Proceedings of the International Conference

INTERNATIONAL JOURNAL on Emerging

OFTrends
CIVILin Engineering and ManagementAND
ENGINEERING (ICETEM14)
30 31, December 2014, Ernakulam, India
TECHNOLOGY (IJCIET)

ISSN 0976 6308 (Print) IJCIET

ISSN 0976 6316(Online)
Volume 5, Issue 12, December (2014), pp. 239-247
IAEME: www.iaeme.com/Ijciet.asp
Journal Impact Factor (2014): 7.9290 (Calculated by GISI)
IAEME
www.jifactor.com

PREDICTING THE STRENGTH OF SELF-COMPACTING

SELF-CURING CONCRETE USING ARTIFICIAL NEURAL
NETWORK

Mohanraj A1, Manoj Prabhakar S2, Rajendran M3

1,2,3
Department of Civil Engineering, Bannariamman Institute of Technology, Erode,
Tamil Nadu, India.

ABSTRACT

In recent years, self-compacting Self curing concrete (SCSCC) has gained wide use for concreting in congested
reinforced structures with difficult casting conditions. For such applications, the fresh concrete must possess high fluidity
and good cohesiveness. The use of fine materials such as Fly ash can ensure the required concrete properties. This
research work focuses on artificial neural networks (ANNs) for evaluating compressive strength of self compacting
concrete (SCC) at 28 days. To evaluate the strength seven input parameters that are the weight of cement, coarse and fine
aggregate, fly ash and three chemical admixtures were identified. Experimental works by casting 17 different trails of
cubes size 150mm was carried out and allowed for curing. All the cubes were tested and the compressive strength was
determined after 28 days of self curing. The experimental results of the tests carried out were used in training Artificial
Neural Network (ANN) model from which an optimum mix of SCSCC was obtained. It is concluded that 2% of super
plasticizers, 0.5% of Viscosity Modifying Agent and Poly-ethylene Glycol is optimum to use in SCSCC mix of M40
concrete. Exceeding which brings down the rate of setting time and strength.
The size of 1000x150x220 mm beam was also casted for the optimum mix and is tested. The experimental result
is compared with ANN result, which suits with minor negligible errors.

Keywords: Artificial Neural Network, Chemical Admixtures, Compressive Strength, Self-Compacting Self-Curing
Concrete

1. INTRODUCTION

There is no standard method for SCC mix design and many academic institutions, ready-mixed industries; precast
and contracting companies have developed their own mix proportioning methods. So in doing trial and error technique
requires a long time and needs more concrete material. To overcome the problems, need a tool for evaluating concrete mix
composition of SCC. This study uses Artificial Neural Network (ANN) as a tool to evaluate the workability test and the
compressive strength of SCC at 28 days. Good curing is not always practical in many cases due to the non-availability of
good quality water. The current trend is incorporating self-curing agents in Self Compacting Concrete. The current trend is
incorporating self-curing agents in Self Compacting Concrete. So, a study may be conducted on self-compacting self-
curing concrete using ANN tool. Abdul Raheman 2013 [1] concludes that, this study of Artificial Neural Network model
will provide an efficient and rapid means of obtaining optimal solutions to predict the optimum mix proportions for
specified strength and workability for sustainable SCC. C. Selvamony et al 2010 [2] Investigated on Self-compacted self-
curing concrete using Lime stone powder and clinkers indicated the use of silica fume in concrete significantly increased

239
 Proceedings of the International Conference on Emerging Trends in Engineering and Management (ICETEM14)
30 31, December 2014, Ernakulam, India

the dosage of Super plasticizer. Silica fume can better reducing effect on total water absorption while quarry dust and lime
powder will not have the same effect at 28 day.

2. MATERIALS AND ITS PROPERTIES

The cement used in this specimen is ordinary Portland cement of 53 grade and the specific gravity is 3.14. The
initial and final setting times were found as 30 and 356 min respectively. The size of coarse aggregate used was 12.5
mm. The specific gravity of it is 2.73. Fine aggregate used was river sand passing through IS sieve 4.75 mm. As mineral
admixture Fly ash was used in this work. The cement is replaced with 5%, 10%, 15% and 20% by weight of cement. To
improve the workability of concrete Conplast SP430 (2% by weight of cementitious material had been used). To make
the concrete as more workable with self compacting character, chemical admixtures of Viscosity Modifying Agent
(VMA) Glenium Stream 2 of 0.5% by weight of cementitious material was used. Poly Ethylene glycol (PEG) was used
for internal curing (0.5% by weight of cement.) Mix designs of Self compacting concretes were developed by means of
trail mixes based on the guidance given in EFNARC. Standard 150mm cube was produced. Typical Trail mix for Self-
compacting Concrete is shown in Table 1.

