You are on page 1of 7

Construction and Building Materials 24 (2010) 17541760

Contents lists available at ScienceDirect

Construction and Building Materials


journal homepage: www.elsevier.com/locate/conbuildmat

Completely random experimental design with mixture and process variables for optimization of rubberized concrete
Raimundo K. Vieira *, Rafael C. Soares, Samantha C. Pinheiro, Otvio A. Paiva, Jos O. Eleutrio, Raimundo P. Vasconcelos
Universidade Federal do Amazonas, Faculdade de Tecnologia, 69077-000 Manaus, Brazil

a r t i c l e

i n f o

a b s t r a c t
A completely random experimental design has been used to simultaneously optimize process variables and cementitious mixture components for rubberized concrete. In this study the concrete is composed of only three mixture components, cement, aggregate (including ne and coarse) and water, and two process variables, size of rubber (chips of waste truck tires) and percentage rubber that replaced the aggregate in concrete, were varied. The calculations of regression and residual ANOVA sums of squares for regression and lack of t are illustrated. These values are shown to be useful for model development. Two different models were evaluated. The bilinearlinear model has no lack of t and is preferred. The signicant terms in this model are capable of describing how the process response surfaces change as mixture level conditions are varied. Optimum condition in the rubberized concrete, considering the level of the variable studied is: 2.4 mm size of rubber, 2.5% rubber that replaced the aggregate, 16% of cement, 76% of aggregate and 8% of water. The optimum mixture conditions with 2.5% of rubber show a concrete value of compressive strength above 20 MPa, can generate concrete suitable for use in structures as well as pavement, curbs, walls and other applications in civil engineering. 2010 Elsevier Ltd. All rights reserved.

Article history: Received 11 December 2009 Received in revised form 29 January 2010 Accepted 2 February 2010 Available online 2 March 2010 Keywords: Rubberized concrete Experimental design Mixture and process variables Rubber tire

1. Introduction Population growth, and the greater use of disable materials have increased considerably the generation of solid residue, to the point of landll cannot accommodate the total demand. One of the major contributors to the solid waste problem is the growing volume of tires. The disposal of tires have become a global problem, the increased storage of discarded has tires created potential re hazards re as well as health risks due to with the proliferation of mosquitoes that breed in stagnant water which accumulates in the tires [1]. Many attempts have been made to solve the problem of tire disposal, such as: (1) the use of tire rubber in asphaltic concrete mixtures, (2) incineration of tires for the production of steam, and (3) reuse of ground tire rubber in a number of plastic and rubber products. The use of tires in asphalt concrete for paving is technically feasible and economically attractive. Nevertheless, the asphalt industry can currently absorb only 3040% of the carcasses of tires generated. [2]. Application of tires as fuel is technically practicable, but it is not economically attractive because of the high initial capital invest-

* Corresponding author. Address: Universidade Federal do Amazonas, Faculdade de Tecnologia, Av. General Rod. O.J. Ramos, 3000, Manaus AM 69077-000, Brazil. Tel.: +55 92 3305 4410. E-mail address: maneiro01@ig.com.br (R.K. Vieira). 0950-0618/$ - see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.conbuildmat.2010.02.013

ment. The use of rubber tires to produce carbon black eliminates the costs of grinding and chopping, but the carbon black produced by pyrolysis of tires is very expensive and lower in quality than that produced with petroleum [2]. Application of tires in plastic and rubber products can be considered a viable alternative; however by itself unable to absorb the enormous volume of solid waste generated from the tire industry. Moreover, the addition of tire rubber to plastic could produce a composite with properties worse than those obtained with virgin plastic. Several studies have been carried out concerning the addition of rubber as a replacement for aggregates in concrete mixtures [3 14]. In general, these studies, have established that the addition of rubber to concrete is capable of changing the characteristics of the concrete, causing an increase in its elastic behavior, but reducing its compressive strength when the particle size of rubber is increased. The principal application of rubberized concrete in civil construction include its use as insulation, by preventing the spread of tension in structures such as in the case of earthquakes, and paving as where high strength is not very important, such as in sub bases and highway crash barriers [2]. Obtaining particles of tire rubber in usable condition can, in some cases, be more expensive than buying mineral aggregate. However, this study is not driven by economics of replacing aggregate mineral

