Professional Documents
Culture Documents
StatisticalModelingofHydraulicandMechanicalPropertiesofPerviousConcreteUsingNDTs PDF
StatisticalModelingofHydraulicandMechanicalPropertiesofPerviousConcreteUsingNDTs PDF
net/publication/323858796
CITATIONS READS
0 652
3 authors:
Norbert Delatte
Oklahoma State University - Stillwater
114 PUBLICATIONS 975 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Xuhao Wang on 19 March 2018.
Abstract: Because of its lower mechanical properties and durability compared to normal concrete, pervious concrete demands better quality
Downloaded from ascelibrary.org by Kamran Amini on 03/16/18. Copyright ASCE. For personal use only; all rights reserved.
control and greater inspection efforts to remain functional. Nondestructive tests (NDTs), ultrasonic pulse velocity (UPV) in particular, were
used to increase the effectiveness of the quality assurance of pervious pavements. Predictive models for hydraulic conductivity (main feature)
and compressive strength (main concern) of pervious specimens were derived using only UPV combined with total void ratio. The results
confirm that UPV offers a technique for rapid assessment of in-place properties of pervious concrete, either to supplement or replace core
testing in some cases. Combined use of UPV and void ratio was found to outperform the models based on a single test result. Based on the
obtained results, statistical models were put forward for predictions of pervious concrete’s hydraulic conductivity and compressive strength.
DOI: 10.1061/(ASCE)MT.1943-5533.0002268. © 2018 American Society of Civil Engineers.
Author keywords: Pervious concrete; Hydraulic conductivity; Compressive strength; Nondestructive tests; Predictive modeling.
2014). They were then tested for ultrasonic pulse velocity, ρw × Vol
hydraulic conductivity (HC), compressive strength (CS), total where V r = total void ratio (%); W 1 and W 2 = weight under water
void ratio, and density. A 10-cm coring drill was used to core and dry weight (kg), respectively; Vol = volume of sample (mm3 );
the pervious specimens with various heights ranging from 15 and ρw = density of water (kg=mm3 ).
to 20 cm. The UPV test was performed in accordance with ASTM C597
The HC, also known as the coefficient of permeability, of the (ASTM 1989). The test is used to estimate the uniformity, quality,
specimens was evaluated through a falling-head method, compat- and strength of the tested specimens through a direct, semidirect, or
ible with ASTM D5084 Method C [ASTM D5084 (ASTM 2016)] indirect method. The direct method is the most accurate method,
for porous materials. The details of the falling head permeability but impractical in the field, and the indirect method is normally
test setup is shown in Fig. 1. All specimens were kept in a water applied instead. The UPV test determines the required time for
tank for 3 days to be fully saturated prior to the HC test. The spec- an ultrasound vibration pulse to travel through a specimen. The
imens were then wrapped in duct tape. Since the diameter of the changes in the wave speed indicate the variability of the material
specimens was smaller than the diameter of the standpipe, speci- properties (i.e., density and dynamic modulus of elasticity).
mens were wrapped using a flexible foam material to adjust their All the specimens were kept at 105°C for 3 days prior to the
size to the standpipe. To ensure accuracy, the HC test was repeated UPV test. A direct UPV test was carried out three times on each
three times for each specimen, and the average was then calculated. core, and the average values were reported. The rough surfaces of
The HC was determined using Eq. (1) (Das and Sobhan 2013) the specimens were trimmed using sandpaper to ease the coupling
of UPV transducers.
Eventually, after all the NDTs were performed, the CS of the
cores was evaluated based on ASTM C39 (ASTM 2015).
Data Modeling
into five equal-sized samples (with five data points in each sample).
One of these samples was retained as the test set (or validation set),
and the remaining samples (four samples) were used as training sets
for fitting a regression model. This cross-validation process was
repeated four times, with each sample used exactly once as the test
set. This part of the study was also used to evaluate the predictive
contribution of predictors, their powers, and their interactions using
stepwise regression analysis (Krzanowski 2000). This means that
the statistical significance of the following exogenous variables was Fig. 2. Correlation among different responses
tested: UPV, VR, and density.
