Available online at www.sciencedirect.

com

Construction and Building

Construction and Building Materials 23 (2009) 353363

MATERIALS
www.elsevier.com/locate/conbuildmat

Optimisation of using lightweight aggregates in mitigating autogenous deformation of concrete

Burcu Akcay *, Mehmet Ali Tasdemir
Faculty of Civil Engineering, Istanbul Technical University, Maslak 80626, Istanbul, Turkey Received 9 June 2007; received in revised form 20 November 2007; accepted 25 November 2007 Available online 11 January 2008

Abstract Experiments were conducted to determine the eects of using dispersed saturated lightweight aggregates (LWAs) as water reservoirs in mitigating the autogenous deformation of high performance concrete and to establish the optimum solutions as a combination of a number of factors aecting the fracture and mechanical characteristics of concrete. For this purpose, in concretes prepared with a constant low water to cement ratio, normal aggregates were replaced by natural LWAs with size fractions of 24 mm or 48 mm at three dierent volume fractions such as 10%, 20% and 30% of the total aggregate volume of concrete. The results indicate that the inclusion of ne fraction of LWAs in concrete reduces the autogenous deformation signicantly compared to that of the coarse fraction. It is also shown that concretes with ne fraction of LWAs have enhanced fracture and mechanical properties compared to those with coarse fraction of LWAs. Increasing the replacement ratio of LWAs mitigates autogenous deformation, while having an unfavourable eect on fracture and mechanical properties of concrete for both size replacements. A multi-objective simultaneous optimisation technique, in which the response surface method (RSM) is incorporated, is used to optimise the mitigation ratio of autogenous deformation and fracture parameters of high strength concretes in an eort to obtain a more ductile concrete with less autogenous deformation. 2007 Elsevier Ltd. All rights reserved.
Keywords: Autogenous deformation; Characteristic length; Fracture energy; Internal curing; Lightweight aggregate; Optimisation; Response surface method

1. Introduction In recent years, early age autogenous deformation and self-desiccation have been considered major problems in new generation high performance and ultra high performance concretes (HP/UHPCs). Autogenous deformation is dened as the self-generated bulk deformation of a closed, isothermal, cementitious material system not subject to external forces [1,2]. Since HP/UHPCs are produced using rich mixes with low water/cement ratio with the addition of high-range water reducers and ultra-ne pozzolanic materials such as silica fume, the material becomes very dense and more homogeneous than normal strength concrete. The amount of water in such concrete will be insuf*

Corresponding author. Tel.: +90 212 2853770; fax: +90 212 2856587. E-mail address: akcayburcu@hotmail.com (B. Akcay).

cient to complete the hydration process, and also the water ingress will be prevented because of the impermeable character of concrete [3,4]. Thus, high strength cementitious materials may exhibit substantial autogenous deformation which has a tendency of cracking at the early stages of hardening. Hence, the matter of utilization of autogenous deformation draws attention nowadays. One of the methods of mitigating autogenous deformation is internal curing which supply water to cementitious materials to keep the small, newly occurred pores under saturated condition. The use of porous and absorbent lightweight aggregates (LWAs) as water reservoirs was rst proposed by Philleo [5], and the method of internal curing of cementitious materials by LWAs has been investigated in a number of recent studies [3,615]. Previous studies have shown that LWA is bonded to the matrix stronger than normal aggregate, producing an

0950-0618/$ - see front matter 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.conbuildmat.2007.11.015

354

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363

enhanced interfacial transition zone (ITZ) [16,17]. This zone therefore prevents the length changes developed under relative humidity (RH) and/or thermal conditions [18]. The interface between the LWA and cement paste is tight and characterized by a mechanical interlocking combined with a chemical interaction in the form of pozzolanic reaction [19]. It is the internal curing that causes the water/cement ratio to increase locally in the paste-aggregate transition zone. The hydration products will be in a greater proportion in this zone than at some distance from the aggregate surface. This results in lling the pores with newly formed hydration products, which might also cause the self-desiccation to occur concomitantly. It is well known that the porosity and permeability of concrete is highly dependent on the properties of ITZ, and also microcracks in the matrix phase have inuences on these properties of concrete. In application, concrete is usually designed on the basis of compressive strength theory, in which the brittleness of concrete is ignored. For this reason, fracture parameters of concrete have to be dened along with its brittleness. In reality, concrete is always subject to crack development under the inuence of shrinkage for instance. This makes the inuence of crack and its size an important factor on concrete fracture [20]. Since the tensile strain capacity of concrete at early ages is not sucient to resist stresses caused by the external and internal restraints, autogenous deformation may lead to the formation of some microcracks. Since LWAs produce more microcracks, the concrete containing these aggregates behaves in a more ductile manner (i.e. less brittle) than normal cementitious materials [4,15,21]. On the other hand, the use of LWA instead of dense aggregate is expected to result in a lower strength. Thus, the amount of LWA for internal curing must be optimised according to the mechanical and fracture properties of cementitious materials. This study aims to determine the eect of the variations in both the volume fractions and the size of the saturated LWAs to mitigate autogenous deformation, which alters the fracture and mechanical properties of concrete, and to provide the optimum solutions as a combination of a number of factors aecting these properties of concrete. The eects of volume fraction and average particle size of LWAs on the loaddisplacement at mid-span curve were investigated by measuring the fracture energy and nal displacement. A multi-objective simultaneous optimisation technique, in which the response surface method (RSM) is incorporated, was used to estimate the optimum solutions as a combination of a number of factors by tting a polynomial model by using analysis of variance (ANOVA). The results were used to provide the optimum solution of the size and volume of LWAs by maximizing mechanical and fracture properties together with the mitigation ratio of autogenous deformation relative to the reference concrete that contains normal weight aggregate only.