R.K. Vieira et al. / Construction and Building Materials 24 (2010) 17541760

1755

with rubber aggregate in concrete mixes, but rather addresses an environmental problem that affects the entire world. This kind solution can be exploited in regions where it is difcult to obtain conventional construction materials, such as the state of Amazonas, in Northern Brazil, which due to its geological formation, does not have rocks easily accessible for use in construction. Because of this, gravel is obtained by dredging of river beds, causing siltation of rivers. Nevertheless, gravel is the most frequently used coarse aggregate most in the commercial production of concrete. The cost of the material is very high because of the long distance that must be traveled to obtain it. Thus the replacement of gravel with tire rubber, even minimally, should decrease damage to the ecosystem of the region. However, in order to use rubber in concrete there are two major problems that need to be addressed: The rst is to determine the largest proportion of rubber tire that can be added to concrete with minimal reduction in the compressive strength. Several experimental studies have been carried out on rubberized concrete to try to solve this problem. The other problem is of determination the particle size of the rubber. For years, studies have been carried out using various methods in order to anticipate the concrete properties in advance. However, such attempts often involved an application of empirical techniques. There have been few studies that have used rigorous mathematical modeling to forecast the proprieties of cementitious material. Topu and Sardemir [15,16] have used the articial neural network and fuzzy logic. These studies mark a new approach to the study of rubberized concrete that in the future will require more elaborate strategies to analyze the incorporation of tire rubber in concrete or mortar cement. In this context this study proposes a new approach which predicts the behavior of tire rubber in concrete, using a single model and implementing an experimental design to study the composition of concrete in a multivariate form. The aim of the present study was to determine the optimal proportion and size of the tire rubber that can be added to concrete as a substitute for aggregate with minimal loss of compressive strength using a completely random experimental design. The experimental design utilizes multivariate optimization, a technique that is being used increasingly in chemistry, chemical engineering, food engineering, pharmacology and others elds. The major advantage of multivariate design is that it has a higher applicability, minimal costs and a high degree of accuracy [17 19]. In addition, interaction among factors can only be discovered by using multivariate strategies. The type of experimental design to be applied in each optimization problem depends on the statistical nature of the variables involved. In general, the variables can be grouped into two broad classes: The mixture variables are known as factors whose properties depend on the proportion of each component and not its absolute quantities. The quantity of each system component must be treated as a variable (variable of mixture), which is not independent of others (i.e., the sum of their proportions must be one), which in mathematical terms can be described as [20]:

most frequently used models are the linear (Eq. (4)), quadratic (Eq. (5)) and special cubic (Eq. (6)). Linear:

^ y a0 a1 z1 a2 z2
Bilinear:

^ y a0 a1 z1 a2 z2 a12 z1 z2
Linear:

^ y b1 x1 b2 x2 b3 x3
Quadratic:

^ y b1 x1 b2 x2 b3 x3 b12 x1 x2 b13 x1 x3 b23 x2 x3


Special cubic:

^ y b1 x1 b2 x2 b3 x3 b12 x1 x2 b13 x1 x3 b23 x2 x3 b123 x1 x2 x3 6

where a are parameters of the process model (the effects); b are the parameters of the mixture model, z are the process variable, x are ^ the mixture variable and y represents the value of the response predicted by the regression model. However, in concrete, the mixture variables such as the properties of the mixture constituents (e.g., aggregates, water and cement), are affected by changes in the process conditions (such as the tire rubber aggregate size and percetagem of tire rubber in the concrete). In this case, when interactions occur between the process conditions and mixture compositions, it is useful to apply a planning strategy which generates models to quantify this type of interaction. Such model can be obtained by combining mixture models and process models. Combined models are obtained by simply multiplying a process variable model by a mixture model. For example, the combined bilinearlinear model is obtained by multiplying Eq. (2) by Eq. (4),

