Professional Documents
Culture Documents
Civil Engineering
a r t i c l e i n f o a b s t r a c t
Article history: Assessment of existing structures especially compressive strength evaluation of concrete structures is an
Received 20 October 2019 important topic for engineers working in construction in most industrial countries. Non-destructive tests
Revised 8 January 2020 (NDT), especially ultrasonic pulse velocity and rebound number tests are mostly wide spread techniques
Accepted 17 February 2020
to predict the compressive strength of exiting concrete structures. This study conducted an experimental
Available online 12 March 2020
program on concrete specimens by non-destructive tests including ultrasonic pulse velocity and rebound
number then an efficient approach namely; Response Surface Methodology (RSM) is applied to estimate
Keywords:
the compressive strength of concrete with more accuracy rather other available models in the literature.
Ultrasonic pulse velocity
Rebound hammer
The sole NDT and the combination of them are also investigated. The results showed that ultrasonic pulse
Compressive strength velocity is the best NDT tests at the beginning ages when it is used in RSM process.
Concrete structures Ó 2020 The Authors. Published by Elsevier B.V. on behalf of Faculty of Engineering, Ain Shams University.
Response surface methodology This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-
nd/4.0/).
https://doi.org/10.1016/j.asej.2020.02.009
2090-4479/Ó 2020 The Authors. Published by Elsevier B.V. on behalf of Faculty of Engineering, Ain Shams University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
940 A. Poorarbabi et al. / Ain Shams Engineering Journal 11 (2020) 939–949
on-site concrete strength prediction [8]. By reviewing the litera- There is no a well-known relation between hardness and
ture, it is concluded that an efficient approach to predict the com- strength of concrete but it can be found using an experimental pro-
pressive strength should be taken into account, while there are cess. However, it should be taken into account that this relation-
some models. Hence, the main purpose of this study is to propose ship is under influence of concrete surface situation such as
an efficient approach to predict the compressive strength of con- saturation degree, carbonation, temperature, surface preparation
crete using non-destructive tests (UPV and RN). and location. Other parameters such as the type of aggregate,
mix proportion, hammer type, and hammer inclination can also
lead to the incorrect results. Concrete specimens must be approx-
2. Ultrasonic pulse velocity (UPV)
imately in the same situation (regarding age, moisture conditions
and carbonation degree). Akashi and Amasaki [22] performed an
UPV method is a well-known non-destructive testing method
assessment the effect of carbonation on the rebound number. Grieb
that conducted by the velocity of compression stress waves
[24] estimated strength using the effect of type of aggregates on
(P-waves). The velocity of P-waves are transforming into a solid
the rebound number. Willetts [25] showed that moisture content
material which it is depended on its material specifications [18].
of concrete has influence on the results of the rebound number
The fundamental concept of UPV test method is based on the
test.
travel velocity of ultrasonic pulses through a material. The pulse
It is obvious that the rebound number obtain only the surface of
velocity equipment includes an emitter, a receiver, and a device
concrete specifications. BS 1881: Part 202 [20] suggests that the
to show the transformation time where the pulse velocity is
measured number is an indication for just the first 30-mm depth
defined as follow:
of concrete. According to Ref. [26], the results of rebound hammer
L are only given the outer concrete layer properties with a thickness
V¼ ð1Þ
t of 30–50 mm.
Due to the difficulty of obtaining a proper correlation between
where V = pulse velocity (km=s); L = path length (mm); and
RN and concrete strength, the rebound number is just appropriate
t = transit time (ls).
to conduct on the large areas of similar types of concrete. Neville
The pulse velocity technique is an efficient method to assess the
[27] presented the reasonable advantages of the rebound hammer
concrete quality because it just depends on the elastic properties of
test in concrete and showed that the test cannot lead to strength
the material and no geometry. An overall view of pulse velocity
test. Hence, it is necessary to combine this test with other well-
testing is given in Fig. 1.
known non-destructive tests.
