See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/250072829

Simulation of concrete slump using neural networks

Article  in  Construction Materials · January 2009

DOI: 10.1680/coma.2009.162.1.11

CITATIONS READS

12 593

1 author:

I-Cheng Yeh
Tamkang University
42 PUBLICATIONS   1,257 CITATIONS   

SEE PROFILE

All content following this page was uploaded by I-Cheng Yeh on 04 July 2018.

The user has requested enhancement of the downloaded file.

Simulation of concrete slump using neural networks

by I-Cheng Yeh
Professor, Department of Information Management, Chung-Hua University,
Hsin Chu, Taiwan 30067, R.O.C.
TEL: 886-3-5186511
FAX: 886-3-5186546
Email: icyeh@chu.edu.tw

Abstract
In this study, an artificial neural network-based modeling system was established to

explore the feasibility of predicting the slump of concrete. Computational simulation

of concrete slump was performed using the trained neural network. The variation in

concrete slump was achieved by varying combinations of factors like the water/binder

ratio (w/b), superplasticizer-binder ratio (SP/B), and water content. From the water

content-slump curves generated using the trained neural networks developed in this

study, two sets of curves have been produced to explore the effects of w/b and SP/B.

It has found that (1) the use of neural network for the modeling of concrete slump

looks very promising, (2) although the water content and SP/B ratio were kept

constant, a change in w/b ratio had a distinct effect on the consistence properties, (3) a

certain saturation level for the SP exists, above which only a small effect of further

dispersion is obtained, and (4) a certain saturation level for the water content exists,

above which only a small effect, even negative effect, on slump is obtained.

Key Words: superplasticizer, slump, consistence, artificial neural networks.

1
1. Introduction

Over the last decade, the High-Performance Concrete (HPC) technology has been

the important subject in concrete research since HPC has numerous advantages in

practical applications. A commentary to the definition states that a HPC is one in

which certain characteristics are developed for a particular application and

environment. Examples of characteristics that may be considered critical for an

application are: ease of placement, compaction without segregation, early age strength,

volume stability, long life in severe environments, etc. Among them, the fine

consistence (workability) is one of the most representative characteristics. For this

reason, the use of such concrete is spreading worldwide quickly [1].

The know-how necessary to produce HPC consists of a specific knowledge of the

properties of the ingredients and of their interactions. The crucial outcome of the

know-how is an extremely low water-binder ratio (w/b) coupled with a satisfactory

consistence at the time of compaction [2].

The superplasticizer (SP) deflocculates the cement particles and thus fluidifies the

mixture so that a very low water content is sufficient for an adequate consistence. The

use of superplasticizers is essential to ensuring that adequate consistence is obtained

at a low ratio of w/b without an excessive cementitious material content. The

effectiveness of admixture-cement combinations varies considerably [3].

Difficult and poor consistence properties are sometimes referred to as problems for

proper utilization of high-performance concrete, but very little information about

consistence properties of such concrete is available in the literature. A literature search

was conducted. Rather than compiling an exhaustive annotated bibliography of the

available literature, some important publications were reviewed, and they are listed in

the References at the end of the paper [4-9]. Lacking such information, optimization

2
of a concrete mix proportion is rarely attained.

Because of high complexity of these relations, conventional regression analysis

could be not sufficient to build an accurate model. The growing interest in neural

networks among researchers is due to its potential in modeling nonlinear multivariate

interrelationships. Some recent applications of neural networks in civil engineering

materials include those references [10-23]. However, little research has been done on

modeling consistence of concrete using neural networks.

In this study, an artificial neural network-based modeling system was established to

explore the feasibility of using a neural network in predicting the slump of concrete.

In addition, the results obtained by neural network are compared with the

experimental values and with those determined from statistical modeling techniques.

Computational simulation of concrete slump was performed using the trained neural

network. The variation in concrete slump was achieved by varying combinations of

factors like the water-binder ratio (w/b), SP-binder ratio (SP/B), and water content.

From the water content-slump curves generated using the trained neural network

developed in this study, two sets of curves have been produced to explore the effects

of w/b and SP/B. Finally, some discussions and conclusions were deduced from these

curves.