^ y d01 x1 d02 x2 d03 x3 d11 x1 z1 d12 x2 z1 d13 x3 z1 d21 x1 z2 d22 z2 x2 d23 x3 z2 d121 x1 z1 z2 d122 x2 z1 z2 d312 x3 z1 z2 7

0 6 xi 6 1;

q X i1

xi 1

where q is the number of components of the mixture and xi are the components. Process variables are factors that are not part of the mixture, although when the levels are variable, it can affect the mixture properties. Process variables such as the particle size of tire rubber, are independent and can change without restriction [20]. The most frequently used models which describe the effects of process variables are the linear models (Eq. (2)) and bilinear models (Eq. (3)). To describe the inuence of the mixture variables, the

where dij is the combination of ai and bj factors. Optimization problems involve mixture and process variables of simultaneous form where the experiments were carried out in a completely random manner having been described in the literature [21,22]. Although there are hundreds of multivariate optimization problems described in the literature, there are few cases where the application of both types of variables occurs simultaneously. Therefore, this type of design often involves the adjustment of large numbers of variables in order to obtain the ideal set of experimental conditions that produce the most desirable results. However, some of these variables are not easily adjusted and cannot be included in a completely random experimental design and still maintain a feasible optimization program. Split-plot procedures [19,2329] can reduce these operational difculties since subsets of experiments are set up. The hard-to-adjust variables are included in one subset and the other variables in a second subset. Complete randomization is restricted since the hard-to-adjust variables are maintained at constant levels for groups of experiments for which all the other variables are adjusted in random order following some experimental design. Then other groups of experiments are performed in random order at different levels of the hard-to-adjust variables until the entire experimental design has been executed. However, there is a price to pay for simplifying laboratory work using the split-plot procedure. The split-plot ANOVA calculation is more complicated than the standard ANOVA since the former has

1756

R.K. Vieira et al. / Construction and Building Materials 24 (2010) 17541760

two kinds of errors, main-plot and subplot errors, whereas there is only one error source for completely randomized designs. In both cases, ANOVA results are used to determine model lack of t and evaluate standard errors for model coefcients in order to determine signicant model terms. However, error determinations for model parameters obtained from split-plot designs are also more complex than those of completely randomized designs. For this reason, this study to add to the literature addressing planning involving process and mixture variables with complete randomization by presenting a simple form to study concrete mixtures, without the need to use complex mathematical or statistical models. 1.1. Analysis of variance Suppose a simple example with a factorial plan that contains only two factors, A and B, with a and b respective levels, and n replicates. If the experiments are carried out in complete random order, the analysis of the variance model can be represented by [25]:

yijk l ai bj dij eijk

where i = 1, 2, . . . , a; j = 1, 2, . . . , b; k = 1, 2, . . . , n; l is the global average of all responses; ai represents the effects of i-th A factor treatment; bj the effects of j-th B factor treatment; (d)ij interaction effects between A and B and eijk represents a random error. The best way to verify whether the effects are statistically signicant is to use a statistical procedure known as the Analysis of Variance, ANOVA, and the hypotheses to be tested is: H0 = s = b = (sb) = 0, so if there are no effect of the factors A and B and Ha = s 0 or b 0, then at least the effect of one factor is signicant. The general principle of ANOVA is to divide the total observations of variability for each factor. Thus, the square sum is calculated, representing the variability attributed to each factor and also that which comes from experimental error. By dividing the square sums value by its respective degrees of freedom square mean is obtained, which is used in F tests, in order to verify the aforementioned hypotheses tests. The ANOVA design with complete randomization of experiments is shown in Table 1 [18]. To verify the statistical signicance of regression, an F test must be carried out by comparing the value of reason MSR/MSr with the value of distribution F tabled, with (p 1, n p) degrees of freedom. The lack of t of the mathematical model is tested by comparing the value of reason MSLof/MSPe with F tabled value with (m p, n m) degrees of freedom. If the value of MSR/MSr exceeds the value of F tabled, the proposed model does not represent lack of t; condence intervals for the coefcients of the regression model can be constructed. It is important to note that in the analysis of