Further, to measure the quality of the models and to confirm the
results obtained by cross-validation methodology, the Akaike infor-
mation criterion (AIC) and Bayesian information criterion (BIC) elastic modulus (E) falls as density decreases and void ratio in-
methodologies were used. The model fit was optimized by mini- creases and also because of the longer path around the voids that
mizing AIC and BIC. These criteria were selected because they the acoustic wave needs to take. On the other hand, the effect of the
provide a trade-off between the model complexity and the quality aforementioned factors on the strength of the concrete is observ-
of the fit of the data, and they also highly penalize for overfitting able. The lower the void ratio is, the higher the density, and the
(Dziak et al. 2012). faster the ultrasonic pulse velocity is, the higher the value of CS
obtained. However, the relationship between these factors with
the HC and/or CS of previous concrete is not well understood,
Experimental Results and Discussion while it can highly improve the engineers’ judgments for quality
control.
Univariate Models Using Individual NDT Results Fig. 3 shows two regions of quality as a function of void ratio.
Fig. 2 demonstrates the correlations between the variables and the The figure shows that all the pervious specimens exhibited a void
responses; the diagonal squares show the names of the investigated ratio higher than 18%. The regions correspond to Region I, which
properties, squares above the diagonal present the correlation co- has a HC between 1,000 and 4,000 cm=h and CS between 10 and
efficients between the properties, and squares below the diagonal 22 MPa, and Region 2, which has a HC between 4,000 and
show the relationship between the variables and responses. From 8,000 cm=h and CS between 4 and 10 MPa. As expected, regres-
the shaded squares, it can be observed that UPV is highly correlated sion analysis of the results shows that there is a close relationship
with other investigated properties in this study. Given basic knowl- between UPV and porosity of the mixtures. UPV is normally used
edge of science, HC is known to increase with increasing void ratio to predict the total porosity of the cementitious materials. However,
[ACI Committee 522R-10 (ACI 2010)]. Previous research con- the relationship between the porosity and permeability of pervious
ducted by Delatte et. al (2009) and Amini et. al (2016) demon-
strates an exponential relationship between the HC and void
ratio. According to their results, Eqs. (8) and (9) can be derived:
HC ¼ 1E þ 08e−0.0012 Void Ratio ð8Þ
The reported R2 values using the above equations are 0.62 and
0.6 for Eqs. (8) and (9), respectively. It is clear that higher statistical
adequacy and model validation should be provided to generalize a
model. This paper attempts to increase the accuracy of the previous
models and predictability of the pervious concrete properties
through univariate and multivariate regression analysis.
For this research, it was hypothesized that the UPV would be
Fig. 3. Concrete quality classification results and relationship between
related to in-place density and void ratio, with velocity decreasing
UPV and void ratio
as density decreases and void ratio increases. This was because the
concrete has been discussed in several studies (Kevern 2006; Low of higher than 3,000 cm=s correspond to mixtures in Region I with
et al. 2008; Mehta and Monteiro 2005; Schaefer et al. 2006). In a void ratio between about 18 and 28% and comparatively adequate
other words, it has been shown that higher interconnectivity would CS. This classification is expected to be useful for researchers and
generally exist in the porous media of the mixtures, with a higher field engineers to estimate the strength and HC of pervious concrete
Downloaded from ascelibrary.org by Kamran Amini on 03/16/18. Copyright ASCE. For personal use only; all rights reserved.
total porosity resulting in higher HC (Ong et al. 2016). That being based on UPV and/or void ratio values.
said, according to the results obtained in this study, application of The statistical parameters found through the simple regression
UPV can provide an adequate estimation of the porous media, analysis of all four univariate models [Eqs. (4)–(7)] to predict the
which can directly point to the hydraulic properties (permeability) HC and CS are given in Table 1, and the respective regression lines
of the pervious concrete. The results show that when the mixtures are plotted in Figs. 4(a and b). The statistical values in Table 1 show
have a void ratio higher than 30%, the UPV is relatively constant that all four models are approximately similar. Before making in-
and less than 3,000 cm=s, and the concrete suffers from very low ferences using regression models, however, the validity of the mod-
CS (Region II, 4 < CS < 10 MPa). On the other hand, UPV values els should be verified. Figs. 5 and 6 illustrate the residuals of these
Fig. 4. Accuracy assessment of the developed different univariate models to express the relationship between UPV results and (a) HC;
(b) CS
models versus the predicted HC and CS, respectively. According to The MSE of Models 1 (linear), 3 (power), and 4 (exp.) are 170,
the basic regression analysis assumption (mean = 0 and sd = 1), 179, and 169, respectively. Although Models 1 and 4 are approx-
Fig. 5 shows that the residuals plot of the exponential model seems imately the same, the lower MSE value confirms that the equation
to meet the assumptions the most. derived based on exponential regression is the most adequate model
In the case of CS (Fig. 6), except for the polynomial model, all for predicting the CS of pervious concrete based on UPV. Also, the
the residual plots are almost similar and possibly adequate for stat- small coefficient found for the second-order-polynomial equation
istical prediction of CS. However, mean-square-error (MSE) analy- suggests that the underlying physics of the problem can be ex-
sis may confirm which of the models can be the most adequate. plained by a first-order model.