3. WORKABILITY OF SCC

Filling ability, passing ability and segregation resistance are the requirements for judging the workability criteria
of fresh SCC. These requirements are to be fulfilled at the time of placing of concrete. Typical Trail mix and workability
property for Self-compacting Concrete are shown in TABLE 1.

Table 1 Trail mixes and Workability tests for SCSCC concrete

Admixtures (%) Workability
Cement Flyash C.A. F.A.
Trails
(Kg) (Kg) (Kg) (Kg) SP Slump V-funnel L-Box
VMA PEG U-Box
337 (mm) (sec) H2/H1
1. 1.18 0.5 2.025 3.024 1 1 0.5 670 6 -- --
2. 1.18 0.5 2.025 3.024 1.2 1.2 0.5 710 7 -- --

3. 1.18 0.5 2.025 3.024 1.5 1.4 0.5 780 10 0.82 24

4. 1.18 0.5 2.025 3.024 1.78 1.6 0.5 500 10 0.91 28

5. 1.28 0.45 2.025 3.024 1.65 1.5 0.5 630 10 0.99 22

6. 1.28 0.45 2.025 3.024 1.5 1.45 0.5 575 11 1 25

7. 1.35 0.34 2.36 3.024 1.65 1.5 0.5 630 9 0.9 23

8. 1.35 0.34 2.36 2.7 1.75 1.6 0.5 670 8 0.9 20

9. 1.50 0.17 2.565 3.105 0.5 2 0.5 700 9 0.88 30

10. 1.42 0.25 2.565 3.105 0.5 2 0.5 670 10 0.82 28

11. 1.34 0.33 2.565 3.105 0.5 2 0.5 660 12 0.76 27

12. 1.59 0.083 2.565 3.105 1 2 1 750 7 0.82 15

13. 1.50 0.17 2.565 3.105 1 2 1 680 6 0.86 22

14. 1.13 0.63 2.65 2.9 0.5 2 0.5 630 12 0.89 28

15. 1.64 0.45 1.90 3.3 0.5 2 0.5 770 8 0.8 21

16. 1.87 0.21 3 3 0.5 2 0.5 790 11 0.97 20

17. 1.68 0.43 3 3 0.5 2 0.5 680 9 0.89 25

18. 1.59 0.083 2.56 3.105 0.5 2 0.5 725 10 0.95 26

240
Proceedings of the International Conference on Emerging Trends in Engineering and Management (ICETEM14)
30 31, December 2014, Ernakulam, India

Figure 1 Slump value for various trails Figure 2 V-Funnel tests for various trails

Figure 3 L-Box tests for various trails Figure 4 U-Box tests for various trails

Figure 5 Compressive strength results for various trails

241
 Proceedings of the International Conference on Emerging Trends in Engineering and Management (ICETEM14)
30 31, December 2014, Ernakulam, India

4. ANN FOR EVALUATING DESIGN MIX

The results of experimental data include 17 data sets, which was collected experimentally. The compressive
strength of SCC at 28 days was determined by the compressive strength machines. The data were randomly divided into
a training phase (6 data sets), testing phase (6 data set), and validation phase (5 data set). For this compressive strength of
SCC at 28 days modelling problem the obvious inputs are the component contents of concrete, including cement, coarse
aggregate, fine aggregate, fly ash, and chemical admixture. That is, the ANN investigated in the developing has seven
units in the input layer and one unit in the output layer. The values of ANN parameters considered in this approach are as
follows: number of hidden layers = 1; number of hidden neurons = 1, 2, 3, and 20; learning rate = 0.001, 0.05, 0.01,
0.1, 0.25 and 0.5; momentum factor = 0.01; and learning cycles (epochs) = 1000 which each cycle covers the entire
database available for training. Training function used is TRAINLM and adaption learning function is LEARNGDM.
Transfer function is TANSIG. The input and output (strength) details are feed in the ANN Tool box. It is trained and then
is simulated. The process is given in fig 6 to 10.