variance of the random procedure there is only one term representing the random error, represented by e in Eq. (8). This term feature incorporates all variability sources that are not due to the factors studied and it includes measurement errors, reading errors, uncontrollable factors, noise of the process, and variability caused by the environment. Supposing that the error is a random variable, it has a normal and independent distribution, with the average of zero and the variance constancy of r2; this hypothesis is known as homoscedasticity. This error, if random, can never be described by the regression model and, is thus called residue and is dened as the difference between the value predicted by the regression model and the experimentally observed value. To verify the mentioned suppositions about residues, the main models diagnostic graph must be veried, using the residue graph vs. normal values expected and residues vs. predicted values. While in the rst graph the residue values must fall in a continuous line that represents the expected values according to the normal distribution, in the second they must present a completely random default, indicating that every systematic variation was included in the model. In addition, a very useful graph is the predicted value vs. experimentally observed value for the planning points and also for external samples, if available, because this graph is indicative of the models predictive capacity. The statistical model coefcients are normally obtained by the least squares method, by the matrix equation: (XtX)1 (Xty), in which X represents the conventional design matrix and y the observed responses vector. The estimated of coefcient errors is obtained by the equation (XtX)1 s2, in which s2 represents the estimated experimental variance, which can be estimated by repeated measures, using MSr if the model does not present lack of t and MSPe if the model presents lack of t.
2. Experimental design One unique type of scrap tire rubber came from used truck tires. The process to recap tires produces one type of rubber in chip form with large interval size. It was not possible to determine the gradation curve for the tire chips as is possible for normal aggregates because they were elongated particles between 1 mm and 10 mm. For this reason, three intervals in diameter size of rubber were chosen: from 1.2 mm to 2.4 mm, from 2.4 mm to 4.8 mm, and greater than 4.8 mm (Fig. 1). Data regarding the properties of the aggregates and the rubber particles are given in Table 2. CP II Z 32 Portland cement (according to Brazilian NBR 5732 code), manufactured in Manaus, and natural river sand were used in this investigation. For the development of the present research, ne and coarse aggregates that were treated as only aggregate. This aggregate was mixed with water and cement, and these three components constituted the mixture variables. The process variables chosen were the rubber size and the percentage of rubber that replaced the aggregate in the concrete. 2.1. Completely random experimental design (CRED) The CRED used the mixture of variables embedded within the process variables for optimization. Three mixture variables, cement, aggregate and water proportion and two process variables, size of tire rubber and proportion of rubber that replaced the aggregates, were simultaneously adjusted. Table 3 shows the three concrete mixtures included in the design and their respective cement, aggregate and water proportions as well as the 23 factorial design for the scaled process variables. The

Table 1 Analysis of variance for planning in which the experiments were performed in complete random order [18]. Source Regression Residues Lack of t Pure error Total % var. explained % max. explainable SS SSR SSr Pm Pni Pm Pni
i j i j i j

DF ^  yi y2 ^  yi y i 2 (p 1) (n p) (m p) (n m) (n 1)

MS SSR/(p 1) SSr/(n p) SSLf/(m p) SSPe/(n m)

Table 2 Material properties and mix proportioning of control mixtures. Material Specic gravity Set of mixture 1 Cement (kg/m3) Tire chips (%)b Gravel Sand (kg/m3) Water (kg/m3)
a b a

SSLf

Pm Pni

^ yij yi 2

2 366 5.0% 560 462 183

3 354 7.5% 541 560 177

 yij yi 2 SSPe i j P P  SST m ni yij y2 i j SSR/SST (SSTSSPe)/SST

Pm Pni

2305 1170 2610 2620 1000

374 2.5% 572 436 187

p represents the number of parameters of the mathematic model, n is the total numbers of experiments and m is the number of planning levels.

As described in Table 3. % Rubber that substituted the aggregate.

