Eskisehir Osmangazi University, Faculty of Architecture and Engineering, Department of Computer Engineering, Meselik Campus, Eskisehir, Turkey

Transcription

1 PREDICTING THE SPLITTING TENSILE STRENGTH OF CONCRETE CONTAINING ZEOLITE AND DIATOMITE UNDER THE EFFECT OF MgSO 4 by ANN Eyyup Gulbandilar 1 *, Yilmaz Kocak 2 1 Eskisehir Osmangazi University, Faculty of Architecture and Engineering, Department of Computer Engineering, Meselik Campus, Eskisehir, Turkey 2 Dumlupinar University, Kutahya Vocational School of Technical Sciences, Department of Construction, Kutahya, Turkey Abstract This study was designed to investigate with artificial neural network (ANN) prediction model for the behavior of concrete containing zeolite and diatomite under the effect of MgSO 4. For purpose of constructing this model, 7 different mixes with 63 specimens of the 28, 56 and 90 days splitting tensile strength experimental results of concrete containing zeolite, diatomite, both zeolite and diatomite used in training and testing for ANN system was gathered from the tests. The data used in the ANN model are arranged in a format of seven input parameters that cover the age of samples, Portland cement, zeolite, diatomite, aggregate, water and hyper plasticizer and an output parameter which is splitting tensile strength of concrete. In the model, the training and testing results have shown that ANN system has strong potential as a feasible tool for predicting 28, 56 and 90 days the splitting tensile strength of concrete containing zeolite and diatomite under the effect of MgSO 4. Key words: artificial neural network, splitting tensile strength, concrete, zeolite, diatomite 1. INTRODUCTION During the previous decades, enormous researchers evaluated the effects of the partial replacement of cement by various types of additions that improve cement properties have been used in the cement. The most frequently used additions are fly ash, silica fume, blast furnace slag, rice husk ash, pummice, zeolite, diatomite nano-silica, trass, burned clay, volcanic tuff and metakaolinite (Wang & Lee 2014; Tabatabaei et al. 2014; Behnood & Ziari 2008; Goñi et al. 2013; Kelestemur & Demirel 2010; Kocak 2010). The use of additional cementitious materials due to the advantages such as economic, technical and environmental considerations has become very common in cement and concrete technology (Fu et al. 2002; Worrell et al. 2000). Two of the materials are zeolite and diatomite (Gerengi et al. 2013, Gerengi et al. 2015; Kocak et al. 2013; Yildiz et al. 2010). Diatomite is a mineral described as consisting of the fossilized siliceous shell of the microscopic single celled alga and possessing the structural properties of amorphous silica. There are nearly fifteen thousand types of diatomite in the nature. Diatomites generally have the shape of a round tray or a long fish. The size of them are μm, when they are dry their specific gravity is , they contain % of SiO 2 and are cellular materials with high water absorption rate (Aruntas & Tokyay 1996). Zeolite is defined as allophones that consist of alkali and alkaline earth cations and have the crystal structure. Zeolites have water molecules in their canals which is one of the most significant properties setting apart them from other mineral groups. When they are heated at o C, these water molecules leave the material continuingly without changing the structure of the zeolite. Another important property of zeolite is ring canals. These canals are full with univalent and bivalent cations such as Na +, K + and Ca + with water molecules clinging on them (Canpolat 2002; Serbest 1999). Nowadays, artificial neural network (ANN) has been used by many researchers to solve a wide variety of problems in civil engineering applications. Kocak et al. (2015) used build models which have architecture in ANN system to evaluate the effect of fly ash and silica fume on compressive strength of cement mortars. For purpose of constructing this model, 8 different mixtures with 240 specimens of the 2, 7, 28, 56 and 90 days compressive strength results of cement mortars containing fly ash and Page 529

