Engineering Structures: L. Orta, F.M. Bartlett T

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Engineering Structures 210 (2020) 110396

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Sensitivity analysis of restrained shrinkage stresses of concrete deck overlays T


a,b,⁎ a
L. Orta , F.M. Bartlett
a
The University of Western Ontario, Department of Civil and Environmental Engineering 1151 Richmond Street, London, Ontario N6A 5B9, Canada
b
Tecnologico de Monterrey, Department of Civil Engineering, Ramon Corona #2514 Zapopan, Jalisco 45201, Mexico1

A R T I C LE I N FO A B S T R A C T

Keywords: Concrete overlays are widely used to rehabilitate bridge decks. This investigation presents parametric and
Shrinkage sensitivity analyses of the mechanical strains due to restraint of shrinkage of the overlay that may cause it to
Parametric analysis crack. The parametric analysis identifies the significant variables and the sensitivity analysis quantifies the
Mechanical strains correlations between these variables and the magnitude of mechanical strains. The strain analysis accounts for
Sensitivity analysis
drying shrinkage using Fick’s law, and for tensile creep strains. Total strains are computed by superposition,
Bridge deck overlays
assuming a linear-elastic concrete behavior, using a time-history approach developed in a previous investigation.
Concrete overlays on concrete slabs and composite systems are considered. Compressive strength of the overlay
concrete, curing period, overlay thickness and environmental humidity contribute the most to the variability of
the analytically predicted mechanical strains.

1. Introduction will be used in the present investigation. Other researchers [8–10] have
proposed simplified methods with shortcomings that are overcome with
Rehabilitation of concrete bridge decks by replacing unsound con- the THAM analysis [4]. Other studies in restrained shrinkage consider
crete with a new concrete overlay is common in Canada [1,2]. When changing few parameters like the degree of restraint in ring experi-
concrete decks deteriorate, the wearing surface becomes inadequate for ments when other parameters are kept the same [11].
traffic and, less frequently, the safety of the structure may be reduced to The overall research objective is to identify factors that contribute
unacceptable levels. Jack–hammer or hydro demolition is typically to the magnitude and variability of the mechanical tensile strains and
used to remove damaged concrete [3] that is then replaced with new stresses due to restrained shrinkage in concrete overlays. To attain this
concrete. Shrinkage of the new concrete overlay is restrained at the objective, a parametric study is first conducted to identify the sig-
interface with the substrate, however, producing tensile stresses in the nificant variables, and then their impact is quantified using a prob-
overlay that may cause cracking. A cracked overlay provides corrosion abilistic-based sensitivity analysis.
protection of the steel reinforcement and is a poor riding surface, par-
ticularly if the cracks are large and widely spaced, so cracking of the 2. Research significance
overlay can be considered a limit state.
The likelihood of overlay cracking is difficult to predict because the Durable and economical partial depth concrete replacement re-
tensile stresses and tensile strength of concrete are uncertain during the quires knowledge and control of the critical factors that cause re-
life span of the overlay. Both quantities depend on other factors that are strained shrinkage cracking so that long service lives can be achieved in
also uncertain including free shrinkage strain, creep strains, and en- practice. To identify potential economical savings, it is essential to
vironmental humidity. identify those factors that are critical to overlay durability, assess cur-
This paper concludes a series by the authors [4–7] that focused on rent practices and propose means to extend the expected un-cracked life
the quantification of stresses due to restrained shrinkage in concrete of the overlay.
overlays: the reader is referred to these references for additional lit-
erature review. The computation of the deterministic stress distribution 3. Analytical model
(in time and space) due to restrained shrinkage in concrete overlays has
been developed using a Time History Analysis Method (THAM) [4] that A parametric study will explore the variability of the mechanical


Corresponding author at: Tecnologico de Monterrey, Department of Civil Engineering, Ramon Corona #2514 Zapopan, Jalisco 45201, Mexico.
E-mail address: luis.orta@itesm.mx (L. Orta).
1
Permanent address.

