You are on page 1of 10

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/316475774

Study of the influence of admixtures on creep of concrete in the Eurocode 2:


Using the Approximate Bayesian Computation method

Conference Paper · August 2017

CITATIONS READS

0 253

5 authors, including:

Elise Zgheib Wassim Raphael


Saint Joseph University, Lebanon Saint Joseph University, Lebanon
12 PUBLICATIONS   8 CITATIONS    74 PUBLICATIONS   200 CITATIONS   

SEE PROFILE SEE PROFILE

Nathalie Salameh Pierre Matar


Saint Joseph University, Lebanon Lebanese University
1 PUBLICATION   0 CITATIONS    25 PUBLICATIONS   88 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

CREEP OF CONCRETE View project

Climate change effects and adaptation of civil infrastructure and buildings View project

All content following this page was uploaded by Pierre Matar on 26 April 2017.

The user has requested enhancement of the downloaded file.


Study of the Influence of Admixtures on Creep of Concrete
in the Eurocode 2 using the Approximate Bayesian Com-
putation Method
Elise Zgheiba,b, Wassim Raphaela, Nathalie Salameha, Pierre Matarb and Alaa Chateauneufc
a
Saint-Joseph University, Science and Technology Campus, Lebanon
b
Doctoral School for Science and Technology (EDST), Lebanese University, Lebanon
c
Institut Pascal, Blaise Pascal University, Clermont University, France

Abstract: Since the predicted creep compliance according to the Eurocode 2 does not
take the effect of admixtures into consideration, a study of the influence of admix-
tures on creep is performed. Based on a large database established from international
laboratories and research centers, a comparison between the experimental creep and
the creep predicted by the Eurocode 2 is performed using statistical methods. An in-
accurate estimation is detected based on the type of the admixture used. The use of a
combination of water reducer and silica fume as admixtures leads to an underestima-
tion of the creep compliance.
In order to overcome this difference, a calibration is required by adding corrective
coefficients to the Eurocode 2 equations, taking into consideration the type and per-
centage of admixtures. The Approximate Bayesian Computation method based on the
rejection algorithm is applied in order to calculate the corrective coefficients.
After implementing the corrective coefficients in the Eurocode 2 compliance formula,
and in order to evaluate our updated Eurocode 2 creep model, statistical methods are
used. An improvement in the results is clearly shown. Using the large experimental
database, the present study demonstrates the importance of the Bayesian model as-
sessment for the updating of the Eurocode 2 creep model, taking into account the ef-
fect of admixtures. The adoption of such a design approach would improve long-term
serviceability of structures subject to creep.