2. Experimental details 2.1. Materials and mixture design Seven concretes were prepared using constant water to binder ratio of 0.28 with addition of 10% silica fume (by weight of cement) and a constant aggregate volume fraction of 0.65 m3/m3. In all concretes was used an ordinary Portland cement (OPC 42.5), of which 7- and 28-day compressive strengths of the standard RILEM-Cembureau cement mortars were 40.0 MPa and 50.6 MPa, respectively. The specic gravity of the silica fume (with 93.7% SiO2; stated in producers report) was 2240 kg/m3. The silica fume was used in dry condition and mixed with cement before the water addition. Crushed limestone with maximum size fractions of 8 mm was used as coarse aggregate, while natural quartz sand and siliceous powder were used as ne aggregates. Eective water absorption of the coarse aggregates (according to ASTM C127) and ne aggregates (according to ASTM C128) were 1.1% and 2.1%, respectively. Particle densities of crushed limestone, crushed sand, quartz sand, and siliceous powder were 2720, 2710, 2650 and 2620 kg/ m3, respectively. The aggregate grading and the maximum particle size of aggregate were also kept constant during preparation. Normal aggregates were replaced by natural LWAs with the size fractions of 24 mm or 48 mm at three dierent volume fractions of 10%, 20% and 30% of the total aggregate volume of concrete. Crushed natural pumice from the Cappadocia area, Central Turkey, was used as LWA. The particle densities (both oven dried and 24-h saturated surface dry (SSD) conditions) and water absorptions of the pumice LWAs were determined according to DIN 4226 (Table 1). Pre-soaked water content was determined using the 30 min water absorption. A commercially available polycarboxylic ether based superplasticizer (Glenium 51 from BASF-YKS) was used to adjust the slump of fresh concrete in the range of 160 20 mm. The values of slump were chosen close to that of self-compacting concrete, because the mixtures with LWAs were intended not to be exposed to any strong vibration for compaction. Before the production of concrete, all LWAs were dried overnight. After cooling the LWAs at room temperature, pre-soak water is sprayed into the LWAs and the container of LWAs was kept closed for 30 min to obtain the water saturated surface dry condition. The LWAs were mixed three times for homogenization during this 30 min period. All mixtures were prepared using a 45 l capacity pan mixer with vertical rotation axis. Since the capacity of the mixer is smaller than the necessary volume of the mixtures, two 35 l identical batches were produced using the same materials. Two layers of polyethylene plastic were inserted inside the oiled steel moulds before casting. The concretes were then covered with these plastic sheets. All specimens were kept in their moulds for 24 h

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363 Table 1 Densities and water absorptions of pumice LWAs Size fraction (mm) Particle density (kg/m3) Oven dry 2/4 4/8 750 730 24 h SSD condition 1070 1020 Bulk unit weight (kg/m3) 410 400 Water absorption, % by dry weight 10 (min) 28 26 30 (min) 35 33 60 (min) 38 35

355

24 (h) 45 41

except for the specimens for autogenous deformation measurements. After de-moulding, each specimen were sealed using two layers of plastic sheets together with two layers of aluminium foil. The specimens were then stored in a room at a temperature of 20 2 C and an RH of 65%. The mixture design, designation codes and some of fresh properties of the concrete are given in Table 2. The numbers following the letter V and L in the designation code denote replacement volume and the replaced size fraction of LWAs, respectively. Exception, however, is the reference concrete (CREF) that contains normal weight aggregates only. 2.2. Test procedure 2.2.1. Autogenous deformation The linear autogenous deformations were determined by measuring displacements on the sealed specimens. In this study, autogenous deformations of concretes were measured on four prism specimens of 70 70 280 mm. Stainless steel studs were settled on specimens using a thin layer of adhesive after initial setting time of approximately 4 h and the side walls were de-moulded. The measurements were taken over 200 mm gauge length according to the ASTM C 41. The extensometer used in the measurements was equipped with a comparator with an accuracy of 0.001 m/m. Specimens of each series were weighted at one and 28 days to monitor whether the sealing layers preserved RH in the specimens. Weight changes for the specimens were not greater than 0.08%.