https://doi.org/10.1016/j.engstruct.2020.110396
Received 27 May 2019; Received in revised form 5 February 2020; Accepted 17 February 2020
Available online 21 February 2020
0141-0296/ © 2020 Elsevier Ltd. All rights reserved.
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

strains assuming practical maximum and minimum limits of the input becomes significant when the thickness of the overlay exceeds that of
parameters to filter out those variables that are not influential. A sub- the substrate [4], so the effect of different surface preparations is not
sequent sensitivity analysis will quantify the contribution of each explicitly considered. The present study accounts for swelling of the
variable to the mechanical strain and thus will identify important substrate due to moisture that has diffused from the overlay, and as-
parameters that can be controlled. The parametric analysis is not sociated compressive creep in the substrate due to this swelling.
computationally demanding and effectively identifies the variables that The cross section is discretized in constant thickness layers and the
have little impact but considers neither correlation between the input solution is obtained at discrete times [4]. The principle of superposition
variables nor the probability distribution that they are sampled from. is applied to compute the mechanical strains at any time:
The sensitivity analysis is conducted to recognize the interaction among
input variables and their probability distribution. ε (y, t ) = εm (y, t ) + εc (y, t ) + εsh (y, t ) (1)
The investigation is focused on mechanical strains, εm(y,t), rather Free shrinkage strain, εsh(y,t), and creep strain, εc(y,t), are needed to
than stresses because, while the cracking strength of concrete, ft(t), is compute the mechanical strain, εm(y,t). The total strain, ε(y,t), is
proportional to the square root of the compressive strength [11] (in- computed assuming that plane sections remain plane and that the cross
creasing with concrete age), the elastic modulus of concrete, Ec(t), is section is free to elongate:
also proportional to the square root of the compressive strength [12], so
the cracking resistance strain εcr, computed as the cracking stress di- ε (y, t ) = ε0 (t ) + y [κ (t )] (2)
vided by the modulus of elasticity, is independent of the concrete age
and compressive strength [5]. A mean cracking strength strain is about where ε0(t) = ε(0,t) is the strain at the top fibre and κ(t) is the curvature
143 μm/m with a coefficient of variation of 34% [5]. of the cross section.
The parametric analysis and sensitivity analysis consider only the Two conditions of equilibrium are imposed to determine ε(y,t). The
mechanical strain at mid-depth of the overlay, εm2(t) = εm(0.5ℓny,t). stress resultants are equal to the axial force and bending moment pro-
Mechanical strains have large gradients close to the top and bottom duced by the external loads, FL(t) and ML(t), respectively:
faces of the overlay in the initial period of drying, (t − t0) ≈ 10 days,
but stabilize shortly after[4]. At the same time, the mechanical strain at ∫ σz (t ) dA = FL (t ) (3]
mid-depth of the overlay, εm2(t), starts to increase and it defines the
time of cracking after the maximum localized strain has exceeded, at ∫ y σz (t ) dA = ML (t ) (4)
very early ages after drying, the cracking resistance, εcr, of concrete[5].
Fig. 1 shows the two types of cross sections considered in the pre- Free shrinkage strains, εsh(y,t), are computed for a given ultimate
sent study: overlays on solid concrete slabs and overlays on composite shrinkage strain using Fick’s second law of diffusion as
steel members. The structure is considered simply supported with a
total length ℓz and having a total deck depth ℓy, including the overlay. ∂εsh ∂ ∂h ∂h ⎤
= βs0 εs ∞ k sh ⎡ ⎛λ (h) ⎞ U(t − t0) +
⎜ ⎟
Overlays on solid concrete slabs are considered to have a nominal ∂t ⎢ ∂y ∂y ∂t ⎥ (5)
⎣ ⎝ ⎠ ⎦
overlay thickness of 150 mm on top of 550 mm thick substrate, so
ℓy = 0.7 m, and the analysis is conducted for a one-metre wide strip, where h = h(y,t) is the internal humidity as a function of location y and
ℓx = 1 m. The 150 mm overlay thickness was chosen because it satisfies time t, βs0 is a constant such that the average shrinkage strain of drying
relevant Ontario specifications [13]. The overall slab thickness would laboratory specimens is equal to the average shrinkage strain [6] pre-
be typical for a short simple span of approximately 14 m. dicted by the Model B3 [15], εs∞ is the ultimate (long-term) shrinkage
Three different steel shapes were investigated for the composite strain for a notional environmental humidity of zero, ksh is a constant of
girder systems to represent slabs on stringers, floor beams or main proportionality that transforms a change in internal humidity into a
girders. The stringer, floor beam and girder consist of a W460x67, a change in shrinkage strain, λ(h) is the diffusivity of the concrete as a
W760x173 and a WWF1400x405, respectively. The width of the slab is function of the internal relative humidity, U(t − t0) is the Heaviside
assumed to be the centre-to-centre spacing of the steel members: 1.5 m step function and t0 is the curing period in days. The first term within
for the slab on stringers, 2.5 m for the slab on floor beams and 3 m for the brackets represent the change of internal humidity due to drying
the slab on girders. The concrete deck is assumed 250 mm thick, in- and the second term the change due to self-desiccation. The ultimate
cluding the overlay of 120 mm. The remaining 130 mm thickness of the shrinkage strain, εs∞, is computed using Model B3 as
original slab is assumed to be sound, and serves as formwork for the
εs ∞ = α1 α2 [0.019w 2.1 f −c28
0.28
+ 270] (6)
new concrete overlay. The shape designations and cross sectional
properties that were used are consistent with those listed by the
where α1 accounts for the type of curing and is equal to 1.2 for moist
Canadian Institute of Steel Construction [14].
curing, α2 accounts for the type of cement and is equal to 1.0 for normal
Previous research [4] has demonstrated that the moisture condition
Type I cement, w is the water content in kg/m3, fc28 is the compression
of the substrate, and hence the substrate surface preparation method,
strength of the cylinders tested at 28 days in MPa. The diffusivity λ(h) is
can impact the mechanical response of the overlay. However this effect
computed as