2. Introduction of Neural Networks

Neural network simulations represent attempts to mirror biological methods of

information processing. The fundamental concept is that of a neuron, a biological cell

that receives electrical inputs from one or many sources and processes those inputs to

generate a unique output. The output may, in turn, be passed on to other neurons [10].

Neural network-based material models differ in some fundamental ways from the

3
traditional mathematical models [12]. Training of the neural network is essentially

carried out through the presentation of a series of example patterns of associated input

and target output values. Each hidden and output neuron processes its inputs by

multiplying each input by its weight, summing the product, and then passing the sum

through a nonlinear transfer function to produce a result. The S-shaped sigmoid curve

is commonly used as a transfer function.

The neural network “learns” by modifying the weights of the neurons in response

to the errors between the actual output values and the target output values. This is

carried out through the gradient descent of the sum of squares of the errors for all the

training patterns. The changes in weights are in proportion to the negative of the

derivative of the error term. One pass through the set of training patterns along with

the updating of the weights is called a cycle. Training is carried out by repeatedly

presenting the entire set of training patterns until the average sum squared error over

all the training patterns is minimized and within the tolerance specified for the

problem [24].

The back-propagation training algorithm is an iterative gradient algorithm designed

to minimize the mean square error between the actual output of a multilayer

feed-forward neural network and the desired output. It requires continuous

differentiable nonlinearities. The following sigmoid logistic nonlinearity is usually

used [25, 26]:

1
y j = f (α j ) = −α j
(1)
1+ e

where α j the net value of sum of the product of weights and inputs, i.e.,

α j = ∑ wij ⋅ xi − θ j (2)
i

In the formulas, y j is the output of the node, xi are the inputs of the node, wij are

4
the weights (connection strengths) between nodes, and θ j is internal offsets in the

output node.

The basic process of back-propagation training algorithm are summarized as

follows [25, 26]

Step 1. Initialize Weights and Offsets

Set all weights and node offsets to small random values.

Step 2. Present Input and Desired Outputs

Present a continuous valued input vector x 0 , x1 ,...x N −1 and specify the desired

outputs d 0 , d1 ,...d M −1 .

Step 3. Calculate Actual Outputs

Use the sigmoid nonlinearity to calculate outputs y 0 , y1 ,... y M −1 .

Step 4. Adapt Weights

Use a recursive algorithm starting at the output nodes and working back to the

first hidden layer. Adjust weights by

wij (t + 1) = wij (t ) + ηδ j x'i (3)

where wij (t ) is the weight from hidden node i or from an input to node j at time t, x' i

is either the output of node i or is an input, η is a gain term, and δ j is an error term

for node j.

(1) If node j is an output node, then

δ j = y j (1 − y j )(d j − y j ) (4)

5
 where δ j is the desired output of node j and y j is the actual output.

(2) If node j is an internal hidden node, then

δ j = x' j (1 − x' j )∑ δ k w jk (5)

k

where k is overall nodes in the layer above node j.

Internal node thresholds are adapted in a similar manner by assuming they are

connection weights on links from auxiliary constant-valued inputs.

Step 5. Repeat by Going to Step 2

Repeat by going to Step 2 until convergence is obtained.

The detail mathematical concepts of the back propagation algorithm are found in

the literature [25, 26].

During training the network performance is monitored by Root-Mean-Square

(RMS) error to achieve a better understanding of the network performance. Once

trained, the values for the input parameters for the project are presented to the

network. Then the network calculates the node outputs using the existing weight

values and thresholds developed in the training process [27]. Such a trained neural

network not only would be able to reproduce the experiment results it was trained on,

but through its generalization capability it should be able to approximate the results of

other experiments. The degree of accuracy in this generalization depends on how

comprehensive the training set is [12].