R.K. Vieira et al. / Construction and Building Materials 24 (2010) 17541760 Table 3 A 23 factorial design with size of tires rubber, proportion of rubber that replaces the aggregate and mixture design with cement, aggregate and water proportions for the CRED. Process 1 2 3 4 5 6 7 8 9 Mixtures 1 2 3
a

1757

Za1 1 0 1 1 0 1 1 0 1

Za2 1 1 1 0 0 0 1 1 1 Cement (X1) 0.16 0.15 0.15

Size rubber (mm) 1.20 2.40 4.80 1.20 2.40 4.80 1.20 2.40 4.80 Aggregate (X2) 0.76 0.77 0.78

% Rubber substitute 2.5 2.5 2.5 5.0 5.0 5.0 7.5 7.5 7.5 Water (X3) 0.08 0.08 0.07

standardized variables.

analytical responses observed for all the experimental conditions, in terms of 28-d compression strength (MPa), are given in Table 4. Different combined models were formed using two process variable models in combination with models of process and mixtures shown in Eqs. (2)(6), as presented in Eq. (7). In order to investigate the mechanical properties of tirerubber concrete, specimens of a cylindrical shape 15 cm in diameter 30 cm in height were fabricated following the leads showed in Table 3. These rubber concrete composites were prepared keeping the water/cement ratio constant at 0.5. Part of the sand (ne aggregate) was substituted for rubber. The weight percentage of rubber used was 2.5%, 5% and 7.5%. The scrap rubber that was used was only cleaned with water. The effects of surface treatment should be the subject of another study using the best size and proportion of rubber in concrete to be determined by this investigation.

shows lack of t. However, the bilinear process and linear mixture variable models resulted in combined models with no signicant lack of t since the lack of t means square/main-plot pure error means square ratio (MSLof/MSPe) of 2.37/2.21 = 1.07 is less than the 95% condence F15,27 tabled value of 2.06. This result indicates an estimate of the variance due to purely experimental uncertainty because there is no evidence that the model presents lack of t at a condence level of 95%. The models coefcients, their standard errors, and t-test values for the bilinearlinear model are given in Table 6. The standard errors were calculated using the ANOVA results of Table 5. There is no exact hypothesis test that can be performed for the l l l t-test, t bk =s:e:bk where s.e. bk represents the standard error involving the kth mixture variable and lth process variable [23]. Possible critical t-values can be investigated using a normal probability plotting of the t-test values in Table 6. This graph is not shown because its results were not conclusive. For the application presented in Ref. [23], Cornell used a t value of three or more to indicate the most signicant model coefcients. Applying this cut-off value to the bilinearlinear model results in a model with signicant x1 term. Thus, considering the interval between greater terms and smaller terms, the cut-off value was not attributed. This consideration maintains the model shown in Table 6, which can be represented in this manner:

^ y 227:41x1 18:25x2 40:51x3 75:82x1 z1 44:26 6:6650:6054:52 4:27x2 z1 111:83x3 z1 57:11x1 z2 2:14z2 x2 4:2761:9754:528:16 138:86x3 z2 73:33x1 z1 z2 27:92x2 z1 z2 120:81x3 z1 z2 61:9766:7810:0075:90 9

3. Results Only two process-mixture variable models were investigated, combining both the linear and bilinear process variable models, Eqs. (2) and (3), with the linear (Eq. (4)), because the degree freedom number for the mixture variables cannot be applied to the others mixture models such as quadratic (Eq. (5)) and special cubic (Eq. (6)). Table 4 contains the ANOVA results for the two combined models involving linear and bilinear process variable models and the linear mixture model. The differences between the part of the process variable model resulted in signicant changes in the extracted compressive strength since the lack of t is determined for the mean square/main-plot pure errors mean square ratio of 4.70/ 2.21 = 2.13. This value is larger than the 95% condence level F18,27 tabled value of 2.0 for the combined liner-linear model. This result indicates than in 95% of the condence level this model

Table 5 CRED ANOVA table for combined linearlinear and bilinearlinear models. Model Linearlinear Source Regression Residues Lack of t Pure error Total % Var. explained % Max. explainable Regression Residues Lack of t Pure error Total % Var. explained % Max. explainable SS 538.03 144.21 84.61 59.61 682.19 78.87 91.26 587.14 95.11 35.50 59.61 682.19 86.07 91.26 DF 8 45 18 27 53 MS 67.25 3.20 4.70 2.21