2.2.2. Mechanical tests The mechanical and fracture tests were all conducted at the age of 28 days. Compression test was performed to obtain the elastic modulus and compressive strength. At least ve cylindrical samples with 100 mm diameter and 200 mm height were tested for each series. Static modulus of elasticity was calculated from the ascending part of the stressstrain curve for stresses up to approximately 33% of the ultimate strength. Standard compression tests were conducted in accordance with the European Standards (EN 206 and EN 12390). Splitting tension tests were performed on disc specimens with 150 mm diameter and 60 mm height. Six specimens tested and the splitting tensile strength was obtained from maximum load using the following equation: fst 2P ph/ 1

where P, h and / are maximum load, height and diameter of disc specimen, respectively. The tests for determining the fracture energy were performed in accordance with the recommendation of RILEM 50-FMC Technical Committee [22]. The deection was measured using a linear variable displacement transducer (LVDT). The load was applied by a closed-loop testing machine (Instron 5500R) with a maximum capacity of 100 kN. The beams prepared for the fracture energy tests were 500 mm in length and 100 100 mm in cross-section. The notch to depth ratio (a/W) of specimens was 0.4 and the notches were formed using a diamond saw. The eective cross-section was reduced to 60 100 mm to accom-

Table 2 Compositional and some of the properties of the fresh mixtures (for 1 m3) CREF Cement (kg) Silica fume (kg) Water (kg) Pre-soaked water (kg) Superplasticizer (kg) Siliceous powder (kg) Natural sand (02) (kg) Crushed sand (24) (kg) Crushed stone (48) (kg) LWA 24 mm (kg) LWA 48 mm (kg) Air (%) Unit weight (kg/m3) 497 50 153 0 10 169 514 520 529 0 0 1.4 2442 CV10L24 496 50 153 17 10 169 513 344 529 48 0 1.5 2329 CV10L48 497 50 153 34 10 169 513 519 353 0 47 1.3 2345 CV20L24 496 50 153 51 10 169 513 173 529 97 0 1.2 2241 CV20L48 497 50 153 16 10 169 514 520 177 0 94 1.2 2200 CV30L24 496 50 153 31 10 169 513 0 528 145 0 1.3 2095 CV30L48 496 50 153 47 10 168 512 519 0 0 141 1.4 2096

356

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363

Linear Autogenous Deformation, microstrain (m/m)

modate large aggregates in more abundance, and the length of support span was 400 mm. The crack mouth opening displacement (CMOD) was used as a feedback control variable to obtain stable tests, and deections were measured simultaneously by means of an LVDT at mid-span. The fracture energy of concrete (GF) was determined using the area under the complete load versus displacement at mid span curve (W0) by the following expression [22]: GF W0 B W a mg S d L 0 2

200 100 0 0 -100 -200 -300 -400 -500 -600 4 8 12 16 20 24 28 Time, days CV30L24 CV30L48 CV20L24 CV20L48 CV10L24 CV10L48 CREF

Linear Autogenous Deformation, microstrain (m/m)

where B, W, S, L, a, m, d0, and g are the width, depth, span, length, notch depth, mass, nal deection of the beam and gravitational acceleration, respectively. At least ve specimens of each mixture were tested. The beams were loaded at a constant rate of 0.02 mm/min. The net exural strength was determined using the following equation assuming that there is no notch sensitivity, where Pmax is the maximum load. fnet 3 P max S 2 B W a 2 3

200 Time, days 0 0 -200 -400 -600 -800 60 120 180 240 300 360 CV30L24 CV30L48 CV20L24 CV20L48 CV10L24 CV10L48 CREF

The characteristic length of the material (lch), which is the measure of brittleness, is given by the expression [23]: lch EGF ft02 4

Fig. 1. Linear autogenous deformation of concrete samples: (a) for 28days and (b) for one year.

where ft0 is the direct tensile strength of concrete which was replaced by splitting tensile strength (fst) in this study. Some previous studies showed that the characteristic length from ctitious crack model can be used for determining the brittleness [24,25]. In high strength concrete with normal aggregate, the characteristic length decreases with increasing compressive strength of concrete and this may indicate that the material becomes more brittle. The characteristic length of a semi-lightweight concrete (i.e. normal aggregates were partially replaced by LWA), however, is higher than that of high strength concrete with normal aggregate [4], meaning that the material becomes less brittle. 3. Results and discussion 3.1. Autogenous deformation The eect of both size and volume of LWAs on linear autogenous deformation of concretes for 28 days and one year are shown in Fig. 1a and b, respectively. It is shown that the linear autogenous deformation of concretes can reach signicant values in a few days after production. The results obtained indicate that while the linear autogenous deformation of reference concrete (CREF) was 537 microstrain (lm/m) at 28 days, substituting the 2 4 mm size fraction of normal aggregate concentration by LWAs at a rate of 10% reduced the autogenous deformation to 410 microstrain (lm/m). On the other hand, when