3. Modeling Concrete Slump Using Neural Networks

6
 The experimental data include 103 mixtures, which are taken from the tests carried

out by Yeh and Chen [28]. The data set is now available on

www.chu.edu.tw/~icyeh/slump_103_dat. Materials in the reference are as follows:

1. Cement is Portland cement (ASTM Type I).

2. Fly ash is produced by power plant.

3. Water quenched blast-furnace slag powder is supplied by steel plant.

4. Water is ordinary tap water.

5. Chemical admixture is the superplasticizer that meets ASTM C494 type G.

6. Coarse aggregate is crushed natural rock with 10 mm maximum size.

7. Fine aggregate is washed natural river sand with fineness modulus of 3.0.

The consistence of fresh concrete was determined by the conventional slump test

(ASTM C143/C143M-00). The descriptive statistics of the data set is listed in Table 1.

These data was randomly divided into training set (78 data) and testing set (25 data).

For this slump modeling problem the obvious inputs are the component contents of

concrete, including cement, fly ash, slag, water, SP, coarse aggregate (CA), and fine

aggregate (FA), and the output is the slump of the concrete. That is, the neural

network developed in the investigation has seven units in the input layer and one unit

in the output layer. After a number of trials, the back-propagation neural network with

one hidden layer containing three hidden units was selected.

The measured slump collected from the literature is plotted against the predicted

slump calculated by the aforementioned neural network model, as shown in Fig. 1 for

training set and Fig. 2 for testing set. Although the correlation between the measured

slump and the predicted values obtained from testing data are somewhat more

scattered than that obtained from training data, it is obvious that rather small scatter of

data around the diagonal line confirms the fact that neural network is an excellent

predictor of the slump.

7
 For comparison purpose, the same training set and testing set are used to build and

assess the polynomial regression model. The following polynomial regression formula

was adopted here:

q q
E ( y ) = ∑ β i xi + ∑∑ β ij xi x j (6)
i =1 i< j

where xi is the qth-component content; β i and β ij are the regression coefficients.

The regression model of slump is listed in Table 2. The measured slump collected

from the literature is plotted against the predicted slump calculated by the regression

model, as shown in Fig.3 and Fig 4. The correlation between the measured slump and

the predicted values, obtained from training data, are rather high. However, that

obtained from testing data are rather low, which indicates the fact that regression is

not an excellent predictor of the slump.

Moreover, the values of RMS error and R2 of the training and testing results for the

regression and neural network models are also listed in Table 3. It is seen that the

neural network model gives smaller RMS error and the larger R2 for both training set

and testing set, especially for testing set.

4. Computational Simulation of Concrete Slump

In this study, computational simulation of concrete slump was performed using the

trained neural network. The variation in concrete slump was achieved by varying

combinations of factors like the water/binder ratio (w/b), SP-binder ratio (SP/B), and

water content. The binder means cementitious material, that is, cement plus fly ash

and slag. The range of each variable is listed as follows:

1. The water-binder ratio (w/b) was varied with 0.3, 0.4, 0.5, 0.6, and 0.7.

8
2. The SP-binder ratio (SP/B), the amount of superplasticizer by weight of binder,

was varied with 0, 1, 2, 3, and 4%.

3. The amount of water was varied from 130 to 250 kg / m 3 .

Besides, the fly ash-binder ratio and slag-binder ratio was kept constant 25% and

25%, respectively, by weight of binder (cement + fly ash + slag); the CA/FA was kept

constant 1.0; the total volume of concrete was 1.000 m 3 .

From the water content-slump curves generated using the trained neural network

developed in this study with the above combinations, two sets of curves have been

shown in Fig. 5 to 9, Fig. 10 to 14, and Fig. 15 to 19, respectively, to explore the

effects of w/b, SP/B, and interaction of w/b and SP/B. Considering the fact that

validity of a slump test is generally recommended for concrete with a slump value

ranging from 25 to 250 mm; therefore, these parts of curves that outside the range

may be unreliable and should be ignored. It was found that

1. Discussions on Fig. 5 to 9

Figure 5 to 9 shows graphically how the slump varied with the SP/B ratio at

different w/b ratios.

(1) Although the water content and SP/B ratio were kept constant, Fig 5 to 9

demonstrates that a change in w/b ratio had a distinct effect on the consistence

properties.

