Heidari, A., Hashempour, M., Tavakoli, D. (2017). Using of Backpropagation Neural Network in Estimating of Compressive Strength of Waste Concrete. Journal of Soft Computing in Civil Engineering, 1(1), 54-64. doi: 10.22115/scce.2017.48040

Ali Heidari; Masoumeh Hashempour; Davoud Tavakoli. "Using of Backpropagation Neural Network in Estimating of Compressive Strength of Waste Concrete". Journal of Soft Computing in Civil Engineering, 1, 1, 2017, 54-64. doi: 10.22115/scce.2017.48040

Heidari, A., Hashempour, M., Tavakoli, D. (2017). 'Using of Backpropagation Neural Network in Estimating of Compressive Strength of Waste Concrete', Journal of Soft Computing in Civil Engineering, 1(1), pp. 54-64. doi: 10.22115/scce.2017.48040

Heidari, A., Hashempour, M., Tavakoli, D. Using of Backpropagation Neural Network in Estimating of Compressive Strength of Waste Concrete. Journal of Soft Computing in Civil Engineering, 2017; 1(1): 54-64. doi: 10.22115/scce.2017.48040

Using of Backpropagation Neural Network in Estimating of Compressive Strength of Waste Concrete

^{1}Associate Professor, Department of Civil Engineering, Shahrekord University, Shahrekord, Iran

^{2}M.Sc. Student, Department of Civil Engineering, Shahrekord University, Shahrekord, Iran

^{3}Ph.D., Department of Civil Engineering, Shahid Rajaee Teacher Training University, Tehran, Iran

Receive Date: 21 May 2017,
Revise Date: 07 July 2017,
Accept Date: 07 July 2017

Abstract

Waste concrete is one of the most usable and economic kind of concrete which is used in many civil projects all around the world, and its importance is undeniable. Also, the explanation of constructional process and destruction of them cause the extensive growth of irreversible waste to the industry cycle, which can be as one of the main damaging factors to the economy. In this investigation, with using of constructional waste included concrete waste, brick, ceramic and tile and stone new aggregate was made, also it was used with different weight ratios of cement in mix design. The results of laboratory studies showed that, the using of ratio of sand to cement 1 and waste aggregate with 20% weight ratio (W20), replacing of normal aggregate, increased the 28 days compressive strength to the maximum stage 45.23 MPa. In the next stage, in order to develop the experimental results backpropagation neural network was used. This network with about 91% regression, 0.24 error, and 1.41 seconds, is a proper method in estimating of results.

Concrete is one of the most usable building materials, its world consumption is more than one tone for one person, which makes it as one of the most usable materials on the earth after water. Also, more than million tons of waste is produced in the world every year. Many of these waste is non-recyclable and accumulate in a variety of tools. Therefore, it is very important to restore this waste to the industry cycle. One of the main parts of this waste is due to tiles and ceramics. Waste of tiles and ceramics are created in transition, during or after burning due to human mistakes or error of construction and even inferior raw materials. Most of this waste is because of demolishing of buildings [1]. The waste increases due to the extensive growth of production and consumption of these valuable materials. Various studies have been undertaken to recycle the waste in concrete in 2000. The investigation of chemical properties of waste tiles based on standards and then making samples with pozzolan and tests of compressive and bending strength, were done. The results showed that it is possible to use waste tiles in concrete, also, they have pozzolanic properties [2]. In 2013, Heidari et al. studied the effects of ceramic aggregates in concrete. For this purpose, the ceramic was replaced as coarse-grained of 0 to 40% as well as sand as 0 to 100%. The results showed that the use of ceramic has no significant negative effect on concrete properties. The optimum sample in the case of ceramic replacement with sand was about 25 to 50 Percent. The best sample, using of ceramic as coarse grained, was between 10 to 20%. In this case, not only the increase in compressive strength is observed, but also the specific weight is reduced without any significant negative effect on water absorption [1]. In order to improve the mechanical properties and reduce the water absorption, Nano particles of silica were used. Final results showed that, using of Nano particles can highly improve the waste concrete parameters because of filling of capillary voids in cement paste [3] . In the bricks industry, due to the high quantity of bricks in the building industry, researchers have proposed different combinations for the manufacture of concrete containing this type of waste with various tests and studies. In 1983, one of the first experimental studies in this field was carried out, and finally it showed a compressive strength of a downward trend and a reduction of 10-35% for coarse grained of replacing brick with natural aggregate. However, compared to conventional concrete with this reduced compressive strength, the tensile strength increases by 11% [4]. In another study in 2014, in the first phase, Tavakoli and Heidari investigated the effect of replacing sand with clay brick in different proportions and observed that there is not a big problem in using of brick as sand in optimum percent and in mechanical properties like compressive strength. However, due to the high water absorption, a longer examination is required [5]. In 1993, Merlet and Pimienta investigated the use of waste concrete in reproduction of constructional concrete. They used the waste concrete in different ratios as fine grained in new concrete for the first time [6]. Later, shrinkage of drying of this concrete was evaluated. The results showed that there is a high shrinkage in comparison of conventional concrete [7], which can be controlled by adding fly ash [8]. Several studies have also been carried out on the use of different neural networks in estimating laboratory values. In 2017, Kalman et al. used neural network to estimate the properties of concrete with waste brick aggregate. The results showed that the neural network can estimate these properties and show very close results to experimental ones [9]. In another study, neural network, ANFIS, and multi-layer regression (MLR) were used to estimate 28 days compressive strength of concrete with waste aggregate. Finally, due to input data to network, although all of the modeling ways acted well, but MLR network showed the most exact answers for estimating the values [10]. In the same field, the neural network and ANFIS were used simultaneously to study the compressive strength of concrete blocks. 102 samples were used to construct the network and parameters such as the ratio of height to thickness of the concrete charter and compressive strength as the input of the model were used. Finally, it was determined that these models can estimate the required values with high precision [11]. Using of neural networks with a mixture of optimization algorithms such as genetic, PSO and etc. which is used to improve the workability, not only exist in all of the fields, but also has a wide usage in concrete production. In a study, in order to simulation of concrete slump, genetic algorithm was used to optimize the bias and weights. In this study backpropagation neural network was processed and developed [12].