Figure 6 Input and Output

Figure 7 Neural Network creation Figure 8 Neural Network Training

242
 Proceedings of the International Conference on Emerging Trends in Engineering and Management (ICETEM14)
30 31, December 2014, Ernakulam, India

Figure 9 ANN Input Figure 10 ANN Output showing strength

5. BEAM DETAILS

The Beam is designed as a singly reinforced beam. The beam is designed as per Code IS 456: 2000. Thus the
beam details are given below:
Beam Size = 1000 X 150 X 220 mm,
Cover = 25 mm
Main rod = 3 nos of 10mm rods
Hanger rods = 2 nos of 8mm rods
Stirrups = 2 legged stirrups of 8mm rods @ 200 mm
fy = Fe 415;
fck = 35Mpa

5.1. A. Load, deflection and crack measurement of RCC beam

The beam as stated in V is tested as per the codal procedure. The two point load is applied at the distance of L/3.
The clear span of beam is adopted as 800mm & the concentric loading is given during the testing of the beam. Deflecto-
meter is said to measure the deflection value of the beam. The test results are given in TABLE 3. Testing of beam is
given in fig 11.

Figure 11 Testing of Beam Figure 12 Tested Beam

243
 Proceedings of the International Conference on Emerging Trends in Engineering and Management (ICETEM14)
30 31, December 2014, Ernakulam, India

Table 3 Load & Deflection Test Results

Loadings Deflection
S.No Remark
(KN) (mm)
1. 0 0
2. 10 0.02
3. 20 0.25
4. 30 0.46
5. 40 0.66
6. 50 0.87
7. 60 1.11 First Crack
8. 70 1.34
9. 80 1.65
10. 90 1.98
11. 100 2.21
12. 110 2.53
13. 120 2.82
14. 130 3.15
15. 140 3.50
16. 150 4.00
Ultimate crack load &
17. 155 4.52
Maximum deflection

5.2. Consistency between ANN Modelling and Experiments

The trained ANN models can be used to simulate the effects of some factors on the strength, and the obtained
functional relations between strength. The following simulation results obtained are as shown in Fig. 10. Fig 6
respectively shows the amount of Cement, Fly ash, coarse aggregates, fine aggregate and percentage of Sp, VMA and
PEG.

5.3. SEM and TEM Analysis

From Fig 13 and Fig 14, high strength (HS) self-compacting concrete samples have shown smaller physical
interface micro-cracks than low strength (LS) self compacting concretes. Fig 15 shows the TEM Images in which the
particle size and its distributions are clearly pictured

Figure 13 SEM Analysis of High Strength concrete Figure 14 SEM Analysis of Low Strength concrete

244
 Proceedings of the International Conference on Emerging Trends in Engineering and Management (ICETEM14)
30 31, December 2014, Ernakulam, India

Figure 15 TEM Analysis

6. DISCUSSION ON RESULTS

From the results we can able to retrieve the following conclusions:

6.1. Discussion on Chemical Admixtures

Out of 18 Trails trail 12 and trail 13 gives the bad strength. This is because of increasing the VMA and PEG to
1%. As it is raised the settling time also increases and thereby the hydration process slows down giving raise to decrease
in compressive strength.
SEM and TEM analysis were performed on the concrete specimen having High and Low strength i.e., on trail
12 and Trail 18. The beam was casted for the mix of trail 18 which gave the satisfactory load carrying capacity and
deflection.

6.2. Discussion on Workability

All the trails satisfy the recommendations of workability tests as per EFNARC Specifications.
Trail 1, trail 8 and trail 10 have satisfactory slump results which gives 7.5% lower than slump value of trail 18. Trail 1
and trail 13 is about 40 % lower in V-Funnel tests comparing with trail 18.
The slump value and V-Funnel results of trail 2 gives about 10% and 36.3 % lower respectively to the values of trail 18.
For the trail 3 the slump value is about 7% higher than the value of trail 18. A trail 3, 4 and 5 has the same V-Funnel
values as that of the values of trail 18. For L-Box tests the trails 3, 10 and 12 is 13.5% lower than value of trail 18. The
value of U-Box test for trail 3 is 7.7% lower than trail 18.
Trail 4 and trail 6 does not satisfy the slump flow value as specified in EFNARC Specifications. But those satisfy the
other workability test results. V-Funnel result of trail 6 and trail 16 is 9% lower than trail 18. The L-Box test value of
trail 4 is about 4.2% lower and trail 6 is 5% higher to the test results of trail 18. The U-Box value of trail 4, trail 10 and
trail 14 is 7% higher than the value of trail 18.
The trail 5, trail 7 and trail 14 have the slump value 13% lower than the slump value of trail 18. About 4% higher L-Box
test result than trail 18 and Trails 5, 13 has U-Box test results which are about 15.3% lower than the trail 18.
For the trail 6 and trail 17 have the same U-Box values and about 4% lower than the value of trail 18.
The trail 7, trail 9 and trail 17 has the same V-Funnel tests and about 10% lower than trail 18 and L-Box tests of trail 7
and trail 8 is about 5.2% lower than trail 18. The U-Box test result of trail 7 is about 11.5% lower than trail 18.
Trails 8 and Trail 15 has 20% lower slump value as compared with the slump value of trail 18. Similarly trail 8 and trail
16 has U-Box value of about 23% lower than trail 18.
Trail 9 has slump value, L-Box and U-Box tests results as 3.4% lower, 7.3% lower and 13.3% higher respectively to the
workability results of trail 18.
Trail 11 has slump value, V-Funnel, L-Box and U-Box tests results as 8.9% lower, 16.6% higher, 20% lower and 3.7%
higher respectively than the workability results of trail 18.
The Slump value and U-Box test results of Trail 12 is about 3.3% higher and 42.3% lower than the trail 18.
Trail 13 and Trail 17 have slump value of about 6.2% lower and L-Box value of trail 13 is about 9.4% lower than trail
18.
Trail 14 and trail 17 have L-Box test value of about 6.3% lower than trail 18.
Trail 15 has slump value, L-Box and U-Box tests results as 5.8% higher, 15.8% lower and 19.2% lower respectively than
the workability results of trail 18.
The slump value and L-Box value of trail 16 is about 8.2% and 2% respectively higher than the test results of trail 18.