1 Introduction
Concrete is subjected to long-term deformations classified into two main categories based on
the presence or the absence of sustained applied load. The first category includes defor-
mations manifested without application of external load known as the shrinkage deformations.
The creep deformations define the second category; they appear during the lifetime of the
structure and they are caused by the sustained applied load. In addition to these two catego-
ries, instantaneous deformations reveal also when the external load is applied. The defor-
mations are affected by any type of admixtures added to the concrete and have an important
impact on the behavior, stability and durability of the structures.
Excessive deflections, difficulties with closure or un-esthetic permanent deflections may re-
sult from an inaccurate creep analysis. In 1978, the first large worldwide creep and shrinkage
database was collected by Bažant and Panula at the Northwestern University (NU). This data-
base coming from American and European institutions consists of results of approximately
400 creep tests and 300 shrinkage tests. The collaboration established between the ACI and
the CEB leads to an extension of this database performed by Müller and Panula. Further ex-
pansion has been undertaken by the RILEM TC 107-CSP Subcommittee 5 [21], leading to
what is known as the RILEM Database. Raphael et al. have used this database in order to pre-
dict the creep compliance for concrete that does not include admixtures in their composition
[22]. A major expansion leads to the current database completely restructured and verified,
and named the NU Database. This database was lately used in order to propose a multi-
objective genetic programming technique for the modeling of complex engineering systems;
this technique was applied to develop a prediction model for the time dependent total creep
compliance of concrete [12]. Using this database, this paper presents a study showing the im-
pact of admixtures on the concrete creep strains.
Admixtures are additions to a concrete mixture that can help in controlling the setting time
and other properties of fresh concrete [20]. In addition, it can improve its performance by
modifying its characteristics and enhancing workability. Concrete admixtures are of two main
types: chemical and mineral.
Chemical admixtures modify properties of concrete; improve its quality during mixing, trans-
porting, placing, and curing. They fall into the following categories: air entrainers, water re-
ducers, retarders, accelerators, corrosion inhibitors and shrinkage reducers.
Mineral admixtures make mixtures more economical, reduce permeability, increase strength,
and influence other concrete properties. They affect the nature of hardened concrete through
hydraulic or pozzolanic activity. Pozzolans are cementitious materials and include natural
pozzolans, fly ash and silica fume.
Gong et al. [14] have proposed a creep model for the axial compression concrete member
suffered from sulfate attack by combining effective stress in concrete damage mechanics the-
ory with the B3 Model for concrete creep [4]. In Model B4 [2], the effect of admixture type
and percentage is taken into consideration by adding scaling factors to p2, p3, p4 and p5 pa-
rameters [17].
The Eurocode 2 does not take the admixtures into consideration in the evaluation of the creep.
Therefore, the present paper treats the impact of the water reducers and the silica fumes ad-
mixtures on the creep strains according to the Eurocode 2 [11] by using the Bayesian method,
which is one of the multiple methods that can be applied in order to predict the creep strains
from measurements. Ghasemzadeh et al. [13] have used the Inverse Analysis method to pre-
dict long-term creep from short-term measurements.
The Bayesian updating [6] is an advantageous method of analyzing experimental data, which
is the case of the present study. Bayesian model assessment is essential for a good evaluation
of structural safety, development of a reliability method that accounts for imperfect states of
knowledge and recognizes all sources of uncertainty arising in structural problems. By adding
correction coefficients, the Approximation Bayesian Computation (ABC) method [26] is ap-
plied in this study to take into account the effect of admixtures in the calculation of creep
strains.
2 Method of Analysis
2.1 Database Parameters and Admixtures
H. Rüsch and O. Wagner performed the first data collection at the Technical University of
Munich. This limited database became the basis for the first CEB creep and shrinkage model.
D.E. Branson and M.L. Christiason’s paper [7] underlying the 1971 ACI-209 Model included
another data. The first large database assembled by Z.P. Bažant and L. Panula at the North-
western University was included in the papers presenting the BP Model [5]. A joint ACI-
RILEM committee was organized, during the international ACI Fall 1979 Convention, to ex-
tend the Northwestern University database. This work was pursued in a Subcommittee
chaired by H. Müller of the RILEM Committee TC107-CSP that led to the RILEM-ACI 209
database in 1992. Some further additions to the database were done, at NU, by G. H. Li, in
2008 [3] and by K.T. Kim, in 2010 [19]. The latest database was assembled at NU during
2010-2013 [18] mainly under the support of the U.S. Department of Transportation. The in-
formation was extracted from many reports, conference proceedings and journal articles. Us-
ing this large database [10], a comparison is performed between the results obtained by
laboratory testing and those given by the theoretical model as indicated in the Annex B of the
Eurocode 2. The creep tests presented in this database have different parameters and condi-
tions including but not limited to water to cement ratio, aggregate to cement ratio, cement
type, concrete strength, effective thickness, age at loading, temperature, relative humidity,
sustained stress, admixtures, etc. Since the admixtures have an influence on the behavior of
concrete, the tests with admixtures are separated from those without admixtures. This large
database contains 117 creep tests containing water reducers and silica fume admixtures simul-
taneously. The percentage of water reducers varies between 0.6% and 4.8 %, whereas the sili-
ca fume percentage varies between 4.17% and 17.6%.
The silica fume (SiO2) is a mineral admixture added to concrete to improve its properties,
particularly, its compressive strength and abrasion resistance. These improvements results
from both the addition of a very fine powder to the cement paste mixture as well as from the
pozzolanic reactions between the silica fume and the free calcium hydroxide in the paste. The
addition of silica fume to the concrete mix increases its compressive strength during the first
days of hydration but barely influences its compressive strength after 28 days [27].
The water reducer (WR) is a chemical admixture that usually reduces the required water con-
tent for a concrete mixture by about 5 to 10 percent. Consequently, concrete containing a wa-
ter-reducing admixture needs less water to reach a required slump than untreated concrete.
The treated concrete can have a lower water to cement ratio. This usually indicates that a
higher strength concrete can be produced without increasing the amount of cement.