In this investigation two phases were considered. In the first phase, experimental studies about using waste materials such as brick, ceramic tile and concrete as fine aggregate. The results showed that this method is not only practical, but also economical. In this method, in order to have the best answer for mix design, the ratio of sand to cement should be 1, also a design with 20% of waste had the best results. In the second phases, numerical methods such as practice of data in Excel and backpropagation neural network were used to develop the prediction of values. The results of this phase showed that neural network with high accuracy of measurement can be used to predict the values.

2. Materials

2.1. Aggregate preparation

In this study, in order to use the constructional waste as aggregate in concrete, it should be crushed enough, also its condition in concrete should be examined. In this kind of concrete, gravel was not used and the used aggregate is in size of sand. At first, constructional waste like, concrete, brick, stone also tile and ceramic waste should be crushed completely with the help of impact crusher and then particles of gravel are separated from sand by the sieve 4. Ordinary sand and waste sand are shown in figure1.

Figure1. A: waste sand. B: ordinary sand

Before doing the experiments, it is essential to make sure about pozzolanic property or lack of it in waste sand, because this property can influence the whole aspects of experimental results. So, the experiment of XRF was done on the samples of these two kind of sand.

Table 1: XRF analysis of sands

Sand constituents in percent

Element

Waste sand

Ordinary sand

1.45

0.59

Al_{2}O_{3}

0.56

0.32

SiO_{2}

53.76

54.21

CaO

0.96

0.06

Fe_{2}O_{3}

0.43

0.26

K_{2}O

2.27

1.04

MgO

0.03

0.01

MnO

1.74

2.15

Na_{2}O

0.08

0.08

P_{2}O_{5}

3.78

0.57

S_{2}O_{3}

0.15

0.01

TiO

33.40

40.58

LOI

Investigating these results showed that there is no pozzolanic property in waste sand and it is similar to stone. Also, grading of these two kinds of sand was done based on figure3. The results of the study of these two graphs based on the Iran standard of grading showed that both of these two kinds of sand are in standard area and they can be used in concrete.

Figure 2. Grading of sands

2.2. Cement

The cement which was used, is 1-525 type of Shahrekord cement factory. Analysis of this material is shown in table 2.

Table 2: Properties of cement

Chemical properties

%

Physical properties

Value

SiO_{2}

20.30

Blain( )

≥3200

Al_{2}O_{3}

5.65

Setting initial time (minute)

90

Fe_{2}O_{3}

3.30

Setting final time(minute)

170

CaO

65.70

Autoclave expansion

≤3.0

MgO

≤1.60

2 days compressive strength

≥24.0

S

≤2.80

3 days compressive strength

≥35.0

CL

≤0.03

7 days compressive strength

≥47.0

F.CaO

≤1.80

28 days compressive strength

≥61.0

I.R

≤0.65

L.O.I

≤1.30

Total Alkali

≤0.70

2.3. Water

Potable water was used in this study. In this water, PH is 7.8, Chloride 40 and Sulphate 29 . So potable water has standard parameters and it is proper for mix design.