Fig. 7. Cross-validation prediction results for multivariate model using combined variables to predict HC
Fig. 8. Cross-validation prediction results for multivariate model using combined variables to predict CS
Multiple Regression Model The results of multiple regression, AIC, and BIC are tabulated in
Tables 2 and 3 for HC and CS, respectively. When fitting a model,
Four multiple predictive regression models (Models 1–4) were in-
adding new variables can possibly increase the likelihood of the
vestigated through different statistical methodology to find the
model; however, this may result in overfitting. For this discussion,
most adequate model. Equations for the case of CS are similar.
BIC and AIC are used to resolve this problem by introducing a
• Model 1 (univariate model): Hydraulic conductivity ∼αUPV
penalty term for the number of parameters in the model. This pen-
• Model 2: Hydraulic conductivity ∼αUPV þ β Void Ratio
alty is larger in the BIC compared to the AIC. Results from Tables 2
• Model 3: Hydraulic conductivity ∼αUPV þ β Density
and 3 confirm the previous findings by cross-validation analysis.
• Model 4: Hydraulic conductivity ∼αUPV þ β Voids Ratio þγ
As expected, Models 2 and 4 show the highest and similar R2 val-
Density
ues. However, because of the lower number of the variables in
The cross-validation analysis was first carried out on all differ-
Model 2, as highlighted in Tables 2 and 3, Model 2 exhibited
ent models, and then AIC and BIC criteria were used to measure
the lowest BIC values for both HC and CS. Therefore, the results
the quality of the models to confirm the results obtained by cross-
indicate that the optimal model order is Model 2 for both HC and
validation methodology. The results show that both Models 2 and 4
CS, as the average test error was minimized at this model order.
produce a significantly better prediction of HC than the univariate
Finally, each model (developed by fitting to 4=5 of the data) was
model (1). Fig. 7 shows that Models 2 and 4 have lower dispersion
used to predict its corresponding test set of the data. The predictions
compared to Models 1 and 3.
are presented in Fig. 9. The best model is chosen as the average of
Therefore, based on the results obtained in this study, a combi-
the five and is presented as Eqs. (10) and (11):
nation of UPV and void ratio can improve predictions of the HC of
pervious concrete. In addition, backward elimination on this full
model showed that all variables should be retained. Similar infer- E½HCjUPV; VR ¼ −4.16 × UPV þ 239.909 × VR þ 10299.36
ences were obtained by fivefold cross-validation analysis. The in- ð10Þ
teraction variables (UPV × VR and UPV × Density) were not
significant in any of the partitions. Similar results were obtained
for CS prediction (Fig. 8). E½fc0 jUPV; VR ¼ 0.0082 × UPV − 0.5 × VR − 0.39 ð11Þ
Table 2. Statistical Results to Measure the Relative Quality of Different Table 3. Statistical Results to Measure the Relative Quality of Different
Models to Predict HC Models to Predict CS
Measures Measures
Model R2 AIC BIC MSE.CV Model R2 AIC BIC MSE.CV
1 0.73 425 429 1,305,996 1 0.71 368 371 135,806
2 0.81 418 423 1,052,755 2 0.79 360 366 105,959
3 0.72 427 432 1,427,247 3 0.69 363 368 119,929
4 0.81 418 424 1,135,642 4 0.79 361 367 103,348
Fig. 10. Residuals plot of the model proposed in (a) Eq. (10); (b) Eq. (11)
The average of the ratio between measured and predicted values core testing in some cases. One difficulty is likely to be coupling
for HC and CS are 0.97 and 1.00, respectively, which confirm the the UPV transducers to the rough surface of a pervious concrete
predictability of the models. There are, however, a few outliers. pavement. This can be eased by trimming the surface using
Overall, the proposed models appear to predict HC and CS data sandpaper.