2.2 Approximate Bayesian Computation Method


The Bayesian calibration [6] provides an automated process for calibrating models [25] by
multiplying the expert knowledge known as a priori distribution [15] by the likelihood com-
ing from the database [16]. The a priori distribution incorporates the uncertainty of the pa-
rameters that shall be calibrated. The objective is to refine these probability distributions by
comparing the results coming from the Bayesian update to the experimental measurements. In
the calibration processes, it is important to distinguish variables from parameters [26]. The
variables are input data [8] that are known but variable in times, while parameters are un-
known constant data modeled by a priori distribution [8]. By multiplying the likelihood func-
tion with the a priori distribution, an a posteriori distribution will be obtained which will be
an update of the knowledge already known by using the latest database provided.
Different methods of Bayesian calibration may be used according to the available information
on the likelihood function [26]. If the likelihood function is analytically known, then Markov
Chain Monte Carlo (MCMC) method can be used. If the likelihood function is unknown ana-
lytically but an analytical approximation can be defined, then the MCMC method based on
metamodeling may be applied. If the likelihood function is unknown analytically, then the
Approximate Bayesian Computation method is used. In this study, the Approximate Bayesian
Computation method based on the rejection algorithm is applied [26].
The Approximate Bayesian Computation rejection algorithm method consists in generating a
random vector for each correction coefficient following an a priori distribution [26]. For each
random variable, the updated compliance is calculated and compared to the experimental
compliance. If the difference between the updated and the experimental compliances is less
than the threshold, then the random variable is accepted; if not, it is rejected. Once this proce-
dure is applied, we obtain at the end a set of correction coefficient values with a known a pos-
teriori distribution or an empirical a posteriori distribution.

2.3 Evaluation Methods


The creep compliance J (t, t0) is the time-dependent strain per unit stress. In order to evaluate
the accuracy of the Eurocode 2 creep compliance prediction based on the experimental tests,
five methods are applied.

2.3.1 The residual method


The residuals are calculated by the difference between the experimental compliance and the
theoretical one, as given in equation (1):
𝑅𝑅𝑖𝑖𝑖𝑖 = (𝑂𝑂𝑂𝑂𝑂𝑂 𝑋𝑋𝑖𝑖𝑖𝑖 – 𝐶𝐶𝐶𝐶𝐶𝐶 𝑋𝑋𝑖𝑖𝑖𝑖 )#(1)
Figure 1 shows the Eurocode 2 residuals versus experimental creep compliance scatter of
concretes containing water reducer and silica fume simultaneously.

Figure 1: Eurocode 2 residuals versus experimental creep compliance of concretes containing water reducer
and silica fume (J (t, t0) in MPa-1)
We notice that Eurocode 2 residuals increase linearly with the experimental creep compliance.
Moreover, the majority of the scatters are located above the X-axis. Therefore, the residuals
values are positive and the experimental creep compliance exceeds the predicted creep com-
pliance. Consequently, the Eurocode 2 underestimates the creep compliance of concretes con-
taining water reducers and silica fume.

2.3.2 The CEB methods


The CEB methods are statistical methods used to evaluate a model relative to the experi-
mental database. The CEB mean deviation MCEB, the CEB coefficient of variation VCEB, and
the CEB mean square error FCEB formulas are presented in [1].

2.3.3 The RMSD method


The Root Mean Square Deviation (RMSD) represents the sample standard deviation of the
differences between predicted values and observed values. It is calculated using equation (2):
1
𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 = � �(𝐶𝐶𝐶𝐶𝐶𝐶 𝑋𝑋𝑖𝑖𝑖𝑖 − 𝑂𝑂𝑂𝑂𝑂𝑂 𝑋𝑋𝑖𝑖𝑖𝑖 )2 #(2)
𝑁𝑁
When RMSD is near zero or have small values, the results of the predicted compliance are
close to the experimental results.

