed better than multiple
regression analysis models. Serkan subas (2009) investigated that the estimation
ability of the effects of utilizing different amount of the class C fly ash on the
mechanical properties of cement using artificial neural network and regression
methods. Experimental results were used in the estimation methods. The developed
models and the experimental results were compared in the testing data set. As a result,
compressive and flexural tensile strength values of mortars containing various
amounts class C fly ash can be predicted in a quite short period of time with tiny error
rates by using the multilayer feed-forward neural network models than regression
techniques. Seyed et.al (2011) studied the application of artificial neural networks to
predict compressive strength of high strength concrete (HSC). A total of 368 different
data of HSC mix-designs were collected from technical literature. The authors
concluded that that the relative percentage error (RPE) for the training set was 7.02%
http://www.iaeme.com/IJCIET/index.asp 303
editor@iaeme.com
and the testing set was 12.64%. The ANNs models give high prediction accuracy, and
the research results demonstrate that using ANNs to predict concrete strength is
practical and beneficial. Vijay et al., (2013) predicted the compressive strength of
concrete using Artificial Neural Network (ANN). The authors compared the predicted
compressive strength with the obtained actual compressive strength of concrete and
also the authors proposed equations for different models. The authors concluded that a
good co-relation has been obtained between the predicted compressive strength by
these models and experimental results. Sakshigupta et.al., (2013) used Artificial
Neural Network (ANN) to predict the compressive strength of concrete containing
nano-silica. The author developed a model for predicting 28 days compressive
strength of concrete with partial replacement of cement with nano-silica for which the
data has been taken from various literatures. The author concluded that compressive
strength values of concrete can be predicted in ANN models without attempting any
experiments in a quite short period of time with some error rates. Wankhade et.al,
(2013) used Artificial Neural Network (ANN) to predict the compressive strength
of concrete. To train the networks back propagation and JordanElman algorithms
are used. Networks are trained and tested at various learning rate and momentum
factor and after many trials these were kept constant for this study. Performance of
networks were checked with statistical error criteria of correlation coefficient, root
mean squared error and mean absolute error. The authors concluded that artificial
neural networks can predict compressive strength of concrete with 91 to 98 %
accuracy.
EXPERIMENTAL INVESTIGATION
M30 grade of concrete were used for the present investigation. Mix design was done
based on IS 10262 2009 (17). The concrete mix proportion 1:1.73:3.2 with w/c 0.45
considered in this study. Twelve HPC mixes were prepared for this test by volumetric
method. The conventional concrete mix CC and Combinations of HPC mixes (S1S11) as given in Table.1. The volume of water is 172.8 lit/m3 and Coarse aggregate
(CA) is 1220 kg/m3 were kept constant while the volume of cement, sand and
Superplasticizer (SP) were varied for all the mixes. The mix Combinations and mix
proportions are given in table 1 & 2. The selected 4 HPC mixes are having the
maximum compressive strength at 28 days including CC & S3, S7, S10 and S11.