245
 Proceedings of the International Conference on Emerging Trends in Engineering and Management (ICETEM14)
30 31, December 2014, Ernakulam, India

6.3. Discussion on Compressive strength

The compressive strength of trail 1, trail 4 and trail 13 is about 50% lower than the target strength achieved in
trail 18.
The compressive strength of trail 2 and trail 16 is about 33% lower than strength achieved in trail 18. Comparing the
compressive strength, trail 3 and trail 14 is about 39.5% lower than the strength achieved in trail 18.
Trails 5, 7, 11 and 17 has lower the compressive strength of about 27% compared to the compressive strength achieved
in trail 18.
Trail 6 and trail 10 has compressive strength of about 22.5% lower than the strength achieved in trail 18. Trails 8 and
trail 15 has compressive strength of about 12.5% lower than strength achieved in trail 18.
The compressive strength of trail 9 is nearly to the target compressive strength achieved in trail 18 which is about 8.7%
lower. The lower compressive strength achieved in trail 12 is about 68.75% compared to trail 18.
Trail 18 is our expected result which achieved the compressive strength of about 35N/mm2.

6.4. Discussion on ANN

The rules obtained by the ANN models are consistent with those by laboratory work. In addition, these
reasonable results indicate that the trained ANN models exhibit good performance. Conclusions drawn from the output
conform to some rules on concrete mix proportioning. To some extent, the ANN models prove reasonable and feasible.
The compressive strength and flexural strength of analytical results satisfactorily matches with the experiment result.

6.5. Discussion on SEM and TEM Images

From Fig 13 and Fig 14, high strength (HS) self-compacting concrete samples have shown smaller physical
interface micro-cracks than low strength (LS) self compacting concretes, which meant that HS SCC had better bonds
than LS concrete, between aggregate and cement. This explains the increase in tensile and compressive strength for the
HS self-compacting concrete compared to the LS concrete. As mentioned earlier, a better bonding due to the smaller
physical interfaces in HS SCC increased the percentages of fractured aggregate compared to LS concrete.
Fig 15 shows the TEM Images in which the particle size and its distributions are clearly pictured. The dark
coloured particle is fly ash is closely arranged. The streams of PEG are also seen which is responsible for internal curing.

7. CONCLUSION

i. From the results, Replacement of fly ash as admixture satisfies all the tests such as Slump flow test, V Funnel, U
Box method and L Box method.
ii. In this research, ANNs model for evaluating the compressive strength of SCC was developed. The study suggests
that the use of ANNs has several significant advantages over other conventional methods. The results of
compressive strength of SCC at 28 days obtained from the developed computer program were compared with
results from experimental studies. The comparisons of results indicate good agreements.
iii. Optimum percentage of chemical admixtures such as SP, VMA and PEG was found as 2%, 0.5% and 0.5%
respectively exceeding which gives low strength and higher setting time.
iv. The optimum dosage of PEG400 for maximum strengths (compressive and flexural strength) was found to be
0.5% for M40 grade of concrete.
v. The deflection of beam is about 4.52 mm which carries the load to about 155 KN.
vi. Self-curing concrete is the answer to many problems faced due to lack of proper curing.
vii. From SEM and TEM analysis it is clear that the high strength of concrete is due to better bonding between the
concrete ingredients.
viii. ANN models attain good prediction accuracy. Some effects of concrete compositions on strength are in
accordance with the rules of mix proportioning. Consequently, the application of ANN models to concrete
strength prediction is practical and has a good future.
ix. This study of Artificial Neural Network model will provide an efficient and rapid means of obtaining optimal
solutions to predict the optimum mix proportions for specified strength and workability for sustainable SCC. The
application of ANN in the field of SCC mix design is very appropriate in order to preserve and disseminate
valuable experience and innovation efficiently at reasonable cost.