2.4 Division of the Creep Compliance into Categories


The division of the creep compliance into categories has been used effectively to study the
impact of the range of creep compliance. Although the residual versus experimental creep
scatter plot shows a linear behavior, Figure 1 shows that for small creep compliances, the
scatters are around the X-axis, which indicates that the Eurocode 2 estimates accurately the
creep compliances, while, for larger values of creep compliances, the residual values increase
and the Eurocode 2 underestimates the creep compliances. In their study [24], Raphael et al.
have divided the creep compliance into three categories as follows:
- Small creep compliance (SC) for 0 < J (t, t0) < 60x10-6 MPa-1.
- Medium creep compliance (MC) for 60x10-6 MPa-1 < J (t, t0) < 120x10-6 MPa-1.
- Important creep compliance (IC) for J (t, t0) > 120x10-6 MPa-1.
In the present paper, the same division into categories was applied for the creep compliance of
concretes containing water reducer and silica fume simultaneously.

3 Results
A prediction of the creep compliance according to the Annex B of Eurocode 2 is performed.
Then a comparison between the obtained results and the experimental measurements is ap-
plied using the CEB and the RMSD methods. Table 1 shows the evaluation method results of
the creep compliance for each category.
Table 1: Evaluation method values for creep compliance before correction

Category of creep Compliance MCEB VCEB FCEB RMSD


Small creep compliance 0.98 23.5 23 7
Medium creep compliance 0.53 58.1 58 44
Important creep compliance 0.44 59.5 59.6 105
Table 1 shows that the Eurocode 2 estimates accurately the small creep compliance and un-
derestimates the medium and important creep compliance. To overcome this underestimation,
correction coefficients are introduced to the Eurocode 2 equation as follows:
𝑊𝑊𝑊𝑊 𝑆𝑆𝑆𝑆𝑆𝑆2
𝐽𝐽(𝑡𝑡, 𝑡𝑡0 ) = �𝐴𝐴 𝑥𝑥 + 𝐵𝐵 𝑥𝑥 � 𝑥𝑥 𝐽𝐽𝐸𝐸𝐸𝐸2 (𝑡𝑡, 𝑡𝑡0 )#(3)
100 100
With:
- A: a correction coefficient relative to the water-reducing admixture,
- B: a correction coefficient relative to the silica fume admixture,
- WR (in %): the percentage of water reducer added to the concrete mixture,
- SiO2 (in %): the percentage of silica fume added to the concrete mixture,
- JEC2 (t, t0): the theoretical creep compliance predicted by the Eurocode 2.
The correction coefficients A and B are calculated using the Approximate Bayesian Computa-
tion rejection algorithm. The results of applying this method on the creep compliance for the
concrete mixture with silica fume and water reducer are shown in Table 2.
Table 2: Correction coefficient values for creep compliance

Category of creep compliance Correction coefficients


Small creep compliance No need for correction
Medium creep compliance A = 14.8, B = 17
Important creep compliance A = 15, B = 24

Table 2 summarizes the results of the correction coefficients. We can notice that since the
Eurocode 2 estimates accurately the creep compliance for the small creep category, no correc-
tion coefficients are available. As for the medium and important creep categories, correction
coefficients are required. Once the values of A and B are determined, equation 3 is applied for
each test in the database and the results are compared to the experimental measurements.
Table 3: Evaluation method values for creep compliance before and after correction

Category of creep compliance MCEB VCEB FCEB RMSD


Small creep Before correction 0.98 23.5 23 7
compliance After correction N/A N/A N/A N/A
Medium creep Before correction 0.53 58.1 58 44
compliance After correction 1 47.2 47.2 27
Important creep Before correction 0.44 59.5 59.6 105
compliance After correction 1 37.9 37.8 61.5