The use of PRSM, firstly proposed by Box and Wilson [29]. This
approach is applied based on mathematical and statistical concepts
that predict the output parameters (responses) by using a function
relating the input parameters. The response is defined as follow:
y ¼ Xb þ ey ð2Þ
which X is input data vector, b demonstrate unknown coefficient
vector and ey is error vector. In addition, Eq. (2) can be restated
for a typical quadratic polynomial as following:
X
k X
k X
k
Fig. 1. Schematic diagram of pulse velocity testing circuit (adapted from ASTM y ¼ b0 þ bi xi þ bij xi xj ð3Þ
C586). i¼1 i¼1 j¼1
A. Poorarbabi et al. / Ain Shams Engineering Journal 11 (2020) 939–949 941
6.1. Materials
Table 4
Mix proportions of concrete.
In this study, concrete specimens are constructed using Port-
land cement type II which its chemical and physical specifications No. W/C Cement Coarse gravel (SSD) Fine gravel (SSD) Sand (SSD)
are given in Table 1. Coarse aggregates are used in the crushed kg/m3
form with two sizes of 4.75–12.5 mm (fine gravel) and 9.5– 1 0.79 273 – 936 863
25 mm (coarse gravel) and the used sand is applied in the natural 2 0.79 244 497 478 936
form that their sieve analysis are given in Table 2. Table 3 shows 3 0.74 291 – 927 980
the physical characteristics of the used aggregates. In terms of 4 0.74 261 493 473 928
5 0.69 313 – 915 968
aggregation, the used aggregates are in good agreement with the 6 0.69 280 487 469 919
standard curves of ASTM C 33. 7 0.65 332 – 920 957
8 0.65 297 529 509 925
6.2. The mix design 9 0.61 354 – 909 949
10 0.61 316 524 504 921
11 0.58 372 – 905 945
In the present study, the number of mix designs are 20. These 12 0.58 332 530 490 908
designs are fitted with ACI 318 which their details are given in 13 0.54 400 – 890 888
Table 4. 14 0.54 357 528 488 860
15 0.5 432 – 935 872
16 0.5 386 550 469 855
6.3. Specimens and experimental process 17 0.45 480 – 908 825
18 0.45 429 533 412 800
Totally, 9 cubic specimen with dimensions of 19 0.42 514 – 884 753
150 150 150 mm were tested at the age of 7, 28 and 90 days, 20 0.42 460 558 419 768
for each mix design. After constructing the mixtures, they were
immediately placed in mold and compressed by a rod in accor-
dance with EN12390-2: 2000 Standard then the specimens were According to the purposes of this research, three different tests
got out from the molds after 24 h and placed in water reservoirs conducted on specimens. These experiments were velocity of
that saturated by lime for the needed time. After the curing, the ultrasound transmissions measurement, return number measure-
specimens were got out from the water and prepared for testing. ment and measuring the compressive strength.
To do this aim, each specimen, after leaving the water, has A measurement of the velocity of ultrasound waves was per-
remained in the laboratory for some time to reach a dry surface formed according to B.S 1881-PART 203. The surface of the gener-
with saturated state. The needed time was varied due to humidity ators was impregnated with a special gel or refractory grease and 5
and environment temperature. readings were made for each lateral side. The generator’s position
Table 1
Chemical compositions and physical specifications of cement.
was in such a way that the entire sample surface and, therefore, the 7. Results and discussion
sample volume were tested. Time and then velocity of ultrasound
transmission were measured in each record. The average results 180 specimens made in different ages (7, 28 and 90 days) in
were recorded as the transfer time and speed of the waves. experimental program. These specimens designed in twenty dif-
In continue, the specimens are tested with a Schmidt hammer ferent mix designs which three cubic specimens for different
based on B.S 1889-PART 202. The specimens were located on the ages were experimented in each mix design. The used non-
machine’s jaws, and then five records were saved for each lateral destructive tests were Ultrasonic Pulse Velocity (UPV), Rebound
side. The average of 20 records saved as the hardness number of Number (RN). In addition, the compressive strength of concrete
the cube specimens. specimens measured and are given in Table 5. By using the
Finally, the specimens were tested based on EN12390-3 (2001) experimental data that are given in Table 5, different models
for measuring compressive strength and their value were recorded. including bilinear, power-power, double exponential, logarithmic
Table 5
Experimental results from non-destructive tests.