the replacement ratio was 20%, the swelling has been observed to occur at the early ages. A possible reason for this is that the pore radius of cement paste at early ages is larger than at the later ages; and hence internal curing water can easily travel into the cement paste at early ages. With ongoing hydration the possible travel distance will be decreased, and the newly occurred smaller pores will not nd enough water to counteract their RH loss. As a result, the capability of LWAs in mitigating autogenous deformation decreases with time, because of the enhancing impermeable character of cement paste. The pores through which the LWAs transport water are getting smaller with the hydration products continue to form. This causes the shrinkage to occur again in concretes with LWAs. In addition to the eects of autogenous shrinkage, the autogenous swelling can also be harmful for cementitious materials. Thus, the amount of replacement of LWAs to mitigate autogenous deformation should be determined with necessary optimisation. When the normal aggregate is replaced by pumice LWAs with 24 mm size fractions, the linear autogenous deformation of concrete at 28 days was mitigated by 24%, 47% and 92%, for volume fractions of 10%, 20% and 30%, respectively. The use of coarse LWAs (48 mm size fractions), however, prevents autogenous deformation of concrete by 11%, 32% and 65% at the same volume fractions. Thus, the results have shown that the ne LWAs are more eective than the coarse LWAs in reducing the autog-

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363

357

enous deformation of concrete. A similar behaviour is also seen in a-year autogenous deformation results (Fig. 1b). One of the most important parameters on autogenous deformation is probably the distribution of internal curing reservoirs in a cementitious material. The eciency of the internal curing is inuenced by the distribution of LWAs as well as their size and volume. Because LWAs can release water approximately 1mm of its outer surface [11,26], the cement paste in these water entrained area will be well cured and the autogenous deformation will be mitigated. It is obvious that the thickness of the water-released layer surrounding the LWA particle is largely dependent on the pore volume and structure of LWA as well as the permeability of cementitious materials. Fine size LWAs with lower aggregate spacing can be used to provide more eective internal curing [6,27,28]. 3.2. Mechanical properties The mechanical test results and fracture parameters of produced concretes are summarizes in Table 3 and are described in the following paragraphs. Since the modulus of elasticity of LWA is lower than that of normal aggregate, the compressive strength of concretes (fc) with LWAs decreased with increasing the particle size and volume of the substituted LWA. A slight decrease in the modulus of elasticity (E) was recorded with increasing the replacement volume of the LWAs, although the change in the size of LWA had no signicant eect on the modulus of elasticity of concrete as expected [4]. As presented in Table 3, the size and volume concentration of LWAs had eects on exural (fnet) and splitting tensile strengths (fst) in a way similar to that observed on the compressive strength results. The use of ne LWA to prevent autogenous deformation of concrete seems more eective than that of coarse LWA in terms of the mechanical properties of concrete. The eect of the size of LWA on exural and also compressive strength can be modeled in that LWAs in concrete phase behave as a defect [15,29]. 3.3. Fracture energy (GF) and characteristic length (lch) The test results have shown that the most signicant eect of using LWA was observed on the fracture energy

of concrete. As seen in Fig. 2 and Table 3, in cases of using both ne and coarse fractions, the area under the loaddisplacement curve decreased with increasing the replacement volume of LWA. Although the slope of the ascending branch of the curves remained almost constant, the measured nal displacement at mid-span increased signicantly with using LWAs. When normal aggregates were substituted with LWAs by volume of 10%, 20% and 30%, concretes with ne fraction of LWAs had greater peak load and also the area under the curve increased with respect to that of the coarser fractions (Fig. 2). Although the ultimate load did not vary signicantly, longer tail was an indication of mean particle size of LWA. The nal displacement increased signicantly with the replacement

b
Load, kN

6 5 4 3 2 1 0 0.0 0.2 0.4 0.6 0.8 1.0 Displacement, mm CREF CV10L48 CV20L48 CV30L48

Fig. 2. Typical load versus displacement curves for concretes contained (a) ne LWAs with the size fraction of 24 mm and (b) coarse LWAs with the size fraction of 48 mm. The results were obtained by three point bending test.

Table 3 Fracture and strength properties of concretes CREF Compressive strength (MPa) Modulus of elasticity (GPa) Area under loaddisplacement curve (N/mm) Fracture energy (N/m) Final displacement at mid-span (mm) Pmax (kN) Net exural strength (MPa) Splitting tensile strength (MPa) Characteristic length (mm) 93 46 508 85 0.313 5.28 8.6 6.7 86 CV10L24 79 40 563 101 0.609 4.34 7.1 6.4 98 CV20L24 62 37 551 95 0.687 3.67 6.0 4.8 155 CV30L24 50 32 432 77 0.618 2.94 4.4 4.6 118 CV10L48 68 40 451 84 0.699 3.81 6.1 5.6 107 CV20L48 61 37 422 77 0.671 3.03 5.0 5.0 115 CV30L48 48 27 322 59 0.475 3.01 4.8 4.1 95