Table 3 summarizes the results of MCEB, VCEB, FCEB and RMSD before and after the Bayesian
correction. By comparing the values obtained before and after correction, we can notice that
for both medium and important creep categories, the mean deviation value became equal to 1,
and the VCEB, FCEB and RMSD values have decreased after correction. Therefore, by applying
the correction coefficients A = 14.8 and B = 17 for 60x10-6 MPa-1 < JEC2 (t, t0) < 120x10-6
MPa-1 and A = 15 and B = 24 JEC2 (t, t0) > 120x10-6 MPa-1, the corrected creep compliances of
concrete mixtures incorporating water reducer and silica fume meet the experimental values.
Figure 2 shows a diagram plot for the MCEB values before and after correction for each creep
category.
Figure 2: The MCEB values of the Eurocode 2 evaluation for each category before and after correction

From Figure 2, we can notice that, after applying the correction coefficients, the mean devia-
tion MCEB has increased from 0.53 to 1 for the medium creep category and from 0.44 to 1 for
the important creep category. Therefore, the corrected creep compliance is nearer to the ex-
perimental values after correction since the MCEB is equal to the expected value 1.

4 Conclusion
This paper treats the effect of water reducer and silica fume added together to concrete mix-
tures, on the creep compliance. The first step consists on evaluating the applicability of the
Eurocode 2 creep compliance formulas to investigated concrete mixtures. We have found that
the Eurocode 2 underestimates the creep compliance for concrete mixture with water reducer
and silica fume. The approach used in the design codes is clearly insufficient for practical
engineering structures when admixtures are added to concrete mixes. In order to predict the
creep compliance for concrete containing water reducer and silica fume, the correction ap-
proach based on the Bayesian updating is applied to the creep compliance equation with addi-
tion of two correction coefficients that take the percentage of admixtures into consideration.
The correction coefficients were calculated according to the Approximate Bayesian Computa-
tion rejection algorithm using Matlab and the experimental database. The Approximate
Bayesian Computation rejection algorithm has proven to be an effective solution for the im-
provement of the creep prediction according to the Eurocode 2 and very interesting results
have been obtained.