(2) At water/binder ratio = 0.3, no combination of SP/B ratio and water content can

reach a slump larger than 50 mm.

(3) At water/binder ratio = 0.4, at a given SP/B ratio ≥2%, the concrete slump

increases at a gradually decreasing rate as the water content increases and starts to

decrease when the water content exceeds a certain optimum value. The optimum

water content has been found to be dependent on the SP/B ratio but is about at 230,

210, 190, and 170 kg / m 3 of the concrete for SP/B=2%, 3%, 4%, 5%, respectively.

9
 Why the slump decreases as the water content increases? A possible explanation is

that too much SP and water makes the concrete breeding.

(4) At water/binder ratio ≥ 0.6, the concrete slump increases sharply when the water

content exceeds a certain threshold, and increases slightly when the slump exceeds

a certain value about 150 mm, and then the SP/B has no effect on the slump when

the slump beyond the value.

(5) Although there are several factors affecting the required water, such as SP/Binder

ratio, water/binder ratio, cement characteristics, aggregate(s) type, and aggregate(s)

grading, according to the slopes of the curves in Fig. 5 to Fig. 9, in general, the

amount of water required to increase or decrease 25.4 mm in slump is about 4 –

5 kg / m 3 . However, there is a minimum water content required to start to produce

slump. The amount of water required to increase or decrease 25.4 mm in slump

differs between this study and the ACI method. ACI 211.1-81 recommends that if

the slump of the batch was incorrect, then increase or decrease the amount of

water by 6.0 kg / m 3 of concrete for each required increase or decrease of 25.4 mm

(1 in.) in slump.

2. Discussions on Fig. 10 to 14 and Fig. 15 to 19

Fig. 10 to 14 shows a large difference in the shape of the slump-SP/B curves

between low- and high-w/b concretes for a wide range of water content. Fig. 15 to 19

shows a large difference in the shape of the SP/B and water content to slump surface

between low- and high-w/b concretes.

As can be seen from these figures, the SP/B ratio had a distinct effect on

consistence properties at low and high w/b ratio.

(1) As can be seen in Fig. 10 and Fig. 15, at water/binder ratio = 0.3, no combination

of SP/B ratio and water content can reach a slump larger than 50 mm.

(2) As can be seen in Fig. 11 and Fig. 16, when w/b=0.4 and water content is greater

10
 than 180 kg / m 3 , there is a maximum slump on each slump curve.

(3) As can be seen in Fig. 12~14 and Fig. 17~19, when w/b=0.5~0.7, a high slump

can be reached with rather high water content (210~250 kg / m 3 ) without SP.

(4) A certain saturation level for the superplasticizer exists, above which only a small

effect of further dispersion is obtained. The saturation level depended on w/b and

water content. The results may be explained by that if the dosage level is greater

than that of saturation, the dispersion is supposed to be more inefficient.

5. Conclusions

The mechanism for consistence of concrete is rather complex, and a number of

factors are involved. Although the present simulating exploration was based on a

limited number of variables (w/b, water content, and SP/B), it appears that the

following conclusions can be drawn.

1. From the previous discussion, the use of neural network for the modeling of

concrete slump looks very promising.

2. Although the water content and SP/B ratio were kept constant, a change in w/b

ratio had a distinct effect on the consistence properties.

3. A certain saturation level for the superplasticizer exists, above which only a small

effect of further dispersion is obtained. The saturation level depended on w/b and

water content. The results may be explained by that if the dosage level is greater

than that of saturation, the dispersion is supposed to be more inefficient.

4. A certain saturation level for the water content exists, above which only a small

effect, even negative effect, on slump is obtained. The saturation level depended

on w/b and SP/B.

5. Due to remaining the constant w/b to reach a given strength, high water content

11
 means high binder is necessary; therefore, it may be more economical not adding

the water content beyond the saturation level into the concrete mixture.

Acknowledgments

This work was supported by the National Science Council, ROC, under Grant

NSC-96-2221-E-216-032.

References

1. Yen T, Tang CW, Chang CS, Chen KH. Flow behaviour of high strength

high-performance concrete. Cement and Concrete Composites. 1989; 21(5):

413-424.