246
Proceedings of the International Conference on Emerging Trends in Engineering and Management (ICETEM14)
30 31, December 2014, Ernakulam, India

REFERENCES

[1] Abdul Raheman and Modani P. O, Prediction of Properties of Self Compacting Concrete Using Artificial Neural
Network, International Journal of Engineering Research and Applications, 3(4), 2013, 333-339.
[2] Ravikumar M S, Selvamony C, Kannan S U and Basil Gnanappa S, Behaviour of self compacted self curing kiln
ash concrete with various admixtures, ARPN Journal of Engineering and Applied Sciences, 4(8), 2009, 24-30
[3] Moosa Mazloom , Application of Neural Networks for Predicting the Workability of Self-Compacting Concrete,
Journal of Scientific Research &Reports, 2 (1), 2013, 429-442.
[4] Ravikumar M S, Selvamony C, Kannan S U and Basil Gnanappa S, Investigation on self compacted self curing
concrete using Lime stone, powder and Clinkers, ARPN Journal of Engineering and Applied Sciences,
5(3),2010, 1-6.
[5] Bai J, Wild S, Ware JA and Sabir BB, Using neural networks to predict workability of concrete incorporating
metakaolin and fly ash, Adv Eng Software,34(12), 2003, 6639.
[6] Farhad Aslani and Shami Nejadi , Shrinkage behaviour of self-compacting concrete, Journal of Zhejiang
University - Science, 13(6),2012, 407-419.
[7] Jadid M N and Fairbairn D R (1996) Neural - network applications in predicting moment curvature parameters
from experimental data, Eng Appl Artif Intell, 9(3),1996, 309-319.
[8] Jagannadha Kumar M V, Srikanth M and Jagannadha Rao K, Strength characteristics of self-curing concrete,
IJRET 1(1), 2012, 51-57.
[9] Maghsoudi A.A, Sh. Mohamadpour and M. Maghsoudi, Mix design and mechanical properties of self
compacting light weight concrete, International Journal of Civil Engineering, 5(3), 2011, 244-251
[10] Rasa E et al, Predicting Density and Compressive Strength of Concrete Cement paste Containing Silica Fume
Using Artificial Neural Networks, 16(1), 2009, 33-42.
[11] Wang J Z, Wen X M, Huang Y X and Jiang, Dynamics and strength analysis of a MK- II screen, J China Coal
Soc, 24(2), 1999, 184-188
[12] Wenzhong Zhu., and Peter J.M.Bartos., Permeation properties of Self Compacting Concrete, Cement and
Concrete research, 33(2),2005, 921-926.
[13] Yang, K. H. and Ashour, A. F, Modelling of Reinforced Concrete Deep Beams, Magazine of Concrete
Research, 60(6), 2008, 441-454.
[14] Subramanian, S. and D. Chattopadhyay, Experiments for mix proportioning of self-compacting concrete, The
Indian Concrete Journal, 2002, 13-20.
[15] Sudarsana Rao H and Ramesh Babu B, Hybrid Neural Network model for the design of beam subjected to
bending and shear, Sadhana , 32(5), 2007, 577-586
[16] Okamura Hajime and Ouchi Masahiro, Self Compacting Concrete, Journal of advanced concrete technology,
1(1), 2003, 5 15.
[17] Rasa E et al, Predicting Density and Compressive Strength of Concrete Cement paste Containing Silica Fume
Using Artificial Neural Networks, 16(1), 2009, 33-42.
[18] EFNARC, Specification and guidelines for self-compacting concrete, European Federation of Producers and
Applicators of Specialist Products for Structures, 2005.
[20] Santosh Patil and Shriniwas Valunjkar, Forecasting of Daily Runoff Using Artificial Neural Networks
International Journal of Civil Engineering & Technology (IJCIET), Volume 5, Issue 1, 2014, pp. 13 - 20, ISSN
Print: 0976 6308, ISSN Online: 0976 6316.
[21] Bhavneet Singh, Bleeding in Concrete International Journal of Civil Engineering & Technology (IJCIET),
Volume 4, Issue 2, 2013, pp. 247 - 249, ISSN Print: 0976 6308, ISSN Online: 0976 6316.

247