Acknowledgement
The authors gratefully acknowledge Professors Zdeněk Pavel Bažant, Fabrice Guérin, David
Bigaud, Mr. Hassen Riahi and Mr. Rafic Faddoul for their help. Many thanks go also to all
persons from RILEM, LCPC, CEBTP, and Northwestern University for their kind support for
providing experimental data. The National Council for Scientific Research, Lebanon (CNRS-
L) and the Research Council of Saint-Joseph University of Beirut have supported this work.
References
[1] ACI Committee 209. ACI 209.2R-08 Guide for Modeling and Calculating Shrinakge
and Creep in Hardened Concrete. U.S.A. 2008.
[2] Z.P. Bažant. RILEM draft recommendation: TC-242-MDC multi-decade creep and
shrinkage of concrete: material model and structural analysis. Model B4 for creep, dry-
ing shrinkage and autogenous shrinkage of normal and high-strength concretes with
multi-decade applicability. Materials and Structures, Vol. 48. 2015.
[3] Z.P. Bažant, and G.H. Li. Comprehensive database on concrete creep and shrinkage.
ACI Materials Journal, Vol.105. 2008.
[4] Z.P. Bažant, and S. Baweja. Creep and shrinkage prediction model for analysis and
design of concrete structures: Model B3. ACI Special Publications, Vol. 194. 2000.
[5] Z.P. Bažant, and L. Panula. Practical prediction of time dependent deformations of con-
crete. Materiaux et Construction, Vol.11. 1978.
[6] G.E.P. Box and G.C. Tiao. Bayesian inference in statistical analysis. Wiley-Interscience
Publication. John Wiley and Sons, INC. 1992.
[7] D.E. Branson, and M.L. Christiason. Time-dependent concrete properties related to
design – Strength and elastic properties, creep and shrinkage, ACI Special Publication
SP-27, Creep, shrinkage and temperature effects in concrete structures, American Con-
crete Institute, Detroit, Michigan, 1971.
[8] K. Campbell. Statistical Calibration of computer simulations. Reliability Engineering
and System Safety, Vol. 91. 2006.
[9] A. Chateauneuf, W. Raphael and R.M. Pitti, Reliability of prestressed concrete struc-
tures considering creep models. Journal of Structure and Infrastructure – Taylor &
Francis Edition, vol. 10, issue 12 , 2014.
[10] NU database of laboratory creep and shrinkage data.
www.civil.northwestern.edu/people/bazant.
[11] CEN. EN 1992-1-1, Eurocode 2: Design of concrete structures – Part 1-1: General
rules and rules for buildings. Brussels. 2004.
[12] A.H. Gandomi, S. Sajedi, B. Kiani, and Q. Huang. Genetic programming for experi-
mental big data mining: A case study on concrete creep formulation. Automation in
construction, Vol. 70. 2016.
[13] F. Ghasemzadeh, A. Manafpour, S. Sajedi, M. Shekarchi, and Z. Hatami. Predicting
long-term compressive creep of concrete using inverse analysis method. Construction
and Building Materials, Vol. 124, 2016, 496-507.
[14] J. Gong, J. Cao., and Y.F. Wang. Effects of sulfate attack and dry-wet circulation on
creep of fly-ash slag concrete. Construction and Building Materials, Vol. 125, 2016, 12-
20.
[15] F. Guérin, B. Dumon, and E. Usureau. Reliability estimation by Bayesian method: Defi-
nition of prior distribution using dependability study. Reliability Engineering and Sys-
tem Safety, Vol. 82. 2003.
[16] Y. Heo, R. Choudhary, and G.A. Augenbroe. Calibration of building energy models for
retrofit analysis under uncertainty. Energy and Buildings, Vol. 47. 2012.
[17] M.H. Hubler, R. Wendner, and Z.P. Bažant. Statistical justification of Model B4 for
drying and autogenous shrinkage of concrete and comparisons to other models. Materi-
als and structures, 2015.
[18] M.H. Hubler, R. Wendner, and Z.P. Bažant. Comprehensive database for concrete
creep and shrinkage: Analysis and recommendations for testing and recording. ACI
Materials Journal, Vol.112. 2015.
[19] K.T. Kim. Creep database V1.1. www.iti.northwestern.edu/publications/bazant. 2010.
[20] CIM Béton, and Ecole Française du Béton. Les constituants des betons et des mortiers.
Paris, France. 2005
[21] H.S. Müller, Z.P. Bažant, and C.H. Kuttner. Database on creep and shrinkage tests.
RILEM Subcommittee 5 Report RILEM TC107-CSP, RILEM, 1999.
[22] W. Raphael, F. Kaddah, F. Geara, and A. Chateauneuf. Information – based modeling of
creep in concrete structures. ICOSSAR, 11th International Conference on Structural
Safety and Reliability, New York, USA, 2013.
[23] W. Raphael, B. Seif-El-Dine and D. El Asmar Selouan. Information based formulation
for Bayesian concrete creep models updating. Proceeding of the ICOSSAR 9th Interna-
tional Conference on Structural Safety and Reliability, 2005.
[24] W. Raphael, A. Chateauneuf, M. Lemaire, J.L. Favre and J.A. Calgoro. Reliability
based assessement of prestressed concrete structures subject to creep. application to a
bridge. Proceedings of the 9th International Conference on Applications of Statistics
and Probability to Civil Engineering (ICASP9), San Fransisco. 2003.
[25] M. Riddle and R.T. Muehleisen. A guide to bayesian calibration of building energy
models. In ASHRAE/IBPSA-USA Building Simulation Conference. Atlanta, GA. 2014.
[26] M. Robillard. Etude des stratégies de gestion en temps réel pour des bâtiments énergé-
tiquement performants. Thèse de doctorat de l’École Nationale des mines de Paris,
France. 2015.
[27] N. Tebbal, Z. Rahmouni, and M. Belouadah. Influence d’un ajout sur le comportement
mécanique des bétons à haute performance soumis à des températures élevées. XXXe
Rencontres de l’AUGC-IBPSA. Chambéry. 2012.

View publication stats

You might also like