2 silica fume used in training and testing for ANN system. In training and testing of the models constituted with an architecture the age of samples (days), Portland cement, fly ash and silica fume were entered as input; while compressive strength values of cement mortars were used as output. The model was trained with 180 data of experimental results. The training and testing results in the ANN model have shown that ANN have strong potential for predicting 2, 7, 28, 56 and 90 days compressive strength values of cement mortars containing fly ash and silica fume (Kocak et al. 2015). Beycioglu et al. (2015) were applied an ANN to predict the compressive strength of clinker mortars. ANN model with two hidden layer was constructed, training and testing stages have been performed using the available test data of 1288 different clinker mortars mix-designs used. The ANN model had seven input parameters and one output parameter. The obtained results from compressive strength tests were compared with predicted results. The results showed that ANN can be alternative approaches for the predicting of compressive strength of clinker mortars (Beycioglu et al. 2015). Most of ANN applications in civil engineering focused on concrete and cement mortars properties such as workability, mechanical behavior and physical properties (Ahmaruzzaman 2010; Subasi 2009; Yaprak et al. 2013; Topcu et al. 2009; Atici 2011; Ashrafi Jalal & Garmsiri 2010). The aim of this study is to build model in ANN system to evaluate the effect of splitting tensile strength of concrete containing zeolite and diatomite under the effect of MgSO 4. For purpose of constructing this model, 7 different mixes with 63 specimens of the 28, 56 and 90 days splitting tensile strength experimental results of concrete containing zeolite, diatomite, both zeolite and diatomite used in training and testing for ANN system were gathered from the concrete tests. The model was trained with 63 data of experimental results. The ANN model had seven input parameters and one output parameter. The obtained results from splitting tensile strength of concrete were compared with predicted results. 2. EXPERIMENTAL STUDY In preparing the concrete samples, Portland cement (PC), diatomite, zeolite, aggregate, well water and hyper plasticizer are used. The PC was CEM I 42.5 R which was provided from Bolu Cement Plant. Diatomite is supplied from ASU Chemistry and Mining Firm and zeolite is supplied from a Turkish Zeolite Firm. Asar River aggregates in Duzce region (0 5 mm crashed sand, 5 19 mm crushed stone and mm) as aggregate are used. Well water from Doganli village in Duzce as mixing water is used. Moreover, the type of fluid 70 produced by AYDOS Construction Chemicals Factory and new generation hyper plasticizer with solid matter content of 34.32%, intensity of (20 o C), ph value of 7.26 (20 o C) are applied as admixture for concrete. The chemical properties of the PC, diatomite and zeolite are given at the Table 1. Table 1. The chemical, properties of PC, diatomite and zeolite Materials PC Diatomite Zeolite Chemical composition, wt.% SiO Al 2O Fe 2O CaO MgO SO Na 2O K 2O S+A+F Loss on ignition Insoluble residue Free CaO Page 530