Fig. 1. Concrete slab and composite girder cross sections.

2
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

Table 1
10 λ 0 ⎞ ⎡ (1 − a0 ) ⎤
Input variables for parametric study.
λ (h) = ⎛ ⎜ ⎢a0 +

ns

⎝ f c28 ⎠ ⎢

1 −h
1 + 1−h
c ( ) ⎥
⎦ (7) Category No. Variable Description Min Max
(unit)
where λ0 = 86.4 mm /day [16] and a0, hc and ns are parameters cali-
2
Substrate 1 Ecs (GPa) Modulus of elasticity 26 38
brated from experimental data[6]. Parameters derived from Model B3
Overlay 2 fc28 (MPa) Compressive strength 25 50
have been adopted because they are believed generally representative 3 ce (kg/m3) Cement content 280 360
of typical concrete mixes, even though they have been shown to be 4 ℓny (mm) Overlay depth 100 200
inadequate for specific mixes[7]. Construction 5 t0 (days) Curing time 2 7
The internal humidity due to self-desiccation is computed as[6] 6 As (mm2/m) Steel reinforcement 0 540
Shrinkage 7 a0 Diffusivity constant 0.03 0.1
h = ha = 1 − a1 [1 − exp(−a2 t na)] (8) 8 hc Diffusivity constant 0.75 0.85
9 ns Diffusivity constant 6 16
where the constants a1, a2 and na are functions of the water-to-ce- 10 λ1 (mm2/ Concrete diffusivity 76.4 96.4
menting materials ratio, w/ce, and the period of curing, t0 [6]. day)
11 a1 Self–desiccation constant 0.017 0.153
Tensile creep strains, εc(y,t), are computed as [4]
12 a2 Self–desiccation constant 0.0198 0.178
13 na Self–desiccation constant 0.168 1.515
dεc dσ (t ′) ⎞ σ (t ) ∂ ∂h
= J(t , t ′) ⎛ z + rc βs0 εs ∞ k sh ⎛ z ⎞ ⎡ ⎛λ (h) ⎞ ⎤ U(t − t0)
⎜ ⎟ ⎜ ⎟
14 βs0 Ultimate shrinkage 0.65 1.95
⎝ dt ′ ⎠ ⎢ ∂y ⎠ ⎥
dt ⎝ ft28 ⎠ ⎣ ∂y ⎝ ⎦ constant
Creep 15 c0 Modulus of elasticity 0.95 1.67
(9) constant
where J(t,t’) is the compliance creep function at time t, t’ is the age of 16 c1 Creep strain constant 0.3 1.1
17 m Creep strain constant 0.33 0.73
loading, σz(t’) is the axial stress at the age of loading, rc is a coefficient
18 nc Creep strain constant 0.2 0.4
calibrated using experimental data and ft28 is the 28-day cylinder ten- 19 rc Creep strain constant 0.1 0.6
sile strength. The first and second terms of this equation represent the Environment 20 hen Relative humidity 0.4 0.8
basic and drying creep [17], respectively. The creep function, J(t,t’), is Modeling 21 Nf Number of fibres 100 200
computed using Dirichlet series [4] as