Mix No. Age (days) UPV (km/s) RN fc (kg/cm2) Mix No. Age (days) UPV (km/s) RN fc (kg/cm2)
1 7 3.55 14.00 113.33 11 7 3.90 19.00 184.00
3.39 20.00 124.44 3.96 19.00 187.11
3.49 17.00 126.22 4.00 21.00 188.00
28 4.05 18.00 166.70 28 4.13 20.00 211.55
4.03 19.00 165.44 4.10 20.00 209.33
4.02 18.00 163.55 4.43 24.00 280.89
90 4.23 21.00 198.67 90 4.43 23.00 248.00
4.23 21.00 191.11 4.47 24.00 233.44
4.29 20.00 195.55 4.39 24.00 252.00
2 7 3.55 15.00 140 12 7 4.17 23.00 217.33
3.58 14.00 143.55 4.23 25.00 225.30
3.37 14.00 135.11 4.29 25.00 225.33
28 4.10 18.00 172.00 28 4.39 26.00 270.67
4.11 19.00 169.78 4.41 27.00 252.00
4.16 20.00 170.22 4.42 27.00 260.44
90 4.41 21.00 196.89 90 4.62 25.00 292.44
4.39 21.00 196.44 4.62 25.00 294.67
4.39 20.00 200.44 4.62 25.00 288.00
3 7 3.70 21.00 166.27 13 7 4.09 22.00 234.00
3.87 21.00 166.70 4.13 22.00 233.78
3.71 19.00 170.22 4.12 22.00 237.00
28 4.17 22.00 204.00 28 4.26 25.00 268.70
4.19 22.00 200.00 4.25 27.00 276.44
4.18 22.00 205.33 4.24 28.00 286.67
90 4.35 23.00 232.44 90 4.54 27.00 288.00
4.43 22.00 213.33 4.48 27.00 279.11
4.37 23.00 231.11 4.51 27.00 281.33
4 7 3.76 18.00 132.44 14 7 4.23 23.00 229.33
3.73 18.00 130.22 4.20 22.00 226.67
3.77 18.00 132.40 4.23 23.00 233.78
28 4.12 18.00 163.55 28 4.39 25.00 292.00
4.12 18.00 168.89 4.40 24.00 285.78
4.13 18.00 158.67 4.43 24.00 280.89
90 4.35 18.00 171.18 90 4.53 28.00 304.44
4.39 18.00 175.11 4.51 28.00 310.67
4.35 18.00 173.78 4.55 26.00 312.00
5 7 3.79 19.00 141.33 15 7 4.21 24.00 286.67
3.75 18.00 144.89 4.18 26.00 279.11
3.68 19.00 140.89 4.23 26.00 273.33
28 3.96 15.00 172.89 28 4.28 24.00 321.78
3.93 15.00 170.22 4.32 27.00 328.44
3.95 14.00 163.55 4.28 26.00 320.89
90 4.23 19.00 198.22 90 4.65 32.00 358.22
4.23 21.00 209.33 4.62 32.00 354.67
4.22 20.00 190.67 4.61 32.00 347.55
6 7 3.88 19.00 143.55 16 7 4.17 23.00 216.89
3.86 18.00 140.89 4.18 23.00 220.89
3.87 18.00 148.00 4.12 23.00 217.33
28 4.08 15.00 184.89 28 4.31 24.00 267.11
4.09 16.00 183.55 4.32 23.00 262.67
4.02 16.00 185.33 4.32 24.00 263.11
90 4.45 18.00 188.89 90 4.66 28.00 329.78
4.34 18.00 197.33 4.66 27.00 328.44
4.39 18.00 204.89 4.70 29.00 323.55
A. Poorarbabi et al. / Ain Shams Engineering Journal 11 (2020) 939–949 943
Table 5 (continued)
Mix No. Age (days) UPV (km/s) RN fc (kg/cm2) Mix No. Age (days) UPV (km/s) RN fc (kg/cm2)
7 7 3.93 21.00 188.44 17 7 4.02 22.00 221.78
3.95 21.00 188.44 4.08 25.00 217.33
3.94 22.00 193.78 4.07 24.00 224.44
28 4.09 20.00 225.78 28 4.24 25.00 288.89
4.09 21.00 233.78 4.23 25.00 281.78
4.11 21.00 228.89 4.17 24.00 270.22
90 4.45 24.00 249.33 90 4.65 25.00 289.78
4.48 25.00 252.89 4.59 28.00 295.11
4.50 24.00 250.22 4.67 27.00 294.22
8 7 3.84 19.00 163.11 18 7 4.25 25.00 278.22
3.92 19.00 153.30 4.19 26.00 279.55
3.91 19.00 151.55 4.08 25.00 283.11
28 4.16 20.00 210.67 28 4.45 28.00 342.22
4.14 20.00 205.33 4.37 30.00 327.11
4.14 21.00 202.22 4.45 30.00 350.67
90 4.42 23.00 229.78 90 4.78 34.00 348.00
4.34 24.00 231.11 4.73 32.00 333.78
4.41 23.00 235.55 4.66 35.00 350.22
9 7 3.93 18.00 179.60 19 7 4.18 23.00 296.44
3.93 18.00 185.78 4.17 24.00 307.44
3.93 19.00 171.11 4.17 25.00 294.22
28 4.09 21.00 205.78 28 4.38 29.00 368.00
4.18 20.00 214.22 4.31 28.00 374.22
4.09 21.00 202.22 4.31 29.00 369.33
90 4.35 24.00 236.89 90 4.68 34.00 385.33