3 In the study, 7 different cements, which are PC, 10 20% diatomite, 10 20% zeolite, 5+5% 10+10% diatomite and zeolite are substituted for Portland cement, are used. For concrete mixture design, materials amounts to be put into the mixture are determined within the framework of the method stated in TS 802 standards (TS ). According to the type and rate of mineral additive, which is substituted for the concrete, seven types of concrete are produced. They are encoded as R, 10D, 20D, 10Z, 20Z 5D5Z and 10D10Z according to the addition rate and the used mineral additive. According to TS EN , consistency of concrete in fresh concrete is stated for each mixing group individually (TS EN ). The materials amounts of the samples in concrete mixture of 1m 3 and the characteristics of fresh concrete are given in the Table 2. Materials Aggregate Table 2. Material quantity in the 1 m 3 for each concrete group Specific R, 10D, 20D, 10Z, 20Z, 5D+5Z, gravity 10D+10Z, Total PC Diatomite Zeolite Hyper plasticizer Water The produced concrete samples are poured into 15x15x15 cm cubic molds without segregation. Splitting tensile strength experiments of concrete samples have been done according to TS EN (TS EN ). Splitting tensile strength experiments are carried out for 28 days, in 23±2 o C water, then on the concrete samples which are cured MgSO 4 of the media factor and in the level 2 (56 and 90 days) of concrete age factor. 3. ARTIFICIAL NEURAL NETWORK Artificial neural network (ANN) consisted of an arbitrary number of simple elements called neurons. Neurons in ANN are, as similar in human brains, interconnected (Adhikary & Mutsuyoshi 2006). ANN represents simplified methods of a human brain and uses new methods to solve problems rather than conventional methods with traditional computations which have difficult solution procedures (Trtnik et al. 2009). Generally, ANN is consisted of an input layer of neurons, one or more hidden layers of neurons and output layer of neurons. The neighboring layers are fully interconnected by weight. The input layer neurons receive information from the outside environment and transmit them to the neurons of the hidden layer without performing any calculation. Layers between the input and output layers are called hidden layers and may contain a large number of hidden processing units. All problems, which can be solved by a perceptron can be solved with only one hidden layer, but it is sometimes more efficient to use two hidden layers. Finally, the output layer neurons produce the network predictions to the outside world (Demir 2008). Figure 1 clearly illustrates the typical neural network which is composed of five main parts such as; inputs, weights, sum function, activation function and outputs (Topcu et al. 2008; Parichatprecha & Nimityongskul 2009). Page 531

4 Inputs Weights Sum function Activation function Outputs X 1 W 1 X 1 W b X 2 W 2 X 2 W Σ (net)j 1-1 O j X n W X n W n Figure 1. The artificial neuron model. The input of a neuron comes from another neuron and it is obtained by multiplying the output of the connected neuron by the synaptic strength of the connection between them. The weighted sums of the input components (net) j are calculated by using Eq. (1) below: ( net) j w ijo i n i1 b (1) Where (net)j is the weighted sum of the j th neuron for the input received from the preceding layer with n neurons, w ij is the weight between the j th neuron in the preceding layer, o i is the output of the i th neuron in the preceding layer and b is a fix value as an internal addition (Topcu et al. 2008). Activation function is a function that processes the net input obtained from sum function and determines the neuron output. In general, for multilayer feed forward models as the activation function (f (net) j) sigmoid activation function is used. The output of the j th neuron (out) j is computed using Eq. (2) with a sigmoid activation function as follows: (Topcu et al. 2009). o j 1 f (net) j (2) α(net ) 1 e j Where α is constant used to control the slope of the semi-linear region. The sigmoid nonlinearity activates in every layer except in the input layer. The sigmoid function represented by Eq. (2) gives outputs in (0, 1). If it desired, the outputs of this function can be adjusted to (-1, 1) interval. As the sigmoid processor represents a continuous function it is particularly used in non-linear descriptions. Because its derivatives can be determined easily with regard to the parameters within (net) j variable (Topcu et al. 2009). 4. ARTIFICIAL NEURAL NETWORK MODEL AND PARAMETERS In training and testing of the ANN model the age of samples, PC, zeolite, diatomite, aggregate, water and hyper plasticizer were entered as input; while splitting tensile strength of concrete were used as output (Table 3). Page 532

5 Input variable Output variable Table 3. The input and output quantities used in ANN model Data used in training and testing the model Minimum Maximum Age of samples (days) PC, g Zeolite, g 0 80 Diatomite, g 0 80 Aggregate Water Hyper plasticizer Splitting tensile strength (MPa) For the training of the model were used 63 of the experimental data and 21 data as the average of these test results were used for testing the trained model. The designed ANN consisted of feed-forward back propagation, two hidden layers, training function (Levenberg-Marquardt), adaptation learning function (learngdm), transfer function (tansig) and performance function (MSE-mean squared error) as demonstrated in Figure 2. Figure 2. The architecture used in the ANN for splitting tensile strength. The neurons used in the system are selected 10 and 5 for the first and second hidden layers, respectively. Momentum rate and learning rate values were determined and the model was trained through iterations. The parameter values obtained from the multilayer feed-forward neural network model were given in Table 4. Table 4. The values of parameters used in models Parameters ANN Number of input layer neurons 7 Number of hidden layer 2 Number of first hidden layer neurons 10 Number of second hidden layer neurons 5 Number of output layer neuron 1 Error after learning 1x10-4 Learning cycle 6 The trained model was tested only with the input values and the predicted results were close to the experimental results. Page 533