Bilinearlinear

11 42 15 27 53

53.38 2.26 2.37 2.21

Table 4 Experimental 28-d compressive strength rubber concrete determinations for the duplicated 54 experiment complete random design for two process and three mixture variables. Formulation numbera 1 R1 z1 = 1; z2 = 1 z1 = 0; z2 = 1 z1 = 1; z2 = 1 z1 = 1; z2 = 0 z1 = 0; z2 = 0 z1 = 1; z2 = 0 z1 = 1; z2 = 1 z1 = 0; z2 = 1 z1 = 1; z2 = 1
b

2 R2 22.90 22.40 20.40 19.40 19.10 20.60 16.50 16.70 15.00 R1 21.00 20.20 21.80 13.80 17.80 14.40 11.90 12.40 16.30 R2 21.20 23.40 21.50 15.50 17.00 16.30 11.00 12.80 14.40

3 R1 21.80 19.70 19.90 17.60 18.80 15.80 10.40 16.10 15.80 R2 22.00 22.00 13.70 16.70 15.80 17.70 12.30 15.30 15.30

Table 6 Parameters, estimated standard errors and t-test ratios for the bilinearlinear combined model. Variables x1 x2 x3 x1z1 x2z1 x3z1 x1z2 x2z2 x3z2 x1z1z2 x2z1z2 x3z1z2 Parameters 227.41 18.25 40.51 75.82 4.27 111.83 57.11 2.14 138.86 73.33 27.92 120.81 Standard error 44.26 6.66 50.60 54.52 8.16 61.97 54.52 8.16 61.97 66.78 10.00 75.90 t-test 5.14 2.74 0.80 1.39 0.52 1.80 1.05 0.26 2.24 1.10 2.79 1.59

22.30 25.10 24.80 17.90 18.50 18.80 14.90 16.60 15.00

The analytical responses are given as compressive strength, in megapascals. a Formulation number from Table 3. b Replicate number.

1758

R.K. Vieira et al. / Construction and Building Materials 24 (2010) 17541760

Fig. 1. Type and size of tire rubber used.

Fig. 2. Graph of the values predicted by Eq. (9) against the experimental values. The line representing exact agreement is also shown.

Fig. 3. Graph of the residuals predicted values against experimental values.

Fig. 2 contains a graph of the experimental values against predicted values from Eq. (9). The deviations in the graph correspond to a root mean square error of 1.25. The points are almost randomly distributed about the line, representing exact agreement, providing there is no evidence of lack of t for the bilinearlinear model. A graph of the residuals against the predicted values is shown in Fig. 3. This graph shows random behavior without a tendency towards residuals for experimental values. It should be noted in Eq. (9) that the inuence of coefcient of x1, is already expected since the cement is a bond element in the mixture, with a signicant contribution in the compressive strength. The presence of the signicant xz crossed terms in the model shows that the mixture of variables response surface depends on the size and proportion of the rubber added. Performing the least-square t of the mixture data for each set of main-plot conditions resulted in the following process models:

x1 0:15; x2 0:78; x3 0; 07 ^ y 17:04 2:83z1 0:22z2

12

The coefcients in these equations are consistent when compared with equations evaluated in each of these three mixture levels and placing the appropriate values of x1, x2, and x3 into Eq. (9). Since there exists a statistical signicance in xz cross terms in Eq. (9), different optimization conditions can be expected for each set of mixture levels. These were obtained from Eqs. (10)(12) and are given in Fig. 4. 3.1. Effect of the percentage of tires rubber replacing aggregates in concrete According to Fig. 4, the most appropriate percentage of rubber that can replace the aggregates in concrete is 2.5%, considering that the optimum mixture is 16% of cement, 76% of aggregate and 8% of water. On the whole, the graphs presented in Fig. 4 where this variable is analyzed, it was veried that the compressive strength increases when the percentage of rubber decreases. This behavior is expected since the rubber which is present in a chemical composition different from conventional components of concrete, in which

x1 0:16; x2 0:76; x3 0:08 ^ y 19:27 3:60z1 0:06z2 x1 0:15; x2 0:77; x3 0:08 ^ y 16:32 3:44z1 0:11z2