c1 i=N τinc overlay thickness, ℓny. The maxima shown for concrete strengths reflect
J(t , t ′) = ∑i =1 γi
(t ′)m
{1 − exp[−(t − t ′)/ τi]}
c0 Ec28 (10) a range of normal to moderately high-strength overlays. The cement
content by volume corresponds to the minimum quantity in kilograms
where c0, c1, m and nc are coefficients calibrated using experimental
of cementitious materials per cubic metre of concrete (kg/m3) as re-
data, Ec28 is the modulus of elasticity of the overlay at 28 days, and γi
commended in the specifications for Design and Control of Concrete
and τi are deterministic modelling parameters.
Mixtures [18]. In this study the water content, w (kg/m3), is computed
The numerical Time History Analysis Model (THAM) was used to
for a given 28-day compressive strength as proposed by Bazant and
predict stresses and deflections of a full-scale composite truss tested at
Baweja [15].
the University of Alberta. The agreement between observed and pre-
dicted stresses and deflections is satisfactory [7] and thus the numerical w = 23ce / (12 + fc28 ) (11)
model is appropriate for use in the current investigation.
The overlay depth depends on a decision by the owner or consulting
4. Parametric analysis engineer, typically based on engineering surveys indicating the depths
where the in-place chloride concentrations exceed threshold limits.
Deterministic analyses of εm2(t) are conducted by varying each The construction variables are the time of curing and the amount of
parameter in turn (between realistic minimum and maximum values) steel reinforcement. The lower limit for the time of curing is two days
while the rest are fixed at their average values. The range of each because a shorter period could lead to poor concretes with insufficient
variable is divided in four intervals of equal size and then five values strength or stiffness to withstand superimposed live loads. Curing for
are sampled, corresponding to the boundaries at each interval. The more than seven days could incur high costs due to project delay.
extrema correspond to values observed in ancillary tests by others or to Typical periods of curing are 4 days for partial-depth overlays [5, per-
values simply deemed realistic. The probability distribution corre- sonal communication Ministry of Transportation of Ontario]. The
sponding to each variable is not accounted for in the parametric study amount of steel ranges from the option of no reinforcement to providing
because the objective is simply to identify influential variables to be twice the minimum for shrinkage and temperature specified in ACI 318
included in the sensitivity analysis, where each probability distribution [19], 0.36%. This percentage is applied to the total area of the overlay
will be accounted for. to compute the reinforcing steel area.
Input variables are grouped in the following seven major categories: The extrema shown in Table 1 for a0, hc and ns were observed by
substrate, overlay, shrinkage, tensile creep, construction, environ- Bazant and Najjar [20] using experimental data. The extrema for a1, a2
mental conditions and number of elements for modeling. Table 1 shows and na correspond to the ± 80% limits of the values predicted using
the seven categories of input variables and their corresponding domains expressions by Orta and Bartlett [6]. Similarly, the minimum and
used to investigate the response of the overlay on the all-concrete maximum values of βs0 are the ± 50% limits of the mean value of 1.3
substrates. [6].
The substrate is characterized by its modulus of elasticity, Ecs. It was The extrema of c1, m and nc were observed by Orta and Bartlett [4],
assumed that at the time of overlay replacing the substrate could be and the extrema for rc were observed by Bazant and Chern [17] using
between a normal-strength concrete with Ecs = 26 GPa (fcs ≈ 30 MPa) experimental data from compression creep tests. The extrema of c0
and a high-strength concrete with Ecs = 38 GPa (fcs ≈ 64 MPa). The correspond approximately to the ± 25% limits of the value of 1.35 [4].
internal relative humidity of the substrate is assumed to be in equili- The limits for the range of environmental humidity, hen, in Canada
brium with the environmental humidity, hen , at the time of the re- were estimated from Fig. A.3.1.3 of the Canadian Highway Bridge
habilitation is conducted. Design Code [21].
The overlay is characterized by its nominal 28-day compressive Finally, the number of elements in a given cross section was varied
strength, fc28, the cementing materials content in the mix, ce, and the from 100 to 200 to investigate the consequences of achieving savings in