6 5. RESULTS AND DISCUSSION Multilayer feedforward network models that contain two hidden layers are used in order to find more reliable solutions. Determination of optimum number of the hidden layers neurons are very important to accurately predict the parameters used by ANN. Starting with a few numbers of neurons and then slightly increasing the number of neurons gives the best approach for finding the optimum number of hidden neurons. The performance of the ANN model is monitored according to chosen performance criteria during this process for each hidden neuron number. This process is repeated until the error becomes acceptably small or no significant improvement is observed. This study uses different neurons in the two hidden layers at the beginning of the process then the neuron number was increased step-by-step adding 1 neuron until no significant improvement is noted. The ANN model tried to be compared according to the absolute fraction of variance (R 2 ), mean absolute percentage error (MAPE) and a root-mean squared (RMS) error criteria. These criteria are defined by Eqs. (3), (4) and (5) respectively (Ozcan et al. 2009). RMS 1 N N i1 t i o 2 i (3) R 2 N 2 ti oi i1 1 (4) N 2 oi i1 MAPE 1 N N i1 ti o i oi 100 (5) Here t is the target value, o is the network output value, N is the total number of pattern. In the training and testing of ANN model from experimental data and average of these test results are used. In the ANN model, 63 data of experiment results were used for training whereas 21 data as average of these test results for testing. Sample number and experimental results with training and testing results obtained from ANN model were given in Fig. 3 and 4, respectively. Page 534

7 12 Splitting tensile strength, MPa Exp-training (28 days) ANN-training (28 days) Exp-training (56 days) ANN-training (56 days) Exp-training (90 days) ANN-training (90 days) Splitting tensile strength, MPa Sample number Figure 3. Comparison of splitting tensile strength experimental and training results with sample number. 12 Exp-testing (28 days) ANN-testing (28 days) Exp-testing (56 days) ANN-testing (56 days) Exp-testing (90 days) ANN-testing (90 days) Sample number Figure 4. Comparison of splitting tensile strength average of test results and testing results with sample number. Also, inputs values and experimental results with testing results obtained from ANN model were given in Table 5. Page 535

8 Predicted splitting tensile strength, MPa Materials, Methods & Technologies Table 5. Comparison of splitting tensile strength average of test results with testing results obtained from ANN Splitting tensile Data used in the model construction strength, MPa As, days PC, Zeolite, Diatomite, Aggregate, Water, Hyper plasticizer, Exp. ANN All results obtained from the studies and predicted by using the training and testing results of ANN model for 28, 56 and 90 days splitting tensile strength were given in Figure 5 and 6, respectively ANN training stage (63 data), R² = RMS = MAPE = Experimental splitting tensile strength, MPa Figure 5. Comparison of splitting tensile strength experimental results with training results of model. Page 536