with suitable accuracy, and may be useful for the prediction of
new data within the range studied in this paper. Fig. 10 shows
the plot of the residuals corresponding to the developed multivari- Acknowledgments
ate models using the UPV and void ratio. It is evident that the re-
siduals are randomly scattered around zero, indicating a good fit. This study is based on work supported in part by the National
Therefore, the proposed models seem adequate for predictions and Science Foundation through Grant No. IIP-0917994. The conclu-
inferences. sions expressed herein are the conclusions of the authors and not
necessarily those of the National Science Foundation. The authors
gratefully acknowledge this support.
Conclusions
This paper used the results of UPVand total void ratio to investigate References
their suitability for the prediction of the HC and CS of pervious
concrete, with no need for information regarding the concrete’s ACI (American Concrete Institute). (2010). “Report on pervious concrete
properties and mixture proportions. It was desired to provide mod- (reapproved 2011).” ACI Committee 522R-10, Farmington Hills, MI.
Amini, K. (2015). “Laboratory testing of high performance repair materials
els for in situ predictions of existing pervious pavements. Based on
for pavements and bridge decks.” Master thesis, Cleveland State Univ.,
the results reported herein, the following main conclusions can
Cleveland.
be drawn: Amini, K., Jalalpour, M., and Delatte, N. (2016). “Advancing concrete
• Fivefold cross-validation analysis indicated that combined use strength prediction using non-destructive testing: Development and
of UPV and void ratio improves the predictions compared to verification of a generalizable model.” Constr. Build. Mater., 102,
those based on only a single test result. This was also confirmed 762–768.
through regression analysis, AIC, and BIC measures. Based on Amini, K., Lesak, A., Sohal, A., Adato, J., and Delatte, N. (2014). “Using
this result, a multivariate regression model is proposed for the ultrasonic pulse velocity to predict properties and performance of per-
predictions of HC and CS of pervious concrete. vious concrete.”Transportation Research Board 93rd Annual Meeting,
• The results confirm that UPVoffers a technique for rapid assess- Transportation Research Board, Washington, DC, 41–49.
ment of in-place properties of pervious concrete, especially ASTM. (1989). “Pulse velocity through concrete.” ASTM C597, West Con-
when combined with void ratio, either to supplement or replace shohocken, PA.
niques.” Corros. Sci., 98, 573–584. Low, K., Harz, D., and Neithalath, N. (2008). “Statistical characterization
Chandrappa, A. K., and Biligiri, K. P. (2016). “Influence of mix parameters of the pore structure of enhanced porosity concretes.” Concrete Tech-
on pore properties and modulus of pervious concrete: An application of nology Forum. Focus on Sustainable Development, National Ready
ultrasonic pulse velocity.” Mater. Struct., 49(12), 5255–5271. Mix Association, Denver.
Chandrappa, A. K., and Biligiri, K. P. (2017). “Investigations on pervious Luck, J., Workman, S., and Higgins, S. (2006). “Hydrologic properties
concrete properties using ultrasonic wave applications.” J. Test. Eval., of pervious concrete.” ASAE Annual Meeting, American Society of
45(5), 20160117. Agricultural and Biological Engineers, Portland, OR, 1.
Chindaprasirt, P., Hatanaka, S., Chareerat, T., Mishima, N., and Yuasa, Y. Malhotra, V., and Carino, N. (2004). Handbook on nondestructive testing of
(2008). “Cement paste characteristics and porous concrete properties.” concrete, CRC Press, New York.
Constr. Build. Mater., 22(5), 894–901. Martin, W., Kaye, N., and Putman, B. (2014). “Impact of vertical porosity
distribution on the permeability of pervious concrete.” Constr. Build.
Ćosić, K., Korat, L., Ducman, V., and Netinger, I. (2015). “Influence of
Mater., 59, 78–84.
aggregate type and size on properties of pervious concrete.” Constr.
Mehta, P. K., and Monteiro, P. J. M. (2005). Concrete: Microstructure,
Build. Mater., 78, 69–76.
properties, and materials, McGraw-Hill, New York.