4.51 23.00 239.55 4.69 35.00 379.11
4.37 23.00 241.33 4.85 34.00 387.11
10 7 4.02 20.00 177.33 20 7 4.36 25.00 307.55
4.01 21.00 173.33 4.32 25.00 312.00
4.03 20.00 175.11 4.32 25.00 306.67
28 4.12 22.00 217.33 28 4.42 32.00 387.55
4.25 22.00 215.55 4.39 32.00 352.89
4.10 21.00 218.67 4.52 31.00 369.78
90 4.44 24.00 217.78 90 4.82 35.00 372.89
4.62 25.00 231.11 4.90 36.00 375.11
4.52 27.00 227.11 4.72 36.00 385.78
and RSM are fitted for ages of 7, 28 and 90 days. Moreover, calculated to have the best accuracy), besides, RSM model is also
these models are also fitted for all data (ages of 7 days, 28 days fitted using UPV and RN tests. The accuracy of each model is shown
and 90 days). in Figs. 2–5 for different ages.
Available models to estimate the compressive strength of con- As can be seen RSM model predicted the compressive strength
crete using UPV and RN tests are presented in Table 6. The relative with more accuracy rather other models with R2 = 0.863, 0.871,
coefficients have calculated for common models in the literature 0.937 and 0.873 for the age of 7, 28, 90 days and all specimens
review including power-power model, bilinear model, double in different ages (7, 28 and 90 days). For instance, in Fig. 2 for
exponential model and logarithmic model based on the experi- the 7 days age, averagely next accuracy is assigned to the double
mental data which presented in this study (these models were fit- exponential model that is in good agreement with the experimen-
ted on the experimental data of this study then their coefficients tal data but its accuracy is less than RSM model. Next one is power-
Table 6
The fitted models based on the experimental data in different ages for estimating the compressive strength using UPV and RN tests.
350 350
300 300
fc (Experimental) 250 250
fc (Experimental)
R2 =0.837 R2 =0.804
200 200
150 150
100 100
50 50
0 0
0 100 200 300 0 100 200 300
f c (Power-Power Model) f c (Bilinear Model)
(a) (b)
350 350
300 300
250 250
fc (Experimental)
fc (Experimental)
R2 =0.848 R2 =0.773
200 200
150 150
100 100
50 50
0 0
0 100 200 300 0 100 200 300
f c (Double Exponential Model) f c (Logarithmic Model)
(c) (d)
350
300
250
fc (Experimental)
R2 =0.863
200
150
100
50
0
0 100 200 300
f c (RSM Model)
(e)
Fig. 2. Performance of models at the age of 7 days: a) Power-Power, b) Bilinear, c) Double exponential, d) Logarithmic, e) RSM.
power model and then bilinear model. The accuracy of logarithmic the accuracy of the fitted models. For this aim, the following crite-
model is at the end place. Fig. 3 shows the accuracy of the fitted ria are investigated [35]:
models for the age of 28 days. Based on this figure, the order of The total error (etot ):
accuracy is such the age of 7 days. Fig. 4 presents the accuracy of PN
1 jExpei Theoi j
models for the age of 90 days. Based on this figure, the best accu- etot ¼ 100 PN ð6Þ
racy is for RSM model, then the order of accuracy is related to: 1 jExpei j
bilinear, logarithmic, power-power and finally double exponential.
in which Expei and Theoi are the experimental compressive strength
Fig. 5 indicates the accuracy of models for total data including the
and the compressive strength results from theoretical models,
ages of 7, 28 and 90 days. The order of accuracy is related to: 1.
respectively. N is also the total samples.