http://www.iaeme.com/IJCIET/index.asp 304
editor@iaeme.com
S.No
Mix designation
Combinations
CC
(C+S+CA)
S1
(C+FA20%)+S+CA)
S2
(C+SF10%)+S+CA
S3
(C+MK10%)+S+CA
S4
C+(S+BA20%)+CA
S5
(C+FA20%)+(S+BA20%)+CA)
S6
(C+SF10%)+(S+BA20%)+CA
S7
(C+MK10%)+(S+BA20%)+CA
S8
(C+FA20%+SF10%)+(S+BA20%)+CA
10
S9
(C+FA20%+MK10%)+(S+BA20%)+CA
11
S10
(C+SF10%+MK10%)+(S+BA20%)+CA
12
S11
(C+FA20%+SF10%+MK10%)+(S+BA20%)+CA
http://www.iaeme.com/IJCIET/index.asp 305
editor@iaeme.com
S.N
o
Mix
Cement(C
) (kg/m3)
Fly ash
(FA)
(kg/m3
)
1
2
3
4
5
6
7
8
9
10
11
12
CC
S1
S2
S3
S4
S5
S6
S7
S8
S9
S10
S11
384
307.2
345.6
345.6
384
307.2
345.6
345.6
268.8
268.8
307.2
230.4
0
76.8
0
0
0
76.8
0
0
76.8
76.8
0
76.8
A
S
C
F
A
Silica
fume
(SF)
(kg/m3
)
0
0
38.4
0
0
0
38.4
0
38.4
0
38.4
38.4
Metakao
lin (MK)
(kg/m3)
Fine
aggregat
e (S)
(kg/m3)
Bottom
ash
(BA)
(kg/m3)
SP
(lit/m3)
Slum
p
(mm)
0
0
0
38.4
0
0
0
38.4
0
38.4
38.4
38.4
665
623
644
649
508
476
500
508
461
467
492
463
0
0
0
0
133
133
133
133
133
133
133
133
2.49
3.37
3.97
3.45
3.84
3.99
4.83
4.49
4.03
3.49
4.60
3.80
55
57
55
56
58
55
52
57
59
58
57
58
N
1
N
2
N
3
S
F
N
4
M
KF
N
5
N
6
N
7
N
8
N
9
B
A
W
S
C
A
Experimental
Compressive
strength for
28
days(N/mm2)
36.5
34.54
37.03
41.34
35.33
37.21
39.31
42.79
37.25
39.22
44.69
40.0
S
P
N10
Input
layer
1. Hidden
layer
fc
http://www.iaeme.com/IJCIET/index.asp 306
editor@iaeme.com
A
S
N1
N11
N2
N12
F
A
N3
N13
S
F
N4
N14
M
K
N5
N15
B
A
N6
N16
N7
N17
N8
N18
C
A
N9
N19
S
P
N10
N20
Input
layer
1.Hidden
layer
2.Hidden
layer
fc
3.Output
layer
http://www.iaeme.com/IJCIET/index.asp 307
editor@iaeme.com
Datas
Input Variables
Age of Specimen (day)
Cement (Kg/m3)
Silica fume (Kg/m3)
Metakaolin (Kg/m3)
Fly ash (Kg/m3)
Bottom ash (Kg/m3)
Sand (Kg/m3)
Coarse Aggregate (Kg/m3)
Superplasticizer (l/m3)
Output variable
Compressive strength
28
230.4
0
0
0
0
461
1220
2.49
120
384
38.4
38.4
76.8
133
665
1220
4.6
36.31
47.38
Table 4 Experimental results with Predicted results from models for 28 days
Compressive strength (N/mm2) 28 days
Mix Designation
Experimental result
ANN-I
ANN-II
% Error
CC-1
CC-2
CC-3
S3-1
S3-2
S3-3
S7-1
S7-2
S7-3
S10-1
S10-2
S10-3
S11-1
S11-2
S11-3
36.310
36.890
36.620
41.020
41.330
41.690
42.040
42.310
44.040
43.960
44.090
46.040
38.980
39.980
41.020
36.823
36.823
36.823
41.493
41.493
41.493
43.225
43.225
43.225
44.124
44.124
44.124
39.980
39.980
39.980
36.452
36.452
36.452
41.302
41.302
41.302
42.951
42.951
42.951
43.950
43.950
43.950
39.951
39.951
39.951
-0.513
0.067
-0.203
-0.473
-0.163
0.197
-1.185
-0.915
0.815
-0.164
-0.034
1.916
-1.000
0.000
1.040
http://www.iaeme.com/IJCIET/index.asp 308
editor@iaeme.com
Mix
Designation
CC-1
CC-2
CC-3
S3-1
S3-2
S3-3
S7-1
S7-2
S7-3
S10-1
S10-2
S10-3
S11-1
S11-2
S11-3
% Error
-0.424
-0.124
-0.474
-0.441
-1.441
0.559
0.029
-0.871
0.429
1.315
-0.785
-0.485
-0.254
-1.754
0.246
Table 6 Experimental results with Predicted results from models for 90 days
Compressive strength (N/mm2) 90 days
Mix Designation
CC-1
CC-2
CC-3
S3-1
S3-2
S3-3
S7-1
S7-2
S7-3
S10-1
S10-2
S10-3