10

11

R.K. Vieira et al. / Construction and Building Materials 24 (2010) 17541760

1759

Fig. 4. Process response surfaces for variation of size and proportion of rubber for each set of mixture variable conditions. Fig. 4ac correspond each set of mixture and (d) compressive strength against least square means of rubber size.

the latter are usually mineral products while tires are composed mainly of polymers of organic origin. Moreover, it is common knowledge that the greater the addition of tire rubber to the concrete, the lower its resistance to compression. However, when comparing the three levels of mixtures, it was observed that in all of them a concrete value of compressive strength above 20 MPa could be obtained. This indicates that these mixtures, even if not optimized, or undergoing any type of surface treatment and without addition of additives can generate concrete that can be used in structures. 3.2. Effect of mixture composition Comparing the three types of mixtures it is easy to determine that the optimal composition is: x1 = 0.16; x2 = 0.76; x3 = 0.08. All mixtures have the same proportion of water/cement, but this set shows a greater relation between cement and aggregate. This result is also expected because the higher proportion of binder (cement), the greater the integration between the components of the mixture and consequently the greater its cohesion, leading to a higher compressive strength. 3.3. Effect regarding rubber size Finally, when considering the particle size of tire rubber, the rst two mixtures do not show a clear distinction between what would be the best size for rubber to be incorporated into the concrete. However, the third mixture (x1 = 0.15; x2 = 0.78; x3 = 0.07) shows that the best size of rubber would be the intermediate. This observation is obvious when comparing only the average of the results of compressive strength against the size of tire rubber added to the concrete, as shown in Fig. 4d. However, the effect of the size

of the rubber chips is very limited since only small amount of rubber chip used in the concrete. These two results seem modest despite the replacement of only 2.5% of rubber tire waste with intermediate size. However, they can be seen as an environmental breakthrough, considering the astronomical amounts produced annually of tires waste that can be used as raw material in construction. This result becomes more relevant when one considers the possible use of rubberized concrete as a structural components in civil construction. Another great advantage is the possibility of using intermediate size rubber, which opens up the chance of using rubber tires without the need for pretreatment. This permits many possibilities to use concrete in pavement, curbs, walls and other applications in civil construction. 4. Conclusions The CRED method has been proven to be quite useful into simplifying operational procedures in which a large number of experiments are required. Model development does not present lack of t. The combined process-mixture variable model is capable of explaining how the process responds to surfaces and optimum process variable levels vary as they are changed in mixture conditions. The optimum condition in the rubberized concrete, considering this level of variables studied is 2.4 mm size of rubber, 2.5% of rubber that replaced the aggregate, 16% of cement, 76% of aggregate and 8% of water. According to the given rubberized concrete data, it was possible to obtain in all of the mixture conditions a concrete value of compressive strength above 20 MPa for 28 days. This indicates that these mixtures, even if not optimized or undergoing any type of surface treatment and without addition of additives, can generate concrete suitable for use in structures as well as pavement, curbs, walls and other applications in civil engineering.