358

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363

volume of LWA, although after 10% replacement of LWA, the nal displacement at mid-span began to decrease. This is probably due to the fact that there is not enough amount of stier normal aggregate, around which the cracks could progress; thus the cracks can easily penetrate into the LWAs and matrix phases. The characteristic length, which is calculated using Eq. (4), increased with the use of LWA, indicating that the material becomes more ductile. The brittleness of concrete diminished with decreasing the substituted mean particle size of LWA. In Fig. 3, the variations in the size of LWA with fracture energy and characteristic length are graphically illustrated. In particular, replacement of normal aggregate by ne LWAs by volume of 10% caused the fracture energy to have increased by 19% with respect to the reference concrete. However, using the coarse fraction at the same volume, although caused a decrease in the peak load, did not change the fracture energy signicantly when compared to the reference concrete. The use of LWAs in concrete had an eect of increasing the characteristic length of the material in all size and volume fraction of LWAs. Since LWAs produce more microcracks, the concrete containing these aggregates behaves more ductile (i.e. less brittle) than the reference concrete. The decrease in the brittleness was more signicant in concretes with ne LWAs. It can be seen in Fig. 3 that substituting normal aggregates by LWAs by up to 30% prevented the autogenous deformation of reference concrete signicantly, although this practice caused also a marked loss in the mechanical properties of cementitious materials. However, the negative eect of this application on the mechanical properties of concrete is shown to be smaller in case of using ne LWAs. It should therefore be noted that in order to mitigate autogenous deformation and cracking in early ages, the volume and size of the LWAs should be optimised with taking the mechanical properties into account.

4. Optimum design of concrete The experimental programs detailed in Section 2.1 have also been designed to nd the optimum solutions as a combination of a number of factors estimated by tting a polynomial model by using analysis of variance (ANOVA). The compressive strength, modulus of elasticity, fracture energy, characteristic length, splitting tensile strength, net exural strength, and autogenous deformation of concretes have been dened according to the size and volume of substituted LWAs. A multi-objective simultaneous optimisation technique is used by incorporating the response surface method (RSM). 4.1. Response surface modelling Since concrete has a non-linear behaviour and there are some microstructural parameters and mechanical properties, the non-linear optimisation problems should be resolved. The response surfaces analyze the eect and signicance of each variable and the RSM has been used by several researchers to nd optimum solutions in cement and concrete industry [3034]. In application, the response surfaces of experimental data are rst analyzed and then the model which ts best the data is established and nally the eciency of the model is determined [30,31]. When the empirical relationship between one dependent and at least two independent variables exists, the whole system can be simultaneously represented by a response surface. In this analysis, the independent variables are volume fraction (volume) and the average size fraction (size) of LWAs, while the dependent responses are compressive strength (fc), modulus of elasticity (E), fracture energy (GF), characteristic length (lch), splitting tensile strength (fst), exural strength (fnet), and autogenous deformation. The tested ranges of the independent responses are as follows: 6:5% 6 volume 6 19:5% 3 mm 6 size 6 6 mm 5

120

200

-410 m/m

160

80

GF , N/m

-285 m/m -40 m/m -537 m/m

-360 m/m -480 m/m -183 m/m

120 lch, mm

80

40 40

Here, the volume of LWAs is considered to be represented by the ratio of volume of total LWA content to the total volume of concretes. These ratios are indicated in the designation codes of concretes. Thus, it should be noted that, for example, the value of LWAs volume of 6.5% is equal to the 10% replacement value which is the ratio of LWAs to the total aggregate content. Using two independent variables the quadratic model for each response can be established using Eq. (6) where the b0, b1, b2, b3, b4, and b5 are the coecients obtained from ANOVA. y b0 b1 volume b2 size b3 volume2
2

GF- 10% lch- 10% 0 FINE

GF- 20% lch- 20% REF

GF- 30% lch- 30%

0 COARSE

b4 size b5 volume size

Fig. 3. Eect of the size of LWA on both fracture energy and characteristic length. The numbers next to the data points represent the linear autogenous deformation values at 28-days.

It is possible to simplify the quadratic model using a backward stepwise technique if the insignicant terms are to be ignored.

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363

359

4.2. Regression analysis For the six concrete series with LWA, a polynomial type of mathematical model for each parameter of experimental data was established. Model tting results for each response are given in Table 4. For example, the tted regression model for mitigation ratio of autogenous deformation (% ratio of autogenous deformation of internally cured concrete to that of the reference concrete at 28 days) is as follows: mitigation ratio of autogen: def : 1:08 volume 8:05 size 0:2 volume
2

close to those obtained from experiments. The results show that the predictiveness of the polynomial regression model is satisfactory. Response surfaces of fracture energy, compressive strength, characteristic length, splitting tensile strength, modulus of elasticity, net exural strength, and the mitigation ratio of autogenous deformation at 28 days have been obtained and some examples of responses used in optimisation are shown in Figs. 47. 4.3. Optimisation Optimisation naturally requires simultaneous assessment of several responses. Thus, in order to optimise the objective functions, the regression model and relationships between design variables and the responses are established (Table 4). For each response, the desirability functions (dj) are obtained and these functions are optimised simultaneously to nd the best settings using a numerical optimisation technique [35]. A desirability function (dj) varies over the range of 0 6 dj 6 1 and each response can be mapped onto the function. In order to establish a solution of multi-objective optimisation, the single composite response (D) is calculated from the following equation: D d 1 d 2 d 3 . . . d n n
1