S11-1
S11-2
S11-3
Experimental
result
37.110
37.280
37.200
43.200
42.500
44.300
44.600
44.100
45.800
46.540
45.820
45.970
41.520
41.360
42.380
ANN-I
37.750
37.750
37.750
43.662
43.662
43.662
44.652
44.652
44.652
46.639
46.639
46.639
42.446
42.446
42.446
http://www.iaeme.com/IJCIET/index.asp 309
ANN-II
37.242
37.242
37.242
43.200
43.200
43.200
44.856
44.856
44.856
46.160
46.160
46.160
41.761
41.761
41.761
% Error
-0.640
-0.470
-0.550
-0.462
-1.162
0.638
-0.256
-0.756
0.944
-0.099
-0.819
-0.669
-0.926
-1.086
-0.066
editor@iaeme.com
% Error
-0.078
0.052
-0.028
0.251
-1.829
-0.159
0.168
-0.992
0.018
0.730
-0.260
-0.400
-0.235
-0.605
0.235
http://www.iaeme.com/IJCIET/index.asp 310
editor@iaeme.com
http://www.iaeme.com/IJCIET/index.asp 311
editor@iaeme.com
ANN-II
ANN-I
ANN-II
ANN-I
ANN-II
120 days
ANN-I
90 days
ANN-II
56 days
ANN-I
28 days
Statistical
parameter
(Training
set)
RMSE
0.787
0.790
0.764
0.763
0.716
0.525
0.591
0.626
MSE
0.619
0.625
0.584
0.583
0.513
0.276
0.349
0.392
MAPE
(%)
1.384
1.345
1.370
1.499
1.453
0.939
1.019
0.916
R2
0.999
0.996
0.999
0.999
0.999
0.9997
0.9998
0.999
CONCLUSIONS
In this Study, using these beneficial properties of artificial neural networks in order to
predict the 28, 56, 90 and 120 days compressive strength values of concrete
containing Industrial Byproducts with attempting experiments were developed two
different architectures namely ANN-I and ANN-II. In two models developed on ANN
method, a multilayer feed forward neural network in a back propagation algorithm
were used. In ANN-I model, one hidden layer were selected. In the hidden layers 10
neurons were determined. In ANN-II model, two hidden layers were selected. In the
first hidden layers 10 neurons and in the second hidden layer 10 neurons were
determined. The models were trained with input and output data. The compressive
strength values predicted from training for ANN-I & ANN-II models were very close
to the experimental results. Furthermore, according to the compressive strength results
predicted by using ANN-I and ANN-II models, the results of ANN-II model are
closer to the experimental results. RMSE, MSE, R2 and MAPE statistical values that
are calculated for comparing experimental results with ANN-I and ANN-II model
results have shown this situation. As a result, compressive strength values of
http://www.iaeme.com/IJCIET/index.asp 312
editor@iaeme.com
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
http://www.iaeme.com/IJCIET/index.asp 313
editor@iaeme.com
[20]
[21]
[22]
P.J.Patel, Mukesh A. Patel and Dr. H.S. Patel. Effect of Coarse Aggregate
Characteristics on Strength Properties of High Performance Concrete Using
Mineral and Chemical Admixtures, International Journal of Civil
Engineering and Technology, 4(2), 2013, pp. 8995.
Topu IB, Sardemir M. Prediction of compressive strength of concrete
containing fly ash using artificial neural network and fuzzy logic, Comp Mater
Sci 2008; 41(3): 30511.
[23]
http://www.iaeme.com/IJCIET/index.asp 314
editor@iaeme.com