3
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

depth tensile strains, particularly for t > 100 days. It is not clear how
to design a concrete mix to achieve certain tension creep character-
istics, however some guidance can be inferred by the results obtained
by Bissonnete et al. [22] that a reduction in the cement paste content
increases the tensile creep of concrete. Specifically, their mix with a
paste content of 22% exhibited 100% more creep strains compared to
their mix with a paste content of 32%. These findings contradict the
compression creep expressions recommended in ACI 209 [12] that
imply that increasing the paste content reduces creep in compression.
Fig. 3c indicates that the ultimate (long-term) shrinkage strain
parameter βs0 has a positive correlation with the mechanical strains, i.e.
increasing the ultimate free shrinkage strain will increase the tensile
mid-height strains and the overlay is more likely to crack, as expected.
Fig. 2. Mechanical strain εm2(t) when input parameters are varied. The figure also suggests that this variable is almost as influential as fc28.
Fig. 3d indicates that the duration of moist curing, t0 is most in-
fluential at short times of drying. The effect of increasing t0 is to delay
computational time by reducing the scale of the analysis. Such savings
the onset of shrinkage by keeping moisture in the overlay while its
are essential to carry out the sensitivity analysis.
modulus of elasticity increases and the overlay gains strength. At the
For the parametric studies of overlays on composite systems, all the
end of the longer period of curing, free shrinkage strains interact with a
input variables are as listed in Table 1 except ℓny and As. The nominal
larger modulus of elasticity of the overlay and therefore larger tensile
overlay thickness, ℓny, is assumed to be 120 mm and ranges from
stresses are generated. The effect is less significant as time increases.
100 mm to 140 mm. The nominal area of steel reinforcement, As, is
Fig. 3e indicates that the area of steel reinforcement, As, has no
assumed to be 250 mm2/m and ranges from 0 to 500 mm2/m.
effect in the overlay before cracking. The numerical method (THAM)
does not consider the behavior after cracking and thus the steel re-
4.1. Results of parametric analysis inforcement is only subjected to mechanical strains in the order of
150 μm/m that correspond to stresses in the order of 30 MPa.
Fig. 2 shows the variation of the mechanical strain εm2(t) for the Finally, the response is not sensitive to the number of elements, Nf
case of concrete substrate. Faint lines correspond to the 85 (i.e. as shown in Fig. 3f. This input variable would have more impact if the
21x4 + 1) simulations when each single input variable in turn is ratio of overlay thickness to total thickness approaches to zero or one,
changed. At t = t0 (typically 4 days) the overlay mid-depth strain varies i.e. 0.1 or 0.9, as finer mesh would be required to accommodate the
from 5 to 45 μm/m and at 104 days (27 years) from 12 to 108 μm/m. large gradient of strains expected.
The maximum value of 140 μm/m occurs at 150 days suggesting that, The metric proposed to rank the input variables is the maximum
for this particular case where fc28 is at its minimum value of 25 MPa, the range index, RX, computed as
overlay is likely to crack. The solid line indicates the result obtained
RX = max[rx (t)] (11)
from a single analysis using the mean value of each input variable.
Open white dots represent the average values of the 85 different si- where rX(t) is the time-dependent range index for a particular input
mulations computed at discrete times. The dots all lie on the solid line variable X, computed as
indicating that the single analysis using mean values of each input
2 2
variable gives an excellent prediction of the mean of the mid-depth ⎡ 1 j= Ns − 1 ⎛ Δεm2 (t ) ⎞ Ω Xj ⎤
mechanical strain averaged over all simulations. Broken lines are
rX (t ) = ⎢
N −1
∑ j= 1 ⎜ ⎟ 2 ⎥
⎣ s ⎝ ΔXj ⎠ Ωm2 (t ) ⎦ (12)
plotted at two standard deviations above and below the mean, where
the standard deviations are computed at discrete times from the data where Δεm2(t) denotes the change in the output variable εm2(t) when
from the 85 different simulations. Assuming εcr(t) normally distributed the input variable is changed ΔX, ΩX is the standard deviation of X,
these boundaries define the range that represents 95% of its possible Ωm2(t) is the standard variation of the output and Ns is the number of
magnitude. simulations per variable, i.e. Ns = 5. The range index normalizes the
Fig. 3 shows the variability of εm2(t) for typical significant and in- average change of output for a given change of input using the variance
significant parameters – although the actual probability distribution of the input and output variables.
associated with each parameter has not been considered, the impact of Table 2 shows the resulting ranking of the input variables based on
each is clear. Similar figures for the other input variables investigated RX. The most influential variables, corresponding to the rows shown
are presented in Orta [5]. The middle line denotes the mechanical shaded in the table, include the compressive strength of the overlay,
strain computed using the mean values of each input variable. The fc28, the ultimate shrinkage strain, βs0, the creep strain, c1, the curing
arrows show the sense of the change of the response as the input time, t0, and the overlay thickness, ℓny. The input variables nc, a1, na, m
variable is increased from its minimum to its maximum value. and a2 are discussed in the section of critical variables for the sensitivity
Fig. 3a indicates that overlay concrete strength, fc28, has a sig- analysis. Some of the least influential variables are the modulus of
nificant impact. The mid-depth mechanical strain is reduced markedly elasticity of the substrate, Ecs, the area of steel reinforcement, As, and
with increasing fc28 between 10 and 104 days. This occurs in part be- the number of elements used to model the cross section, Nf, among
cause the ultimate free shrinkage strain and creep parameters depend others.
on this variable: larger fc28 values correspond to smaller ultimate Fig. 4 shows the variation of the mid-height mechanical strain εm2(t)
shrinkage strains [Eq. (6)], reduced diffusivity [Eq. (7)] and larger Ec28 with time for the three overlays on composite substrates. The various
values that cause smaller creep strains [Eq. (10)]. For fc28 = 25 MPa, a lines types and markers shown are identical to those shown in Fig. 2.
maximum εm2(t) of 140 μm/m occurs at approximately 100 days that is The restraining effect of composite substrates is similar to that observed
likely to cause overlay cracking. It would, therefore, seen prudent to set for the concrete substrates and shows little variation with the size of the
the minimum fc28 to be 30 MPa, to reduce the likelihood of overlay steel shape.
cracking, as is currently specified in OPSS 930 [13]. The corresponding range indices RX are also summarized in Table 2.
Fig. 3b indicates that increasing the tension creep variable c1, that The typical range of variation is small for the different steel shapes, and
changes the magnitude and rate of creep strains, causes reduced mid- the values are similar to the corresponding RX value computed for the