7
2

In this case, (volume), (size), and (volume) are the only signicant terms. Since the LWAs were added to the reference concrete to mitigate autogenous deformation of this concrete, the eects of CREF series were taken into account in tting analysis. It is obvious that the b0 terms of each response belong to the experimental results of reference concrete. For each response, the r value of regression model has been found greater than 0.94. The comparison between the results obtained from both regression models and experiments are given in Table 5. It has been shown that the values obtained from model calculations are very
Table 4 Model tting results for each response Responses Coecients b0 fc E GF lch fst fnet Mitigation ratio of autogenous deformation at 28 days 93 46 85 86 6.7 8.6 0 b1 2.77 1.15 1.66 22.77 0.33 0.38 1.08 b2 1.69 5.14 8.61 44.33 0.75 0.46 8.05

Correlation coecient r b3 0.006 0.05 0.14 0.72 0.007 0.002 0.2 b4 0.69 0.70 1.61 5.44 0.11 0.12 b5 0.23 0.123 0.82 0.0077 0.036 0.9967 0.9953 0.9998 0.9389 0.9780 0.9937 0.9993

Table 5 The results of experimental and model tting of each concrete fc CV10L24 CV20L24 CV30L24 CV10L48 CV20L48 CV30L48 Experimental Fitted Experimental Fitted Experimental Fitted Experimental Fitted Experimental Fitted Experimental Fitted 79 78 62 64 50 49 68 69 61 59 48 49 E 40 40 37 39 32 33 40 41 37 37 27 29 GF 101 101 95 95 77 77 84 84 77 77 59 59 lch 98 103.5 155 144 118 123.5 107 101.5 115 126 95 89.5 fst 6.4 6.3 4.8 5.1 4.6 4.5 5.6 5.7 5.0 4.7 4.1 4.2 fnet 7.1 7.2 6 5.8 4.4 4.5 6.1 6 5 5.2 4.8 4.7 Mitigation ratio of autogenous deformation 24 23.2 47 48.7 92 91.2 11 11.8 32 30.3 65 65.8

360

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363

Fig. 4. Variation of compressive strength fc with volume and size of LWAs.

Fig. 6. Variation of splitting tensile strength fst with volume and size of LWAs.

Fig. 5. Variation of characteristic length lch with volume and size of LWAs.

Fig. 7. Variation of the value of the mitigation ratio of autogenous deformation according to the CREF series (at 28-days) with volume and size of LWAs.

where D is the geometric mean of the individual desirability functions and n is the number of responses. D is then maximized over the feasible region of the design variables given in Eq. (5) [36]. For maximizing and minimizing each response, the desirability function is obtained using Eqs. (9) and (10), respectively,

dj

8 0 > > <h > > :

Y j min fj max fj min fj

iwtj

Y j 6 min fj and 0 < d j < 1; min fj < Y j < max fj Y j P max fj 9

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363

361

dj

8 1 > > <h > > : 0

max fj Y j max fj min fj

iwtj

Y j 6 min fj and 0 < d j < 1; min fj < Y j < max fj Y j P max fj 10

where dj, Yj, min fj and max fj are the desirability function, the tted value, and minimum and maximum actual (experimental) values of jth response included in the optimisation, respectively. The power value wtj is weighting factor of the jth response. In this work, for optimisation, a commercially available Design-Expert software package was used. In order to nd out an optimum solution for a concrete with minimum autogenous deformation, maximum

a
0.340

0.270

0.200

0.130

0.060

6.00 5.25 4.50 13.0 3.75 3.00 6.5 9.8 16.3

19.5

strength and less brittleness, a simultaneous maximization of characteristic length (lch), splitting tensile strength (fst), compressive strength (fc), and mitigation ratio of autogenous deformation is required. The three responses lch, fc and fst are considered to be of equal importance, i.e. wtj = 1, while the response of the mitigation ratio of autogenous deformation is considered to be of greater importance, i.e. wtj = 2. The solution of this multi-objective optimisation gives the optimal values of design variables as follows volume: 13.1% and size: 3 mm (Fig. 8). The predicted response values are as follows: the mitigation ratio of autogenous deformation = 49.4%, lch = 115 mm, fst = 5.1 MPa, and fc = 61.1 MPa. In above optimisation, the response values of CREF reference concrete have been ignored because of the ranges given in Eq. (5), although in quadratic tting model they have been taken into account. In the rst situation, eect of LWAs on toughening mechanism has also been discarded. For example, the characteristic length of CV10L24 series is signicantly higher than that of CREF series. If the CREF is included as a design point, it is obvious that the ranges of 0% < volume < 6.5% and 0 mm < size < 3 mm will be included in the optimisation. However, CREF has been inserted in the design layout as a design point. In this situation, the optimal values of design variables have been found as volume: 17% and size: 2.2 mm. The predicted response values are: mitigation ratio of autogenous deformation = 49.5%, lch = 117 mm, fst = 5.4 MPa, and fc = 67.4 MPa as shown in Fig. 9. Since the prevention of autogenous deformation is being more important for the high strength and high performance concrete, the response of autogenous deformation is considered to be at the maximum importance (i.e.

