You are on page 1of 7
Tamkang Journal of Science and Engineering, Vol. 12, No.1, p. 1-7 (2008) Estimating the Strength of Concrete Using Surface Rebound Value and Design Parameters of Concrete Material Department of Cwvil Engineering and Engineering Informatics, Chung-tua University Hein Chu, Titan 300, RO.C. Abstract This study estimates the strength of concrete, an attempt is also made to increase the accuracy of Calculating the strength, using the nondestructive test (NDT) surface hardness rebound value, material design parameters and regression analysis. The strength of the concrete specimens was 130-480 kglem’, and their ages were 7-38 days, In total, 166 standard specimens of conerete were grouped into 146 traning examples and 20 test examples to estimate concrete compressive strength Regression analysis was performed to establish a mathematical formula, Study results indicate that the corel son coefficient may reach 0.9622, indicating thatthe proposed value. Therefore, engineers may use this comprehensive concrete strength Key Words: Strength of Concrete, Rebound Value, Design Parameters, Regression Analysis 1. Introduction Concrete has significantly influenced the nature of engineering projects, Conerete, as a composite material, is generally composed of cement, sand, aggregate, water, ‘mineral admixtures and chemical admixtures Considerable work has been conducted to develop rapid, nondestructive tests (NDTS) that provide a repro- ducible measure of concrete quality in a structure [1] Unfortunately, as is usually the case in concrete testing, all these NDT generate results that are affected by vari= cous parameters such as aggregate type and size, age, mo- {shure content, and mix proportions [1]. Therefore, the correlation between measured properties and strength differs for various concretes and must be limited to the concrete in question, However, the NDTS are also conve nient and have been used for many years in quality man- agement of engineering materials, These tests are useful in determining the differences in concrete quality from *Comesponding author. F-malchkou@etu edu one part of a structure to another, Developed in Germany in 1930, the rebound hammer test (REIT), based on ASTM. (C805 and BS 4408 Part 4, can be utilized for testing con- crete surface hardness [1,2]. In 1948, Schmidt developed the Schmidt rebound hammer test [3,4]. This device is universally used because of a hardened steel hammer im- pacted on the concrete by # spring. The RHT is a con- venient NDT. The surface of hardened concrete is struck with the hammer, and concrete compressive strength is estimated vi the surface hardness rebound value. In 1979, the ASTM listed the rebound hammer testing method (ASTM C 805-79) asa standard testing method, explsin- ing that this method can be used to estimate the unifor- rity of concrete and detect areas of inferior quality within a concrete structure; however, it is not a substitute for concrete strength testing methods. The general view held by many users of the Schmidt rebound hammer is that it is useful in assessing concrete uniformity and in comparing fone concrete against another, bul can only be used as a rough indication of concrete strength in absolute terms (1) ‘When the RH is performed, kinetic energy from the 2 on-Chel Liv ota. impact and amount of lost kinetic energy allect the re- bound value. Typically, the amount of energy lost during contact between the pole and concrete must be deter- ‘mined Via the stress-strain relationship of the concrete therefore, rebound energy is correlated with the concrete strength and rigidity. However, the accuracy of RH need tobe improved in real applications when estimating con- crete strength using the surface rebound value. Low strength conerete will have a low rebound va- ue. However, when two concrete specimens have the same strength and different rigidities, the resulting re- bound values may not equal each other [5]. The amount of energy lost with low-rigidity eonerete is greater than that lost with high-rigidity concrete. The reason for this difference may be associated with material parameters, For instance, the amount of coarse aggregate and how aggregate is mixed in a concrete mixture affect the con- crete rigidity, thus affecting the rebound value. “Thus, this study analyzes the proportions in conere! mixtures. Design parameters are used as input data to create a rebound model, to enhance the accuracy of de- smining concrete strength, In the conventional material modeling process, re gression analysis is an important tool for constructing ‘model, In this study, seven design parameters, ie. amounts of cement, coarse aggregate, fine aggregate, slag, fly ash, chemical admixture, water-o-binder ratio (X1-7) and re= bound value of concrete, were used to build the regres sion formula, 2. Experimental Work Figure 1 presents the research flow chart. The com- pressive strength of concrete is estimated using the sta- tistical regression analysis model. Additionally, the root mean square of error (RMSE) and coefficient of detemaintion are used to determine ‘model liability, The esearch mothod ie as flows (0) The rebound value with concrete design parame- rs, using statistical regression to find the com pressive strength, Sutstical regression is used to identify compressive strength based on he rebound valu nd concrete design parameters Intoral, 146 specimens with diferent mix propor sions are made inthis stay, Different methods forthe HRT of standard cylinders (f=15 cm, 1. ~ 30 em) are categorized as follows. (1) Measure 20 distributed test points and calculate the average. (2) Take 5 points, pei form 4 measurements at each point, and calculate the av- erage. (3) Fora single point, perform HIRT 20 times and calculate the average, The largest mean rebound value in this study is the rebound value used as input data .The largest rebound value obtained using method (3) 20 tests are made atthe same point and calculate the average. In this model, 146 concrete specimens with different mixture proportions are used as raw data, Regression an- alysis is applied to the variables (Table 1). Table 1 lists the ranges of values for applied parameters, Regression analysis was utilized to construct a model estimating the concrete strength, (2) Model Confirmation ‘When the model was completed, 20 sets of test data are used to determine the accuracy of regression est Material mixture Rebound proportion parameter hammer test ¥ Regression analysis Model i ‘Statistical tests for ‘Compared with determination coefficient and RMS value I Confirmation J Estimated compressive strength Figure I. Research flow chart Estimating the Strongth of Concrete Using Surface Rebound valuo and Design Parameters of Concrote Matera «3 Table 1. Regression analysis output and input variables (The lower and upper bounds of each component) Variable E ponent Range of values Input variable Input variable X; Input variable X; Input variable X. Input variable Xs Input variable X« Input variable X; Input variable X¥ Input variable X, Input variable X yo Output variable Y ‘Age (ays) ‘Cement (kgim) Coarse aggregate (kg/m) Fine aggregate (kg/m") Slag (ke/m") Fly ash (kg/m?) ‘Chemical admisture (kg/m?) Water-to-binder ratio Moisture content (%) Rebound value (7) Compressive strength (kgfem’) 140-280 0807-1031 790-910 60-150 040-110 17-86, 0.45-0.62 07-38 0.015-0.060 12-30 130-480 sates. The absolute deviation values and pereentages are utilized for comparisons with true values. As the abso. Ite deviation values and percentages decreases, the de- agree of difference between the predicted compressive strength and true compressive strength decreases, and model accuracy inereases, Thus, the accuracy ofthe mode! is confirmed. 3. ndings and Analysis Input parameters employed in this sty inelude mo inure content, The influence of maisure conten on re- Bound vale is 99-80% (average 24%), The diferent mnehods forthe RHT applied to standard eylnders (15 em, L30 cr) area follows. For one point, apply the RT 20 simes and calulate the average. Tes results indicate that round valu increases by 20-70% when mlpetests remade onthe sam point, with method (3) (20 tests onthe sme point). Applied to all data coleted forthe 146 concrete samples, ther: soling regression (equation of Single Point eximaton) is as fllows y=23.085% - 145.02 a whores compressive strength xs the rebound valu and the corsaton coecient i 0.916, With method (3) (20 tests om the same point), cone eres compressive suonath vals diel due fom the single rebounding pint deviate by 1-53 average 11%) (strength error, 27.26 keffem’), The rebound value with designed parameters and seatisical regression ate used to estimate the compres- sive strength of concrete. In this model, statistical regres- sion analysis is applied to all data collected for the 146 concrete samples. The resulting regression is as follows: Y 237.66 ~0.695 (X1) ~0.292 (X2)- 0.501 (X3) = 0,530 (X4) = 1.117 (XS) + 1.013 (X6) 606.478 (X7) ~ 3.673 (XB) 30.994 (X9) = 12.887 (X10) @ where Y is concrete compressive strength and X1 is amount of cement; Table 1 presents X2-10, By employing the regression equation formulated in this study, which incorporates material design parame- ters such as age and moisture content, errors of estimated and actual values of concrete compressive strength, Ob- tained from the regression analysis reasoning formula and the accuracy ate high To test the results ofthe regression analysis, this ex- amines the RMSE as a criterion for evaluating the extent of data error in regression analysis. The RMSE forrmula RMSE = ® where T= the test output value of the jth order on the ith sam= ple (04~ the objective output value ofthe jth order on the ith sample n= sample number N= number of output variables 4 on-Chel Liv ota. Table 2 lists the parameter values obtained, by re- gression analysis for regressic results analysis for statistical regression. The F- value is 168.5738 and the P-value is 3,97E-71. The P-value for this model is far smaller than the standard level of sig. nificance (0 = 