9 Predicted Splitting tensile strength, MPa ANN testing stage (21 data) R² = RMS = MAPE = Experimental splitting tensile strength, MPa Figure 6. Comparison of splitting tensile strength average of test results with testing results of model. The linear least square fit line, its equation and the R 2 values were shown in these figures for the training and testing data. As it is visible in Fig. 5 and 6 the values obtained from the training and testing in ANN model are very closer to the experimental results. The result of testing phase in Fig. 5 and 6 shows that the ANN model is capable of generalizing between input and output variables with reasonably good predictions. The statistical values for all the station such as RMS, R 2 and MAPE were given in Table 6. Table 6. The splitting tensile strength statistical values of proposed ANN model Statistical parameters ANN Training set Testing set RMS R MAPE While the statistical values of RMS, R 2 and MAPE from training in the ANN model were found as , and , respectively, these values were found in testing as , and , respectively (Table 6). All of the statistical values show that the proposed ANN model is suitable and predict the 28, 56 and 90 days splitting tensile strength values very close to the experimental values. 6. CONCLUSION In this study, ANN was used for the prediction the 28, 56 and 90 days splitting tensile strength values of concrete containing zeolite, diatomite, both zeolite and diatomite. In the model developed in ANN system, a multilayered feed-forward neural network with a back-propagation algorithm was used. In the multilayer feed-forward neural network model, two hidden layers were selected. In the first hidden layer 10 neurons and in the second hidden layer 5 neurons were determined. This model was trained with input and output experimental data. Using only the input data in trained models the 28, 56 and 90 days splitting tensile strength values of concrete containing zeolite, diatomite, both zeolite and diatomite were found for testing the model. The splitting tensile strength values are very closer to the Page 537

10 experimental data obtained from training and testing for ANN model. The statistical parameter values of RMS, R 2 and MAPE that calculated for comparing experimental data with ANN model results have shown obviously this situation. As a result, splitting tensile strength values of concrete containing zeolite, diatomite, both zeolite and diatomite can be predicted in the multilayer feed-forward neural network model in a quite short period of time with tiny error rates. The conclusions have shown that ANN system is practicable methods for predicting splitting tensile strength values of concrete containing zeolite, diatomite, both zeolite and diatomite. ACKNOWLEDGEMENT The authors would like to thank Duzce University Presidency of Scientific Research Projects that provided financial support with the project code number HD.009 and Duzce Yigitler Beton that enabled the tests to be carried out. REFERENCES Adhikary, BB & Mutsuyoshi, H 2006, Prediction of shear strength of steel fiber RC beams using neural Networks, Construction and Building Materials, vol. 20, no. 9, pp Ahmaruzzaman, M 2010, Review on the utilization of fly ash, Progress in Energy and Combustion Science, vol. 36, Aruntas, HY & Tokyay, M 1996, The use of diatomite as pozzolanic material in blended cement production, Cement and Concrete World, vol. 1, no.4, pp Ashrafi, HR, Jalal, M & Garmsiri, K 2010, Prediction of load displacement curve of concrete reinforced by composite fibers (steel and polymeric) using artificial neural network, Expert Systems with Applications, vol. 37, pp Atici, U 2011, Prediction of the strength of mineral admixture concrete using multivariable regression analysis and an artificial neural network, Expert Systems with Applications, vol. 38, pp Behnood, A & Ziari H 2008, Effects of silica fume addition and water to cement ratio on the properties of high-strength concrete after exposure to high temperatures, Cement and Concrete Composites, vol. 30, pp Beycioglu, A, Emiroglu, M, Kocak, Y & Subasi, S 2015, Analyzing the compressive strength of clinker mortars using approximate reasoning approaches ANN vs MLR, Computers and Concrete, vol. 15, no. 1, pp Canpolat, F 2002, Utilization of natural zeolite and industrial waste in improving cement performance in cement production, PhD Thesis, Suleyman Demirel University, Sakarya (in Turkey). Demir, F 2008, Prediction of elastic modulus of normal and high strength concrete by artificial neural network, Construction and Building Materials, vol. 22, no. 7, pp Fu, X, Wang, Z, Tao, W, Yang, C, Hou, W, Dong Y & Wu, X 2002, Studies on blended cement with a large amount of fly ash, Cement and Concrete Research, vol 32, no. 7, pp Gerengi, H, Kocak, Y, Jażdżewska, A, Kurtay, M & Durgun, H 2013, Electrochemical investigations on the corrosion behaviour of reinforcing steel in diatomite and zeolite containing concrete exposed to sulphuric acid, Construction and Building Materials, vol. 49, pp Gerengi, H, Kurtay, M & Durgun, H 2015, The effect of zeolite and diatomite on the corrosion of reinforcement 4 steel in 1 M HCl solution, Results in Physics, vol. 5, pp Page 538