Desirability

B: size

A: volume

Desirability
6.00
0.11 4 0.15 9

0.5

0.4

Desirability

5.25

0. 203

0.2

0. 248

B: size

0. 293

0.24 8 0. 203

0.1

4.50

0.0

3.75
6.00 19.5 4.50 3.00 9.8 1.50 0.00 0.0 4.9 14.6

Prediction 0.338
3.00 6.5 9.8 13.0 16. 3 19.5

B: size

A: volume

A: volume

Fig. 8. (a) Response surface and (b) contour plot of the composite desirability (D) when the mitigation ratio of autogenous deformation, lch, fst, and fc are maximized simultaneously.

Fig. 9. Response surface plot of the composite desirability (D) when mitigation ratio of autogenous deformation (with wtj = 2), lch, fst, and fc are maximized simultaneously.

362

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363

0.200

0.150

Desirability

0.100

deformation of concretes can be obtained by using the variable volume and size of LWAs. After analyzing the data obtained from the experiments, a decision might be made to perform another experimental program to obtain the desired responses. As a further study, a wider range of variables can be used, and the concretes with dierent water to cement ratios can be investigated by using optimisation based mixture design. Acknowledgements

0.050

0.000

6.00 4.50 3.00 19.5 14.6 1.50 4.9 0.00 0.0 9.8

This research was carried out in the Faculty of Civil Engineering at Istanbul Technical University. The authors acknowledge the Grant from DPT (State Planning Organization, Project: 2003K120630). References
[1] Jensen OM, Hansen PF. Autogenous deformation and RH-change in perspective. Cem Concr Res 2001;31:185965. [2] Jensen OM. Autogenous phenomena in cement-based materials. Aalborg University; 2005. [3] Weber S, Reinhardt HW. A new generation of high performance concrete: concrete with autogenous curing. Adv Cem Based Mater 1997;6:5968. [4] Tasdemir MA, Tasdemir C, Grimm R, Ko nig G. Role of aggregate fraction in the fracture of semi-lightweight high strength concrete. In: Ko nig G, Dehn F, Faust T, editors. Proceedings of the 6th international symposium on utilization of high strength/high performance concrete, Leipzig; 2002. p. 145366. [5] Philleo R. Concrete science and reality. Material science of concrete II. Westerville: American Ceramic Society; 1991. p. 18. [6] Bentz DP, Snyder KA. Protected paste volume in concrete extension to internal curing using saturated lightweight ne aggregate. Cem Concr Res 1999;29:18637. [7] Kohno K, Okamoto T, Isikawa Y, Sibata T, Mori H. Eects of articial lightweight aggregate on autogenous shrinkage of concrete. Cem Concr Res 1999;29:6114. [8] Takada K, Van Bruegel K, Koenders EAB, Kaptijn N. Experimental evaluation of autogenous shrinkage of lightweight aggregate concrete. In: Tazawa E-I, editor. Proceedings of international workshop autoshrink98, Japan. London: E&FN SPON; 1999. p. 22939. [9] Bentur A, Igarashib S-I, Kovler K. Prevention of autogenous shrinkage in high-strength concrete by internal curing using wet lightweight aggregates. Cem Concr Res 2001;31:158791. [10] Zhutovsky S, Kovler K, Bentur A. Autogenous curing of highstrength concrete using pre-soaked pumice and perlite sand. In: Persson B, Fagerlund G, editors. Self-desiccation and its importance in concrete technology proceedings of the 3rd international research seminar in Lund; 2002. p. 16174. [11] Lura P. Autogenous deformation and internal curing of concrete. PhD Thesis, Technical University of Delft, Netherlands; 2003. [12] Geiker MR, Bentz DP, Jensen OM. Mitigating autogenous shrinkage by internal curing in high-performance structural lightweight concrete. In: Ries JP, Holm TA, editors. SP 218. American Concrete Institute; 2004. p. 14348. [13] Lura P, Bentz DP, Lange DA, Kovler K, Bentur A. Pumice aggregates for internal water curing. In: Proceedings of the international RILEM conference on the advances in concrete through science and engineering, Evanston, IL; 2004. p. 13751. [14] Lam H, Hooton RD. Eects of internal curing methods on restrained shrinkage and permeability. In: Persson B, Bentz D, Nilsson L-O, editors. Self-desiccation and its importance in concrete technology proceedings of the 4th international research seminar in Lund; 2005. p. 21028.