1760

R.K. Vieira et al. / Construction and Building Materials 24 (2010) 17541760 [12] Khaloo AR, Dehestani M, Rahmatabadi P. Mechanical properties of concrete containing a high volume of tirerubber particles. Waste Manage 2008;28:247282. [13] Topu B, Bilir T. Experimental investigation of some fresh and hardened properties of rubberized self compacting concrete. Mater Des 2009;30(8): 305665. [14] Topu B, Bilir T. Analysis of rubberized concrete as a three phase composite material. J Compos Mater 2009;43(11):125163. [15] Topu B, Sardemir M. Prediction of rubberized concrete properties using articial neural network and fuzzy logic. J Mater Process Technol 2008;199:10818. [16] Topu B, Sardemir M. Prediction of rubberized mortar properties using articial neural network and fuzzy logic. Constr Build Mater 2008;22:53240. [17] Box GEP, Hunter JS, Hunter WG. Statistics for experimenters. New Jersey: John Wiley and Sons; 2005. [18] Bencio BN, Scarminio IS, Bruns RE. Planejamento e Otimizao de experimentos. Campinas: Editora da Unicamp; 1995. 299p. [19] Wooding WM. The split-plot design. J Qual Technol 1973;5:1633,. [20] Kowalski SM, Vining GG, Cornell JA. Split-plot designs and estimation methods for mixture experiments with process variables. Technometrics 2002;44:729. [21] Dingtad G, Egelandsdal B, Naes T. Modeling methods for crossed mixture experimentsa case study from sausage production. Chemom Intell Lab Syst 2003;66:17590. [22] Campisi B, Vojnovic D, Chicco D, Phan-Tan-Luu R. Melt granulation in a high shear mixer: optimization of mixture and process variables using a combined experimental design. Chemom Intell Lab Syst 1999;48:5970. [23] Cornell JA. Analyzing data from mixture experiments containing process variables: a split-plot approach. J Qual Technol 1988;20:223. [24] Cornell JA. Process variables in mixture problem for categorized components. J Am Stat Assoc 1971;66:42. [25] Montgomery C. Design and analysis of experiments. New York: Wiley; 1991. [26] Mge I, Ns T. Split-plot design for mixture experiments with process variables: a comparison of design strategies. Chemom Intell Lab Syst 2005;78(1-2):8195. [27] Langhans I, Goos P, Vandebroek M. Identifying effects under a split-plot design structure. J Chemometr 2005;19:515. [28] Narinder SS, Piepel GF, Tormod N. Product and process improvement using mixture-process variable methods and robust optimization techniques. J Qual Technol 2009;41(2):18198. [29] Bortoloti JA, Bruns RE, Andrade JC, Vieira RK. Split-plot design optimization for trace determination of lead by anodic stripping voltammetry in a homogeneous ternary solvent system. Chemometr Intell Lab 2004;70:11321.

Acknowledgements This research was funded by Fundo Setorial de Infra-Estrutura (CT-INFRA) through MCT/CNPq and Fundao de Amparo a Pesquisa do Estado do Amazonas (FAPEAM). RKV thanks the CAPES for the program PRO-ENGENHARIA. Thanks are also due to FAPEAM for a fellowship to RCS.

References
[1] Eldin NN, Senouci AB. Engineering properties of rubberized concrete. Eng Prop Rubber Concr 1992;19:91223. [2] Raghavan D, Huynh H, Ferraris CF. Workability, mechanical properties, and chemical stability of a recycled tyre rubber-lled cementitious composite. J Mater Sci 1998;33:174552. [3] Rostami H, Lepare J, Silverstraim T, Zandi I. Use of recycled rubber tires in concrete. In: Proceedings of the international conference concrete 2000; 1993. p. 3919. [4] Li Z, Li F, Li JSL. Properties of concrete incorporating rubber tyre particles. Mag Concr Res 1998;50(4):297304. [5] Topu B. Properties of rubberized concretes. Cem Concr Res 1995;25(2): 30410. [6] Topu B. Properties of concretes containing waste rubber chips. In: Third international symposium on environmental geotechnology, San Diego, CA, USA; 1996. p. 38694. [7] Fedroff D, Ahmad S, Savas BZ. Mechanical properties of concrete with ground waste tire rubber. Washington, DC: Transportation Research Board; 1996. p. 6672 [Transportation Research Record No. 1532]. [8] Khatib ZK, Bayomy FM. Rubberized Portland cement concrete. J Mater Civ Eng 1999;11(3):20613. [9] Gneyisi E, Gesoglu M, zturan T. Properties of rubberized concretes containing silica fume. Cem Concr Res 2004;34(12):230931. [10] Albano C, Camacho N, Reyes J, Feliu JL, Hernndez M. Inuence of scrap rubber addition to Portland I concrete composites: destructive and non-destructive testing. Compos Struct 2005;71:43946. [11] Ghaly Ashraf M, Cahill IV James D. Correlation of strength, rubber content, and water to cement ratio in rubberized concrete. Can J Civil Eng 2005;32: 107581.

You might also like