RSM, 2. bilinear, 3. power-power, 4. logarithmic and 5. double
Other parameters are mean square error (MSE), average abso-
exponential. Hence, it can be concluded that RSM model has the
lute magnitude error (AAE), standard deviation (SD) that are pre-
best accuracy among available models and its accuracy is not
sented as follow, respectively:
under influence of the age of specimens. PN Theoi Expei 2
In addition, for better comprehension and comparison, other 1ð Expei
Þ
well-known statistical parameters are also performed to check MSE ¼ ð7Þ
N
A. Poorarbabi et al. / Ain Shams Engineering Journal 11 (2020) 939–949 945
400 400
300 300
fc (Experimental)
fc (Experimental)
R2 =0.862 R2 =0.855
200 200
100 100
0 0
0 100 200 300 400 0 100 200 300 400
f c (Power-Power Model) f c (Bilinear Model)
(a) (b)
400 400
300 300
fc (Experimental)
fc (Experimental)
R2 =0.867 R2 =0.819
200 200
100 100
0 0
0 100 200 300 400 0 100 200 300 400
f c (Double Exponential Model) f c (Logarithmic Model)
(c) (d)
400
300
fc (Experimental)
R2 =0.871
200
100
0
0 100 200 300 400
f c (RSM Model)
(e)
Fig. 3. Performance of models at the age of 28 days: a) Power-Power, b) Bilinear, c) Double exponential, d) Logarithmic, e) RSM.
PN Theoi Expei methods which are well-known in the literature (such as Double
1 Expei
AAE ¼ ð8Þ Exponential).
N Table 7 shows that the total error of RSM model is less rather
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi other models (8% better rather double exponential, 13%
uP better rather power-power, 23% better rather bilinear and 29%
u N Theoi Theoav g 2
t 1 ðExpei Expeav g Þ better rather than logarithmic models). This table is also shown
SD ¼ ð9Þ
N1 RSM has the best performance in other statistical parameters.
Table 8 indicates the mentioned statistical parameters for the age
The statistical parameters to evaluate the accuracy of the avail-
of 28 days. This table also proves that RSM is the best fitted model.
able models and RSM are given in Tables 7–10. Then, the value of
For example, in MSE parameter, RSM is better about 1%, 15%, 24%
these parameters for each model is compared to RSM as follows:
and 42% rather double exponential, power-power, bilinear and
logarithmic, respectively.
XOthermethods XRSM
100 ð10Þ Table 9 is also given the statistical parameters for the age of
XOthermethods
90 days of specimens. This table is also shows the efficiency of
In which, XRSM is the value of the RSM for the specified statisti- RSM model. For instance, in AAE parameter, RSM is better
cal parameter and Xother methods is the value of the other mentioned approximately 32%, 63%, 8% and 10% rather double exponential,
946 A. Poorarbabi et al. / Ain Shams Engineering Journal 11 (2020) 939–949
400 400
fc (Experimental)
R2 =0.926 R2 =0.930
200 200
100 100
0 0
0 100 200 300 400 500 0 100 200 300 400 500
fc (Power-Power Model) fc (Bilinear Model)
(a) ( b)
400 400
300 300
fc (Experimental)
fc (Experimental)
R2 =0.911 R2 =0.928
200 200
100 100
0 0
0 100 200 300 400 0 100 200 300 400
fc (Double Exponential Model) fc (Logarithmic Model)
(c) 400 (d)
300
fc (Experimental)
R2 =0.937
200
100
0
0 100 200 300 400
fc (RSM Model)
(e)
Fig. 4. Performance of models at the age of 90 days: a) Power-Power, b) Bilinear, c) Double exponential, d) Logarithmic, e) RSM.