4
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

Fig. 3. Variation of mechanical strain εm2(t) with various parameters: (a) overlay concrete strength; (b) basic creep coefficient; (c) ultimate free shrinkage strain
parameter; (d) curing time; (e) steel reinforcement area; and f) number of fibres.

concrete substrate. category the area of steel reinforcement is fixed with a value corre-
sponding to the minimum shrinkage and temperature steel ratio of
0.18% [19]. For the shrinkage category the autogeneous shrinkage
4.2. Critical variables for sensitivity analysis
variables a1, a2 and na are computed from w/ce and t0 as recommended
by Orta and Bartlett [6]. Parameters a0, λ0, hc and ns are fixed at 0.05,
The parametric study results can be used to identify critical vari-
86.4 mm2/day, 0.8 and 11, respectively. Since the parametric study
ables in each category for the subsequent sensitivity analysis. The ob-
results shown in Table 2 indicate that the mechanical strain demand is
jective of filtering the input variables is to save computational time by
sensitive to a1, future experimental data should be obtained to confirm
neglecting variables that are not significant and so enhance the simu-
that the expressions proposed by Orta and Bartlett [6] accurately pre-
lation by improving the modeling of the distributions of the remaining
dict autogeneous shrinkage. For the creep category variables nc and m
input variables.
are not independent and their possible combinations are not unique [4].
The input variables for the sensitivity analysis are the overlay
The variables m and nc are assumed to be constant with values of 0.53
compressive strength fc28, the overlay thickness ℓny, the curing time t0,
and 0.28, respectively, because they represent the best fit to experi-
the ultimate shrinkage strain parameter βs0, the basic creep strain
mental data [4]. The variable rc for drying creep does not rank highly in
coefficient c1 and the environmental humidity hen.
Table 2 and so it is fixed at its mean value of 0.35. Similarly c0 is as-
The rest of the input variables are considered fixed. The modulus of
sumed constant with a value of 1.35. The number of elements Nf is fixed
elasticity of the substrate, Ecs, is fixed at 29.5 GPa. For the overlay
at 100.
category the cement content ce is fixed at value of 350 kg/m3 corre-
sponding to the minimum cementing materials for concrete used in
flatwork based on Tables 9–7 in CAC [18]. For the construction

5
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

Table 2 Table 3
Parameter ranking from parametric study. Statistical parameters for input variables.
Variable Ranking Parameter RX Variable Type Bias COV Distribution Source