2. Aitcin PC, Neville A. High-performance concrete demystified. Concrete

International. Jan. 1993; 21-26.

3. Domone PLJ, Soutsos MN. An approach to the proportioning of high-strength

concrete mixes. Concrete International. Oct. 1994; 26-31.

4. Naik TR, Ramme BW. High-strength concrete containing large quantities of fly

ash. ACI Materials Journal. 1989; 86(2):110-116.

5. Kwan AKH. Use of condensed silica fume for making high-strength,

self-consolidating concrete. Can. J. Civil Engineering. 2000; 27(4):620-627.

6. Punkki J, Golaszewski J, Gjorv OE. Workability loss of high-strength concrete.

ACI Materials Journal. 1996; 93(5):427-431.

7. Shen DH. Mix proportioning technology and mechanical properties of HPC with

pozzolans. Proceedings of Conference of High Performance Concrete, Taipei,

Taiwan, 1997; 107-111.

12
8. Faroug F, Szwaborski J, Wild S. Influence of superplasticizers on workability of

concrete. Journal of Materials in Civil Engineering. 1999; 11(2): 151-157.

9. Nehdi M, El-Chabib H, El-Naggar MH.. Predicting performance of

self-compacting concrete mixtures using artificial neural networks. ACI Materials

Journal. 2001; 98(5):394-401.

10. Brown DA, Murthy PLN, Berke L. Computational simulation of composite ply

micromechanics using artificial neural networks. Microcomputers in Civil

Engineering. 1991; 6(2):87-97.

11. Tang XW, Yen T. Prediction of the ultimate shear strength of high-and

normal-strength concrete beams without stirrups using neural networks. in High

Strength/High Performance Concrete, G. Konig et al. (eds), Volume 1, 2000;

595-608.

12. Ghaboussi J, Garrett JH, Wu X. Knowledge-based modeling of material behavior

with neural networks. Journal of Engineering Mechanics. 1991; 117(1): 132-153.

13. Kim JI, Kim DK, Feng MQ, Yazdani F. Application of neural networks for

estimation of concrete strength. Journal of Materials in Civil Engineering 2004;

16(3):257-264.

14. Peng J, Li Z, Ma B. Neural network analysis of chloride diffusion in concrete.

Journal of Materials in Civil Engineering. 2002; 14(4):327-333.

15. Stegemann JA, Buenfeld NR. Mining of existing data for cement-solidified wastes

using neural networks. Journal of Environmental Engineering. 2004; 130(5):

508-515.

16. Kim H, Rauch AF, and Haas CT. Automated quality assessment of stone

aggregates based on laser imaging and a neural network. Journal of Computing in

Civil Engineering. 2004; 18(1):58-64.

17. Pratt D, Sansalone M. Impact-echo signal interpretation using artificial

13
 intelligence. ACI Materials Journal. 1992; 89(2):178-187.

18. Basma AA, Barakat S, Al-Oraimi S. Prediction of cement degree of hydration

using artificial neural networks. ACI Materials Journal. 1999; 96(2): 167-172.

19. Haj-Ali RM, Kurtis KE, and Akshay R, Neural network modeling of concrete

expansion during long-term sulfate exposure. ACI Materials Journal. 2001; 98(1):

36-43.

20. Nehdi M, Djebbar Y, Khan A. Neural network model for preformed-foam cellular

concrete. ACI Materials Journal. 2001; 98(5):402-409.

21. Yeh IC. Modeling concrete strength with augment-neuron networks. Journal of

Materials in Civil Engineering. 1998; 10(4):263-268.

22. Yeh IC. Modeling of strength of high performance concrete using artificial neural

networks. Cement and Concrete Research. 1998; 28(12):1797-1808.

23. Yeh IC. Design of high performance concrete mixture using neural networks and

nonlinear programming. Journal of Computing in Civil Engineering. 1999;

13(1):36-42.

24. Goh ATC. Neural networks for evaluating CPT calibration chamber test data.

Microcomputers in Civil Engineering. 1995; 10:147-151.