B:size

A: volume

Fig. 10. Response surface plot of the composite desirability (D) when mitigation ratio of autogenous deformation (with wtj = 10), lch, fst, and fc are maximized simultaneously.

wtj = 10) and the optimisation was repeated. The solution of this multi-objective optimisation gives the optimal values of design variables as follows volume: 17.3% and size: 3.13 mm (Fig. 10). The predicted response values are: mitigation ratio of autogenous deformation = 79.6%, lch = 116 mm, fst = 4.84 MPa and fc = 57.9 MPa. 5. Conclusions Experimental results have shown that the use of presoaked pumice LWAs is eective in mitigating autogenous deformation of concrete. The replacement of normal aggregate by LWA even in as small volume as 10% can play a signicant role in decreasing the linear autogenous deformation of concrete. In particular, the replacement of normal aggregate by ne fraction of LWA has been shown to be more eective in reducing the autogenous deformation compared to the use of the coarse fraction of LWAs. The compressive strength of concrete has been found to be inversely proportional to the substituted volume of LWAs. The compressive strength of cement paste with ne fraction of LWAs is greater than that of the coarser fractions. The use of LWAs in concrete has an eect of increasing the ductility of material which can also be clearly observed at the load versus displacement at mid-span graphs. As the mean particle size of the LWAs decreases, the specic fracture energy, exural strength and the displacement at mid-span curve increase. It has been shown that the amount of microcracks is governed by the low particle size of aggregate substituted, and a longer tail is an indication of more microcracks. The possible optimum solutions for desired mechanical and fracture properties and also the value of autogenous

B. Akcay, M.A. Tasdemir / Construction and Building Materials 23 (2009) 353363 [15] Akcay B, Pekmezci BY, Tasdemir MA. Utilization of articial lightweight aggregates in hardened cement paste for internal water curing. In: Balazs GL, Borosnyoi A, editors. Proceedings of the b keep concrete attractive, Budapest; 2005. p. 37480. [16] Sarkar S, Chandra S, Berntsson L. Interdependence of microstructure and strength of structural lightweight aggregate concrete. Cem Concr Comp 1992;14:23948. [17] Swamy RN, Lixian W. The ingredients for high performance. In: Holand I et al., editors. Proceedings of international symposium on structural lightweight aggregate concrete. Oslo: Norwegian Concrete Association; 1995. p. 62839. [18] Kayali O, Haque MN, Zhu B. Drying shrinkage of bre-reinforced lightweight aggregate concrete containing y ash. Cem Concr Res 1999;29:183540. [19] Zhang MH, Gjorv OE. Microstructure of the interfacial zone between lightweight aggregate and cement paste. Cem Concr Res 1990;20:6108. [20] Hu XZ, Wittman FH. Size eect on toughness induced by crack close to free surface. Eng Frac Mech 2000;65:20911. [21] Akcay B, Tasdemir MA. Eects of lightweight aggregates on autogenous deformation in concrete. In: Konsta-Gdoutos M, editor. Measuring, monitoring and modeling concrete properties. Springer; 2006. p. 16370. [22] RILEM 50-FMC. Determination of fracture energy of mortar and concrete by means of three-point bend tests on notched beams. Mater Struct 1985;18:28590. [23] Hillerborg A. Theoretical basis of method to determine fracture energy GF of concrete. Mater Struct 1985;18:2916. [24] Tasdemir MA, Karihaloo BL. Eect of aggregate volume fraction on the fracture parameters of concrete: a meso-mechanical approach. Mag Concrete Res 2001;53:40515. [25] Lange-Kornbak D, Karihaloo BL. Design of ber reinforced DSP mixes for minimum brittleness. Adv Cem Mater 1998;7:89101.

363

[26] Akcay B. Eects of lightweight aggregates on autogenous deformation and fracture of high performance concrete. PhD Thesis, Istanbul Technical University, Istanbul; 2007. [27] Akcay B, Tasdemir MA. Inuence of lightweight aggregates on internal curing and fracture of concrete. In: Jensen OM, Lura P, Kovler K, editors. Proceedings of the international RILEM conference, volume changes of hardening concrete: testing and mitigation, Lyngby, Denmark, PRO 52; 2006. p. 3140. [28] Kovler K, Jensen OM. Novel technologies of concrete curing. Concrete Int 2005;27(9):3942. [29] Tasdemir MA. Elastic and inelastic behaviour of structural lightweight aggregate concrete. PhD Thesis. Istanbul Technical University, Istanbul; 1982 [in Turkish with English summary]. [30] Bayramov F, Tasdemir C, Tasdemir MA. Optimisation of steel bre reinforced concretes by means of statistical response surface method. Cem Concr Comp 2004;26:66575. [31] Karihaloo BL, Lange-Kornbak D. Optimisation technique for the design of high-performance bre-reinforced concrete. Struct Multidiscip Opt 2001;21(1):329. [32] Soares RC, Mohamed A, Venturini WS, Lemaire M. Reliability analysis of non-linear reinforced concrete frames using the response surface method. Reliab Eng Syst Safe 2002;75:116. [33] Grabiec AM, Piasta Z. Study on compatibility of cement-superplasticiser assisted by multicriteria statistical optimisation. J Mater Proc Technol 2004;152:197203. [34] Duprat F, Sellier A. Probabilistic approach to corrosion risk due to carbonation via an adaptive response surface method. Prob Eng Mech 2006;21:20716. [35] Derringer G, Suich R. Simultaneous optimization of several response variables. J Qual Technol 1980;12(4):2149. [36] Myers RH, Montgomery DC. Response surface methodology: process and product optimization using designed experiments. New York: John Wiley & Sons; 2002.