Concrete W460 W760 WWF1400 βs 0


I 1.00 0.20 Normal Bazant and Baweja [14]
fc28 D 1.15 0.27† Lognormal Bartlett [23]
Overlay 1 fc28 8.3 8.1 7.5 6.5 ℓny D 1.06 0.30† Lognormal Kennedy et al. [24]
Creep 2 nc 3.5 2.9 3.5 4.2 c1 I 1.00 0.11 Lognormal Orta and Bartlett [13]
Shrinkage 3 βs0 3.5 3.5 3.7 3.8 t0 D 1.00 0.40 Lognormal Assumed
4 a1 3.5 4.0 3.9 3.7 hen I 1.00 0.58 Uniform Assumed
Creep 5 c1 3.3 2.9 3.3 3.4
Shrinkage 6 na 3.1 3.1 3.1 3.1 †
Assumed double the COV for a Type I random variable.
Construction 7 t0 3.0 3.0 3.0 3.0
Overlay 8 ℓny 3.0 1.7 1.3 1.0
Creep 9 m 2.6 3.0 2.4 2.0
coefficient of variation of 0.27 is approximately twice that re-
Shrinkage 10 a2 2.4 2.5 2.4 2.2 commended by Bartlett [23] and so allows fc28 to be represented as a
Environment 11 hen 2.4 1.8 1.8 2.0 decision random variable, Type D. The distribution was assumed log-
Shrinkage 12 hc 0.7 0.7 0.7 0.6 normal to avoid negative magnitudes and was truncated at 20 MPa as
Creep 13 rc 0.6 0.4 0.4 0.3
convergence becomes problematic for lower concrete strengths.
Overlay 14 ce 0.2 0.2 0.2 0.2
Shrinkage 15 ns 0.1 0.1 0.1 0.1 The total concrete deck thickness is assumed to have a nominal
Creep 16 c0 0.1 0.3 0.2 0.1 value of 700 mm for a concrete substrate or 250 mm for a composite
Substrate 17 Ecs 0.1 0.1 0.1 0.0 steel substrate. The nominal overlay depth in either case is 150 mm
Shrinkage 18 a0 0.1 0.0 0.0 0.0 with a bias coefficient of 1.06 [24] and a coefficient of variation of 0.3.
19 λ0 0.1 0.1 0.1 0.1
Modeling 20 Nf 0.0 0.0 0.0 0.0
This coefficient of variation is also approximately twice the re-
Construction 21 As 0.0 0.0 0.0 0.0 commended value by Kennedy et al. [24] to explore a larger range of
overlay thicknesses.
The mean duration of curing of 4 days corresponds to the minimum
5. Sensitivity analysis curing in specifications for partial-depth and full-depth overlays in
Ontario Canada [5, personal communication MTO] and the coefficient
The results are quantified using correlation coefficients, ρX(t), and of variation was selected to reflect short and long curing periods, ran-
sensitivity indices, SX(t). A positive correlation coefficient means that ging between the maxima of t0 shown in Table 1. The distributions of
increasing the magnitude of an input random variable increases the ℓny and t0 are also assumed lognormal to avoid negative values.
mechanical strain. A large sensitivity index indicates an input random The factor for the creep compliance function c1 reflects test-to-
variable that contributes significantly to the variance of the mechanical predicted values obtained in a previous investigation and its nominal
strain. The correlation coefficient can be used to rank the impact of value of 0.71 is also reported [4]. The random environmental humidity,
each input variable on the magnitude of the mechanical strain, whereas hen, is assumed time-independent and takes values from 0.4 to 0.9 to
the sensitivity index indicates the effectiveness of controlling the reflect locations with very low and very high relative humidities.
variability of an input variable to reduce the variability of the me- Each random variable is sampled using the Latin Hypercube
chanical strain. Sampling (LHS) technique [25]. The number of samples per variable is
The accurate calculation of the sensitivity indices requires prob- denoted Ns and the sampling process is repeated for all k input vari-
ability distributions of the variables. Two types of randomness are ables. All the sample points are stored in a matrix M which has Ns rows
considered. Type D denotes a decision random variable that depends on and k columns.
a decision, like time of curing, and so can be controlled by the designer The analysis generates the output vector as
or contractor. Type I denotes an inherently random variable with an
Y (t) = g(M) (13)
established nominal value that cannot be controlled as efficiently, like
the environmental humidity hen. where Y(t) denotes the evaluation of the mechanical strain εm2(t) using
Table 3 summarizes the six variables considered for the sensitivity THAM[4] as the function g(·).
analysis, indicating whether the variable is type D or Type I and pro- The correlation coefficient is computed as
vides the bias coefficient, coefficient of variation (COV), distribution
ρX (t ) = bX (t ) [ΩX /ΩY (t )] (14)
type and typical range for each input variable.
The variable βs0 corresponds to calibration of the ultimate (long- where ΩX denotes the standard deviation of X, ΩY (t) denotes the
term) free shrinkage strain using Model B3. The coefficient of variation standard deviation of Y(t) and bX(t) corresponds to the fitted coefficient
is reported by Bazant and Baweja [15], the mean value is assumed 1.3 of a linear regression analysis.
[6] and the distribution is assumed normal. To compute the sensitivity indices, three sample matrices are
The variable fc28 is based on an assumed nominal compressive needed [26]. The first two matrices, M1 and M2, are sampled in-
strength of 30 MPa [5], with a bias coefficient of 1.15 [23]. The dependently using the LHS technique [25]. The third matrix is

Fig. 4. Mechanical strains εm2(t) for different composite substrates: (a) W460; (b) W760; (c) WWF1400.