25. Lippmann RP. An introduction to computing with neural nets. IEEE ASSP

Magazine, April 1987, 4-22.

26. Welstead ST. Neural Network and Fuzzy Logic Applications in C/C++, New York:

John Wiley & Sons, 1994.

27. Lam L, Wong YL, and Poon CS. Effect of fly ash and silica fume on compressive

and fracture behaviors. Cement and Concrete Research. 1998; 28(2):271-283.

28. Yeh IC, Chen JW. Modeling workability of concrete using design of experiments

and artificial neural networks. Journal of Technology. 2005; 20(2):153-162.

14
List of Figures

Fig. 1 The measured and predicted slump of neural network for training set

Fig. 2 The measured and predicted slump of neural network for testing set

Fig. 3 The measured and predicted slump of polynomial regression for training set

Fig. 4 The measured and predicted slump of polynomial regression for testing set

Fig. 5 The water content-slump curves at w/b=0.3

Fig. 6 The water content-slump curves at w/b=0.4

Fig. 7 The water content-slump curves at w/b=0.5

Fig. 8 The water content-slump curves at w/b=0.6

Fig. 9 The water content-slump curves at w/b=0.7

Fig. 10 The SP/B-slump curves at w/b=0.3

Fig. 11 The SP/B-slump curves at w/b=0.4

Fig. 12 The SP/B-slump curves at w/b=0.5

Fig. 13 The SP/B-slump curves at w/b=0.6

Fig. 14 The SP/B-slump curves at w/b=0.7

Fig. 15 The SP/B-water-slump surface at w/b=0.3

Fig. 16 The SP/B-water-slump surface at w/b=0.4

Fig. 17 The SP/B-water-slump surface at w/b=0.5

Fig. 18 The SP/B-water-slump surface at w/b=0.6

Fig. 19 The SP/B-water-slump surface at w/b=0.7

List of Tables
Table 1. Descriptive statistics of the data
Table 2. Regression model of slump
Table 3. RMS error and R2 of the training and testing results

15
 300

250
Predicted Slump (mm)

200

150

100

50 95% confidence limit

0
0 50 100 150 200 250 300
Measured Slump (mm)
Fig. 1 The measured and predicted slump of neural network for training set

300

250
Predicted Slump (mm)

200

150

100

50 95% confidence limit

0
0 50 100 150 200 250 300
Measured Slump (mm)
Fig. 2 The measured and predicted slump of neural network for testing set

16
 300

250
Predicted Slump (mm)

200

150

100

50

0
0 50 100 150 200 250 300
Measured Slump (mm)
Fig. 3 The measured and predicted slump of polynomial regression for training set

350

300
Predicted Slump (mm)

250

200

150

100

50

0
0 50 100 150 200 250 300 350
Measured Slump (mm)
Fig. 4 The measured and predicted slump of polynomial regression for testing set

17
 300
0%
1%
250
2%
3%
200 4%
5%
Slump (mm)

150

100

50

0
100 150 200 250
Water (kg)

Fig. 5 The water content-slump curves at w/b=0.3

300
0%
1%
250 2%
3%
200 4%
5%
Slump (mm)

150

100

50

0
100 150 200 250
Water (kg)

Fig. 6 The water content-slump curves at w/b=0.4

18
 300
0%
1%
250
2%
3%
200 4%
5%
Slump (mm)

150

100

50

0
100 150 200 250
Water (kg)

Fig. 7 The water content-slump curves at w/b=0.5

300
0%
1%
250
2%
3%
200 4%
Slump (mm)

5%
150

100

50

0
100 150 200 250
Water (kg)

Fig. 8 The water content-slump curves at w/b=0.6

19
 300
0%
1%
250
2%
3%
200 4%
5%
Slump (mm)

150

100

50

0
100 150 200 250
Water (kg)

Fig. 9 The water content-slump curves at w/b=0.7

20
 300
130
140
250
150
160
200 170
Slump (mm)