power-power, bilinear and logarithmic, respectively. In addition, tests, function including RN has the highest accuracy especially in
Table 10 reports the accuracy of RSM model among the available the age of 90 days. In addition, the age of 90 days leads to higher
models for all specimens. Of course, standard deviation is roughly accuracy, among two-parameter functions. It is obvious that by
constant for all models (even in this parameter, RSM has the lowest aging, the accuracy is also raised.
value that shows its efficiency) but other statistical parameters
proves the highly accuracy of RSM rather other models. It is 8. Conclusion
noticed that RSM is not sensitive to the age of concrete specimens
while the accuracy of other models are under influence of age. In this paper, a model proposed to estimate the compressive
As mentioned RSM has the best performance among models, strength of concrete using non-destructive tests. To this aim,
hence this model is selected to investigate more details about RSM used as an accurate model to estimate the compressive
the influence of the measured non-destructive tests (sole test or strength and it compared with other available models in the liter-
combined tests) on the accuracy of prediction. ature. The results showed the accuracy of RSM is more and can pre-
Table 11 shows the accuracy of prediction using different non- dict the compressive strength with reasonable accuracy in
destructive tests in different ages. Based on this table, among sole different ages of concrete in comparison with other models. The
A. Poorarbabi et al. / Ain Shams Engineering Journal 11 (2020) 939–949 947
400 400
300 300
fc (Experimental) R2 =0.855 R2 =0.861
fc (Experimental)
200 200
100 100
0 0
0 100 200 300 400 500 0 100 200 300 400 500
fc (Power-Power Model) fc (Bilinear Model)
(a) (b)
400 400
300 300
R2 =0.831 R2 =0.844
fc (Experimental)
fc (Experimental)
200 200
100 100
0 0
0 100 200 300 400 500 0 100 200 300 400
fc (Double Exponential Model) fc (Logarithmic Model)
(c) (d)
400
300
R2 =0.873
fc (Experimental)
200
100
0
0 100 200 300 400 500
fc (RSM Model)
(e)
Fig. 5. Performance of models for all of experimental data (7 days + 28 days + 90 days): a) Power-Power, b) Bilinear, c) Double exponential, d) Logarithmic, e) RSM.
Table 7 Table 8
Statistical parameters for the fitted models using UPV and RN tests for the age of Statistical parameters for the fitted models using UPV and RN tests for the age of
7 days. 28 days.
Theoretical models etot MSE AAE SD Theoretical models etot MSE AAE SD
Power-Power 9.136 0.791 5.458 1.023 Power-Power 8.218 0.665 5.204 1.022
Bilinear 10.405 1.044 6.558 1.028 Bilinear 8.408 0.738 5.398 1.023
Double exponential 8.688 0.681 5.152 1.023 Double exponential 8.293 0.569 5.066 1.023
Linear Logarithmic 11.203 1.263 7.099 1.032 Linear Logarithmic 9.231 0.978 6.027 1.027
RSM 7.989 0.547 4.747 1.022 RSM 7.911 0.563 4.904 1.022
influence of concrete age is not highly effect on RSM while concrete compressive strength of concretes using the sole RN is more effi-
aging effects on other available models in the literature. In addi- cient, while the combination of RN and UPV could not improve
tion, the effect of sole and combined non-destructive tests were the accuracy in comparison with sole tests. Hereby, the results
examined. Results showed that the improvement to estimate the are summarized as follow:
948 A. Poorarbabi et al. / Ain Shams Engineering Journal 11 (2020) 939–949
Ali Poorarbabi is a Ph.D. candidate in Civil Engineering Mehdi Azhdary Moghaddam is an associate Professor in
Department, University of Sistan and Baluchestan. He Civil Engineering Department, University of Sistan and
received a MS.c degree from Islamic Azad University Baluchestan. He received a Ph.D. degree from University
(Kerman branch), Iran in the field of Civil Engineering – of Ottawa, Canada at 1997 in the field of Civil
Hydraulic Structures. His main research interest Engineering- Hydraulic Structures. His main research
includes: non-destructive tests and concrete structures. interest includes: dams and concrete structures.