0.05); therefore, significance is excel- Tent, indicating that the regression line is valid. The correlation coefficient is 0.9622 for this model, and the coefficient of determination (R2) is 0.9258, RMSE is 24.088, and goodness of itis good. Thus, avery strong positive relationship exists between independent and de- pendent variables, Table 3 shows the correl input material paramet ingredients on conerete mixes and other input variables ‘on concrete compressive strength. A positive correlation sts between rebound value, age, cement, slag, chemi cal admixture, and fly ash amount; concrete compressive jon coefficients for each ‘Table 3 also lists the effects of Table 2. The residual error of regression analysis table strength is compared with correlation coefficients of re- gression analysis (Table 3). Table 3 also lists the effects of ingredients on concrete mixes and other input vari- ables on conerete compressive strength, Table 3 shows the test results of 20 repetitive measurements (of the con crete sample-Rebound value, age, cement, sag, fly ash and chemical admixture) are all positively correlated with concrete compressive strength, and have correlation co- efficients of 0.916, 0.738, 0.431, 0.362, 0.248 and 0.331, respectively, The water-to-binder ratio, moisture content, course ageregate and fine aggregate are all negatively correlated with concrete compressive strength, and have correlation coefficients of -0.455, -0.417, -0.296 and 0.289, respectively. The trend of correlation coefficients generally complies with the physical and chemical cha- racteristis of conerete with general aggregates, As too ‘many variables are included in this regression analysis, correlation coefficients are 0.455-0.916, Thus, this stu- Degrees offivodom Sumofequares Valuccofmean square ‘Fvalue —-Povalue Rages a10 978139. 9781395 165738 ——S97E-TI Residual eror 135 7830282 580.2439, Total 14s 1056472 ‘Remar The Fovelue and P-value are regression params of epression aalyis Table 3, onsaton cocticients for cach ingredient coment CO Fine Fly Chemical. 4. Moisture Rebound Compressive “Aggregate Aggregate ash Admixture "Ritio 2° Content Value Strength Come ——«.000 Course 0395 1,000 ‘Agarogste Fine Agateget 0673-0221 1.000 Slee 0457-0777 0186 1.000 FlyAsh 04590203. 0644 0.010 1.000 Chemical 0394-0227 0301 0.185 0.757 1.000 ‘Admintre Woterio- 0895 0382 0656 -0.558-0698 0.630 1,000 binder Ratio Age 0.109 0037 0.133 00017-0074 0.036 0.088 1,000 Moise 03990314 0270 0.388.075 0.299 04110157 1.000 Content Rebound Value 0479 0.335 0320 0.386 0261 0342 0.489 0.604 .0.452 1,000 Compressive 0431-0296 0289 0362 0268 0331-0455. 0738 -04IT 0916 1.000 Strength Estimating the Strongth of Concrete Using Surface Rebound value and Design Parameters of Concrete Maternal § dy takes the largest correlation coefficient as an example for discussion. A maximum positive correlation, 0.916, exists between rebound value and concrete compressive strength. Water-to-binder ratio, coarse aggregate, sand amount and concrete compressive strength of concrete are negatively correlated The correlation coefficient for rebound value is 0.9622, which is higher than correlation coefficients for other va- riables. Furthermore, in standardizing residual errors for rebound values (Figure 2), the distribution of residual er- rors has a central tendency, and no pattern exists in re- sidual errors. No problem of serial correlation exists: thus, the model is permissible. Figure 3 shows the correlation between compressive strength and rebound value; the coefficient of determina~ tion for this model is 0.9258, indicating that the differ- ence between expected values and estimated values is not large. The RMSE for this model is 24.1 kg/cm? ‘Additionally, from the normal distribution of eom- pressive strength estimated by sample percentages (Fig- ure 4), specify has an approximate straight-line graph witb litte variation. Figure 4 shows the accumulated per- centage for strengths of different specimens in 146 train- ing examples in the model developed in the study with actual compression strength values of 131-477 kgffcrs? Concrete compressive strength in this study approximates ‘anormal distribution; this is also confirmed from the fact that, with 20 test examples, en error percentage of 0- 26% exists when compressive strength is 148.9-402.5 kgflem? average erzor is only 5.43% and strength devia- tion is only 13.05 kgifem*. Thus, the data are approxi- ‘mately normally distributed. The difference between ex- pected values and estimated values is not large, suggest- ing thatthe regression analysis is valid. However, when the seven design parameters (X1-7), age (X8) and moisture content (X9) of conerete also used as input variables in this model, the correlation coefii« cient increased to 0.9622. Testing ‘Testing the statistical regression model By employing the regression equation formulated in this study, which incorporates material design parame- ters such as age and moisture content, ers of estimated and actual values of concrete compressive