180
150 190
200
210
100
220
230
50 240
250
0
0% 1% 2% 3% 4% 5% 6% 7%
SP/B ratio

Fig. 10 The SP/B-slump curves at w/b=0.3

300
130
140
250
150
160
200 170
Slump (mm)

180
150 190
200
210
100
220
230
50 240
250
0
0% 1% 2% 3% 4% 5% 6% 7%
SP/B ratio

Fig. 11 The SP/B-slump curves at w/b=0.4

21
 300
130
140
250
150
160
200 170
Slump (mm)

180
150 190
200
210
100
220
230
50 240
250
0
0% 1% 2% 3% 4% 5% 6% 7%
SP/B ratio

Fig. 12 The SP/B-slump curves at w/b=0.5

300
130
140
250
150
160
200 170
Slump (mm)

180
150 190
200
210
100
220
230
50 240
250
0
0% 1% 2% 3% 4% 5% 6% 7%
SP/B ratio

Fig. 13 The SP/B-slump curves at w/b=0.6

22
 300
130
140
250
150
160
200 170
Slump (mm)

180
150 190
200
210
100
220
230
50 240
250
0
0% 1% 2% 3% 4% 5% 6% 7%
SP/B ratio

Fig. 14 The SP/B-slump curves at w/b=0.7

23
 300

250

Slump (mm)
200

150
100
50
0
130

5%
150

170 4%
3%
190
2%

210
1%

/B
Wate

230
rc 0%

SP
onten

250
t (kg)

Fig. 15 The SP/B-water-slump surface at w/b=0.3

300

250
Slump (mm)

200

150

100
50
0
130

5%
150

4%
170

3%
190

2%
210

1%
/B

Wate
230

rc 0%
SP

onten
t
250

(k g)

Fig. 16 The SP/B-water-slump surface at w/b=0.4

24
 300

250

Slump (mm)
200

150

100

50
0
130

5%
150

4%
170
3%
190 2%

210
1%

/B
Wate

230
rc

SP
onten 0%
t

250
(k g)

Fig. 17 The SP/B-water-slump surface at w/b=0.5

300

250

200
Slump (mm)

150

100

50
0
130

5%
150

4%
170

3%
190

2%
210

1%
/B

Wate
230

rc
SP

onten 0%
t (k g)
250

Fig. 18 The SP/B-water-slump surface at w/b=0.6

25
 300

250

200
Slump (mm)

150

100

50

0
130

5%
150

4%
170

3%
190
2%
210 1%

/B
Wate

230
rc onten

SP
t 0%
(kg)

Fig. 19 The SP/B-water-slump surface at w/b=0.7 250

26
 Table 1. Descriptive statistics of the data
Cement Fly ash Slag Water SP CA FA Slump
kg / m 3 kg / m 3 kg / m 3 kg / m 3 kg / m 3 kg / m 3 kg / m 3 (mm)
min 137.0 0.0 0.0 160.0 4.4 708.0 640.6 0
max 374.0 193.0 260.0 240.0 19.0 1049.9 902.0 290
average 229.9 78.0 149.0 197.2 8.5 884.0 739.6 180
stdev 78.9 60.5 85.4 20.2 2.8 88.4 63.3 88

Table 2. Regression model of slump

βi β ij coefficient
coefficient
Cement Fly ash Slag Water SP CA
Cement 4.99E-04
Fly ash -2.20E-03 1.42E-03
Slag -1.90E-03 2.39E-03 1.57E-03
Water -7.49E-04 -2.93E-03 4.47E-03 -1.06E-03
SP -7.99E-05 1.29E-02 -8.65E-02 -7.42E-02 -2.64E-02
CA 1.07E-02 -2.59E-03 -3.36E-03 -6.21E-04 5.27E-03 1.79E-02
FA 9.11E-03 3.12E-03 2.40E-03 9.28E-04 -2.58E-03 5.83E-03 -4.86E-04

Table 3. RMS error and R2 of the training and testing results

Neural Network Regression Model
Training Set Testing Set Training Set Testing Set
RMS error (mm) 33.2 41.2 52.4 108.0
R2 0.833 0.860 0.647 0.302

27

View publication stats