2.4. Superplasticizer

The superplasticizer which was used in this study, is high water reducer based on polycarboxylate copolymer that has a wide range in ultra-high performance concrete. This brown superplasticizer with 1.08 density, less 0.01 CL ion, and 6 to 8 PH is manufactured to be used in UHPC, SCC, impermeable, and concrete with high workability. This superplasticizer with high distribution property of cement particles is an admixture to produce UHPC concrete with minor ratio of water to cement. Moreover, it doesn't have any CL ion, so it is usable in producing the reinforced concrete. Also, with less using of this material, a homogenous mix can be made.

3. Mix design

In this investigation two phases with ratios of sand to cement, 1 and 1.5 are considered and each phase has 11 mix designs, which ordinary and waste sand were used as percent of total aggregate from 0 to 100 percent in each design. Also, in both phases, the ratio of water to cement was 0.5. Table of mix design for this concrete is shown as follows. Meanwhile, W is used in order to name each design, and the number of after it shows the percent of waste sand.

Table 3: Mix design

Superplasticizer

(weight ratio of cement)

Type of sand

Sample

Phase 1

Phase 2

Ordinary sand

(%)

Waste sand

(%)

0.3

0.4

0

100

W100

0

0.3

10

90

W90

0

0.2

20

80

W80

0

0.1

30

70

W70

0

0.1

40

60

W60

0

0

50

50

W50

0

0

60

40

W40

0

0

70

30

W30

0

0

80

20

W20

0

0

90

10

W10

0

0

100

0

W0

After preparing the aggregate, total weight of 5 Kgs cement and aggregate based on the mix design, was poured into the mixer, then the water contained of superplasticizer was added. This concrete was mixed for 3 minutes which was adjusted based on ACI304-37 standard. Finally, this made concrete was poured into the 5 centimeters cube molds.

4. Experimental results

After curing the concrete samples in water, compressive strength tests were done, and the results were analyzed. Figure 3 shows concrete samples with different percent of waste.

The most important experimental test is compressive strength test. This test is based on ASTM C-109 standard. In this investigation, 1, 7 and 28 days samples were broken by 2000KN under load hydraulic jack and then the results were determined. Figure 4 shows the results of phase 1 and 2.

Figure 4. Results of compressive strength in phase 1 and 2

As it was shown, samples which were cured for 28 days had higher compressive strength in comparison with other samples. Also, 28 compressive strength and 7 days compressive strength were close because of using the cement type 1-525 which can cause strength in short term. Moreover, with increasing the ordinary sand, strength will increase, too. This fact is because of replacing high strength ordinary sand with low strength waste sand. So, mechanical properties of microstructure of concrete will improve. Nevertheless, in the first phase, the sample with 20% waste sand had the highest strength. Also, in the second phase, the sample with 30% of waste had the highest strength.

5. Neural network

Neural networks are derived from human biological intelligence. Each human neurons cell consists of three main parts called axon, dendrite and soma, which dendrites are the agent of reception information from other cells and axons are the agent of information transmission from other cells. The soma or cell body is also the place where the data is entered into the cell. The neural network also has the same mechanism in which the neuron receives signals that are affected by the weights and biases, as well as the neuron cell as the processor and the information get processed. The processed information is transferred to the neuronal cell again and this cycle is repeating. On the other hand, the neural network can be similar to the model multivariate linear regression is also known. Input neurons similar to independent variables and output neurons like the dependent variable acts and the weights have the same function width of the source or constant sentence in the regression. In general, the role of neurons in the neural network is information processing. This is the case in artificial networks that is the same as the activation function [13].

Since the estimation and prediction of values using neural networks is fast and relatively new, it is therefore necessary to avoid conclusive results from other values that were not possible due to the limitation of laboratory conditions. Assured. First, we estimate the values for fitting the results using Excel software. It is shown in linear fitting data of figure 5.

Figure 5. Fitting data in Excel

Excel fittings with low regression cannot estimate the values and they are unreliable. So, in this stage, backpropagation neural network with artificial intelligence basic and optimization algorithms were used. In order to prepare the network, at first layered status and input and output data were determined. For a basic network, five kinds of input, two hidden layers, and three kinds of output based on compressive strength were used. Figure 6 shows this graph.

Figure 6. Layered status for neural network

Before importing the data and training the network, in order to decrease the errors, formula 1 was used to normalized the data [14].