strength vary aL 0-26% (average, 5.43%) (Table 4), of which the accu- racy is more than double that estimated with 20 tests ap- plied to one point (equation (1)). . Rebound value Table 4 Compares the accuracy of regression esti- Figure 2. Standardization of residual error for the rebound tes with true compressive strength, Finally, this study a orznzes tester te teession model Based 08 torent the repesion eas etnate, ate ey a 4 901 [5 copcuceoe atts Fetlicem | allt” baw Eee siuit i ia jie 4 zal b z ¢ § Rebound Figure. Corelation between compressive strength and re- bound value igure 4, Normal distribution of compressive strength est- ‘mated by sample percentage. on-Chel Liv ota. Table 4. Compares the accuracy of regression estimates with true values Concrete compressive Regression formule Hem stcength (kgilem") estimates (katfem®) _APsolue efor value 1 Oat 2 Qos 3 (34s 4 (152 5 (99.2 6 (9303 1 (sa 8 09 9 10 ul 2 B 4 15 16 17 18 19 20 ‘Mean value 13.045 error mean values is 13.045 kgt/em?, absolute error per- centage is 5.43%, and the correlation coefficient i 0.9258, Obtained from the regression analysis reasoning formule and the accuracy thereby obtained are high. Study results show that the accuracy of regression formula is based on known parameters of material mix- ture proportions involving fly ash, slag and the chemical admixture. The range of estimating strength is 130480 kgf/em?, Thus, the established estimation model has bet- ter accuracy than the equation (1) The quality of the cement paste produced is pri 1e water-cement ratio marily an inverse funetion of (w/e). Restated, concrete strength is a function of total void content of the material [6]. The accuracy of the es- tablished estimation model is based on known parame- s of material mixture proportions, involving fly ash, slag and chemical admixture [7 Pozzolanie materials are known to enhance the uni= ormity of concrete. Adding fly ash to the mix introduces variable that increases paste quality and, consequently, overall concrete quality, Many studies have shown that when the water-binder ratio (w/b) is used instead of the ‘water-cement ratio as the basis for mix design, the accu racy of the strength prediction increases [8,9] 4, Conclusion and Recommendations The following conclusions are based on analyses and discussions. 1, Through multiple statistical regression analyses of estimating concrete compressive strength using the rebound value model, a eal scaling RMSE of 24.08, kefiem? was abtained, When the seven design pa rameters (XI-7), age (X8) and moisture content (X9) of concrete are input variables in this model, ‘the correlation coefficient increased to 0.9622. This indicates that combining the rebound hammer test value with design parameters of composite con- crete materials and using statistical regression an- alysis to estimate concrete strength has an average accuracy of within an error of 5.5%, ‘Simulation and test resulls of compound material ‘parameters processed by the statistical regression formula indicate that using statistical regression with the surface hardness rebound value, and de- sign parameters of concrete materials inereases the accuracy when estimating concrete strength, 3, Study results show that, regarding surface hard- ness rebound value (an NDT) and design para- Estimating tho Strongth of Concrete Using Surtaco Rebound value and Design Parameters of Concrete Material = meters of materials for estimating concrete strength, the accuracy of Calculating concrete strength is based on input design parameters of materials in- volved, such as water-binder ratio, fy ash, slag, chemical admixture, age, and moisture content References [1] Mindess, Sidney. and Young, JF, "Concrete Pretice- Hall," Ine, Englewood Cliffs N. Jp. 441 (1981). (2] Kumar, Meta, Pand Paulo, J. M., Monteiro “Con- crete Structure, Properties, and Methods,” p. 347 (1987), [3] Schmide. E,“The Conerete Test Hammer (Der Betom pruf hammer)” Schweiz Bauz (Zurich) Vol. 68, p.378 (1950) [4] Schmide, F., “The Conerete Szlerometer,” Proc, It, Symp, Nodestructive, Testing en Material and Struc- tures Vol. 2, RILEM, Paves, p. 310 (1954). [5] Lin, L-C., “The Test Conerete Strength the Non-De- struction Test Technology.” the Taiwan area mixes the conerete industry trade association published confer- ence proceeding in advanee, pp. 23-25 (1999), {6) Aitein, PC. and Neville, A., "High Performance Con- crete Demystified,” ACI Concrete International, Vol 15, ACI, pp. 21-26 (1993), (7) Yeh, 1-C., "Computer-Aided Design for Optimum Concrete Mixture,” Cement and Concrete Composites, Vol. 29, pp. 193-202 (2007), {8] Yeh, I-C., "Modeling Slump Flow of Conerete Using Second-Order Regressions and Artificial Neural Net- works,” Cement and Concrete Composites, Vol. 29, pp. 474-480. [9] Yeh, L-C, “Analysis of Strength of Concrete Using Design of Experiments and Neural Networks," Jour- nal of Materials in Civil Engineering, ASCE, Vol. 18, pp. 597-604 (2006). Manuscript Received: Apr. 24, 2007 Accepted: Jun, 27, 2008

You might also like