11 Goñi, S, Frias, M, Vigil de la Villa, R & García R 2013, Sodium chloride effect on durability of ternary blended cement: Microstructural characterization and strength, Composites: Part B, vol. 54, pp Kelestemur, O & Demirel, B 2010, Corrosion behavior of reinforcing steel embedded in concrete produced with finely ground pumice and silica fume, Construction and Building Materials, vol. 24, pp Kocak, Y 2010, A study on the effect of fly ash and silica fume substituted cement paste and mortars, Scientific Research and Essays, vol. 5, no 9, pp Kocak, Y, Gulbandilar, E & Akcay, M 2015, Predicting the compressive strength of cement mortars containing FA and SF by MLPNN, Computers and Concrete, vol. 15, no. 5, pp Kocak, Y, Tasci, E & Kaya, U 2013, The effect of using natural zeolite on the properties and hydration characteristics of blended cements, Construction and Building Materials, vol. 47, pp Ozcan, F, Atis, CD, Karahan, O, Uncuoglu, E & Tanyildizi, H 2009, Comparison of artificial neural network and fuzzy logic models for prediction of long-term compressive strength of silica fume concrete, Advances in Engineering Software, vol. 40, pp Parichatprecha, R & Nimityongskul, P 2009, Analysis of durability of high performance concrete using artificial neural Networks, Construction and Building Materials, vol. 23, pp Serbest, D 1999, The Use of natural zeolites in the industry of the light construction, Master's Thesis, Anadolu University, Eskisehir (in Turkey). Subasi, S 2009, Prediction of mechanical properties of cement containing class C fly ash by using artificial neural network and regression technique, Scientific Research and Essay, vol. 4, no. 4, pp Tabatabaei, R, Sanjaria, HR & Shamsadini, M 2014, The use of artificial neural networks in predicting ASR of concrete containing nano-silica, Computers and Concrete, vol. 13, no. 6, pp Topcu, İB, Karakurt, C & Saridemir, M 2008, Predicting the strength development of cements produced with different pozzolans by neural network and fuzzy logic, Materials and Design, vol. 29, pp Topcu, İB, Saridemir, M, Ozcan, F & Severcan, MH 2009, Prediction of long-term effects of GGBFS on compressive strength of concrete by artificial neural networks and fuzzy logic, Construction and Building Materials, vol. 23, pp Trtnik, G, Kavčič, F & Turk, G 2009, Prediction of concrete strength using ultrasonic pulse velocity and artificial neural Networks, Ultrasonics, vol. 49, pp TS , Design of concrete mixes, Turkish Standards, Ankara (in Turkish). TS EN , Testing fresh concrete- Part 2: Slump test, Turkish Standards, Ankara (in Turkish). TS EN , Testing hardened concrete - Part 6: Tensile splitting strength fo test specimens, Turkish Standards, Ankara (in Turkish). Wang, X-Y & Lee, H-S 2014, Prediction of compressive strength of slag concrete using a blended cement hydration model, Computers and Concrete, vol. 14, no. 3, pp Worrell, E, Martin, N & Price, L 2000, Potentials for energy efficiency improvement in the US cement industry, Energy, vol. 25, no. 12, pp Yaprak, H, Karaci, A & Demir, I. 2013, Prediction of the effect of varying cure conditions and w/c ratio on the compressive strength of concrete using artificial neural networks, Neural Computing and Application, vol. 22, pp Page 539

12 Yildiz, K, Dorum, A & Kocak, Y 2010, The investigation of the effect of minerological molecular electrokinetical and thermal compliance of pumice, zeolite and CEM I cement on high strength concrete, Journal of the Faculty of Engineering and Architecture of Gazi University, vol. 25, no. 4, pp Page 540