(1)

Where is normalized data, X input or output parameter, minimum input or output parameter, is the maximum input or output parameter. In order to, create the network, 70% of data for training, 15% for evaluating and 15% for testing were used. The final answer is in figure 7.

Figure 7. Regression of neural network

As it is shown, network regression is higher and more acceptable than excel. Network errors are shown in figure 8. In this figure, neural network with high regression and less errors could create the best answer in estimating the values.

Figure 8. Error of neural network

Maximum time consumption in this study, was 1.46 seconds, which was recorded for core i7 CPU and 16 gigabyte ram. This value is different for other systems. Totally, this item is proper because of selecting two hidden layers and considering Levenberg Marquardt training function which is one of the fastest functions of toolbox of MATLAB.

6. Conclusion

This study has two experimental and numerical parts which the results are as follows:

Adding construction waste as aggregate in concrete is an acceptable and not only compressive strength has not diminished, but has also been remarkable for some designs. For example, the maximum compressive strength for phase 1 is sand-to-cement ratio 1 and the waste is 20% up to 45.23 MPa.

The results of compressive strength in 7 and 28 days samples, were so close, which was because of using of cement type 1-525 and its short term strength.

In order to develop and estimate the experimental data, two layers back propagation neural network was used.

The results of fitting data of neural network were more accurate and closer to real values in comparison with fitting data of Excel.

Finally, it was concluded that back propagation neural network is accurate and accelerator in estimating and fitting processing the experimental data.

References

Tavakoli, D., Heidari, A., & karimian, M. (2013). Properties of concretes produced with waste ceramic tile aggregate. Asian journal of civil engineering (BHRC), 14(3), 369-382.

Ay, N., & Unal, M. (2000). The use of waste ground ceramic in cement production. Cement and Concrete Research, 30(3), 497-499.

Heidari, A., & Tavakoli, D. (2013). A study of the mechanical properties of ground ceramic powder concrete incorporating nano-SiO2 particles. Construction and Building Materials, 38, 255-264. doi:https://doi.org/10.1016/j.conbuildmat.2012.07.110

Akhtaruzzaman, A. A., & Hasnat, A. (1983). Properties of concrete using crushed brick as aggregates. Concr. Int, 5(2), 58-63.

Tavakoli, D., Heidari, A., & Hayati Pilehrood, S. (2014). Properties of Concrete made with Waste Clay Brick as Sand Incorporating Nano SiO2. Indian Journal of Science and Technology, 7(12), 1899-1905.

Merlet, J., & Pimienta, P. (1993). Mechanical and physical–chemical properties of concrete produced with coarse and fine recycled aggregates. Paper presented at the Demolition and reuse of concrete and masonry.

Zega, C., & di Maio, A. (2011). Use of recycled fine aggregate in concrete with durable requirements. Waste Manage, 31, 2336-2340.

Jeong, G. (2011). Processing and properties of recycled aggregate concrete. Paper presented at the MSc dissertation in Civil Engineering.

Kalman Šipoš, T., Miličević, I., & Siddique, R. (2017). Model for mix design of brick aggregate concrete based on neural network modelling. Construction and Building Materials, 148, 757-769. doi:https://doi.org/10.1016/j.conbuildmat.2017.05.111

Khademi, F., Jamal, S. M., Deshpande, N., & Londhe, S. (2016). Predicting strength of recycled aggregate concrete using Artificial Neural Network, Adaptive Neuro-Fuzzy Inference System and Multiple Linear Regression. International Journal of Sustainable Built Environment, 5(2), 355-369. doi:https://doi.org/10.1016/j.ijsbe.2016.09.003

Zhou, Q., Wang, F., & Zhu, F. (2016). Estimation of compressive strength of hollow concrete masonry prisms using artificial neural networks and adaptive neuro-fuzzy inference systems. Construction and Building Materials, 125, 417-426. doi:https://doi.org/10.1016/j.conbuildmat.2016.08.064

Chandwani, V., Agrawal, V., & Nagar, R. (2015). Modeling slump of ready mix concrete using genetic algorithms assisted training of Artificial Neural Networks. Expert Systems with Applications, 42(2), 885-893. doi:https://doi.org/10.1016/j.eswa.2014.08.048

Heidari, A., Tavakoli, D., & Fakharian, P. (1392). Approximation of special sheet values using artificial neural network. Quarterly Journal of Engineering Modeling, 11(35).

Doğan, E. (2009). Reference evapotranspiration estimation using adaptive neuro-fuzzy inference systems. Irrigation and Drainage, 58(5), 617-628. doi:10.1002/ird.445