Das, B., and Sobhan, K. (2013). Principles of geotechnical engineering,
Neithalath, N., Sumanasooriya, M. S., and Deo, O. (2010). “Characterizing
Cengage Learning, Stamford, CT.
pore volume, sizes, and connectivity in pervious concretes for per-
Delatte, N., Mrkajic, A., and Miller, D. (2009). “Field and laboratory evalu-
meability prediction.” Mater. Charact., 61(8), 802–813.
ation of pervious concrete pavements.” Transp. Res. Rec., 2113(1),
Neithalath, N., Weiss, J., and Olek, J. (2006). “Characterizing enhanced
132–139.
porosity concrete using electrical impedance to predict acoustic and hy-
Deo, O., and Neithalath, N. (2010). “Compressive behavior of pervious
draulic performance.” Cem. Concr. Res., 36(11), 2074–2085.
concretes and a quantification of the influence of random pore structure
Ng, K., Sun, Y., Dai, Q., and Yu, X. (2014). “Investigation of internal frost
features.” Mater. Sci. Eng. A, 528(1), 402–412.
damage in cementitious materials with micromechanics analysis, SEM
Deo, O., and Neithalath, N. (2011). “Compressive response of pervious imaging and ultrasonic wave scattering techniques.” Constr. Build.
concretes proportioned for desired porosities.” Constr. Build. Mater., Mater., 50, 478–485.
25(11), 4181–4189. Ong, S., Wang, K., Ling, Y., and Shi, G. (2016). “Pervious concrete physi-
Draper, N. R., and Smith, H. (1981). Applied regression analysis, Wiley, cal characteristics and effectiveness in stormwater pollution reduction.”
New York. InTrans Project. Rep. 197, Iowa State Univ., Ames, IA.
Dziak, J. J., Coffman, D. L., Lanza, S. T., and Li, R. (2012). “Sensitivity Park, S., and Tia, M. (2004). “An experimental study on the water-
and specificity of information criteria.” Technical Rep. No. 12-119, purification properties of porous concrete.” Cem. Concr. Res., 34(2),
Pennsylvania State Univ., State College, PA. 177–184.
El Mir, A., and Nehme, S. (2015). “Application of non-destructive test on Popovics, J. (2005). “Pervious concrete pavements.” Mater. Eval., 63(1),
self-compacting concrete.” IABSE Symp. Rep., Vol. 105, International 50–55.
Association for Bridge and Structural Engineering, Geneva, 1–8. Schaefer, V. R., Wang, K., Suleiman, M. T., and Kevern, J. T. (2006). “Mix
Güneyisi, E., Gesoğlu, M., Kareem, Q., and İpek, S. (2016). “Effect of design development for pervious concrete in cold weather climates.”
different substitution of natural aggregate by recycled aggregate on per- Technical Rep., Iowa State Univ., Ames, IA.
formance characteristics of pervious concrete.” Mater. Struct., 49(1–2), Shu, X., Huang, B., Wu, H., Dong, Q., and Burdette, E. (2011). “Mix de-
521–536. sign development for pervious concrete in cold weather climates.”
Gupta, R. (2014). “Monitoring in situ performance of pervious concrete in Constr. Build. Mater., 25(8), 3187–3192.
British Columbia: A pilot study.” Case Stud. Constr. Mater., 1, 1–9. Tennis, P. D., Leming, M. L., and Akers, D. J. (2004). Pervious concrete
Hjorth, J. (1993). Computer intensive statistical methods: Validation, pavements, Portland Cement Association, Skokie, IL.
model selection, and bootstrap, Chapman & Hall, London. Wang, X., Taylor, P., Wang, K., and Lim, M. (2015). “Monitoring of setting
Huang, B., Wu, H., Shu, X., and Burdette, E. (2010). “Laboratory evalu- time of self-consolidating concrete using ultrasonic wave propagation
ation of permeability and strength of polymer-modified pervious con- method and other tools.” Mag. Concr. Res., 63(3), 151–162.
crete.” Constr. Build. Mater., 24(5), 818–823. Washer, G., and Fuchs, P. (2004). “Ultrasonic testing of reactive powder
Jin, X., and Li, Z. (2001). “Dynamic property determination for early-age concrete.” IEEE Trans. Ultrason. Ferroelectr. Freq. Control, 51(2),
concrete.” ACI Mater. J., 98(5), 365–370. 193–201.