6
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

generated for each random variable by replacing the j-th column of M2 overlays on composite substrates are subjected to less restraint and so
with the j-th column of M1 to form the matrix M3j, for j = 1,2,…k. The the restraint-induced mechanical strains are smaller. Khan et al[11]
analysis generates three output vectors Y1(t), Y2(t) and Y3j(t) using M1, observed a similar influence of restraint rigidity in ring experiments.
M2 and M3j, respectively: Y1(t) and Y2(t) each represent the magnitude Thick overlays may not be cost-effective, however, and other con-
of the mechanical strain εm2(t) when all the variables are sampled in- trollable variables can be investigated before opting to increase the
dependently, Y3j(t) represents the magnitude of the mechanical strain overlay thickness.
εm2(t) when all the input variables are resampled except the random The correlation coefficients for the basic creep factor c1 are ap-
variable Xj. proximately zero as shown in Fig. 5d. These results suggest that there is
The first-order sensitivity index SX(t) is computed as [26] not a clear linear correlation between creep and mechanical strains, or
at least a better characterization than a linear model is needed. This
YT1 Y3j − Ns (mY1)2 finding is discussed afterwards.
SX (t) =
YT1 Y1 − Ns (mY1)2 (15) Fig. 5e shows that the correlation coefficient for the curing time t0 is
positive, being extremely important for drying periods < 10 days, i.e.
where mY1 is the mean of Y1(t).
for t < 14 days. As the drying period is increased, ρX(t) decreases ra-
For additive, monotonic and non–linear models, the sum of the
j= N pidly irrespectively of the type of substrate. The effect of curing for
sensitivity indices at any time equals one, βX = ∑j= 1 s SXj (t) = 1.
longer than 4 days is only critical during the first 10 days, suggesting
Additive models are models where the variance of the output is exactly
that the overlay may be susceptible to cracking during this time but if
the sum of the variances of each independent input variable. For models
no cracks are detected from an inspection at 10 days the effect of curing
that are not additive, the parameter βX < 1 and the reminder, 1 − βX,
is unlikely to be a cause of later cracking.
represents the fraction of the variance of the mechanical strain εm2 that
Finally, Fig. 5f shows that the correlation coefficient for the en-
cannot be explained solely on the variance of each input variable and it
vironmental humidity is negative, as expected, t > 7 days. This con-
is attributable to the combination of input variables.
firms that the restrained shrinkage phenomenon is more likely to be a
problem in environments with low relative humidity. Knowing the
5.1. Results of sensitivity analysis magnitude of the environmental humidity accurately is therefore par-
ticularly important for predicting long-term mechanical strain de-
The total number of simulations required is N = Ns(k + 2). In the mands.
current investigation, 1000 points per variable were used to conduct Fig. 6 shows the resulting sensitivity indices SX(t) when the overlay
the sensitivity analysis, resulting in 8000 simulations. acts with a concrete substrate or with a composite system for the three
Fig. 5 shows the variation of ρX(t) for different input variables for different steel beams investigated. A sensitivity index close to one in-
the four substrates considered, one concrete slab and three composite dicates the variance of the output is sensitive to variation of the par-
steel beams, over a period of 50 years of drying. The logarithmic time ticular input variable.
scale has been selected to highlight the early and late effects of each Fig. 6a shows that the sensitivity indices for βs0 are similar when the
variable on the mechanical strain demand. A positive ρX(t) indicates substrate is a concrete slab or steel beams. Over the lifetime of the
that larger values of the variable are associated with larger values of the overlay, the value of SX(t) remains constant at approximately 0.1. Thus
mechanical strain demands. roughly 10% of the variance of the mechanical strain demand is due to
Fig. 5a indicates that the correlation coefficients for the ultimate the variance of the ultimate shrinkage strain coefficient.
free shrinkage strain parameter, βs0, are quite similar for concrete or The relative high sensitivity indices for fc28 shown in Fig. 6b in-
composite steel substrates. It has a fixed positive value of approximately dicate that it is a very important variable. The sensitivity index in-
0.3 throughout the life of the overlay and consequently its impact on creases until the drying period reaches approximately 100 days when
the mechanical strain demands is significant. As expected, increasing the maximum indices of 0.5 and 0.65 occur for concrete and composite
βs0 will increase the mechanical strains in the overlay and substrate. substrates, respectively. At 1 000 days the sensitivity index reduces to
Fig. 5b shows that the correlation coefficients for the compression approximately 0.1 for composite substrates and to a value averaging
strength of the overlay are similar for concrete or composite steel 0.25 for concrete substrates.
substrates. The ρX(t) value is initially positive, for times less than seven Fig. 6c shows the sensitivity indices for the thickness of the overlay.
days (concrete substrate) or less than five days (composite steel sub- Overall, the variance of the mechanical strain demands for concrete
strate) because overlays with large fc28 will also have large moduli of substrates is slightly more influenced by ℓny than for composite sub-
elasticity and consequently large stresses due to restrained shrinkage at strates. For concrete substrates the maximum index of 0.3 occurs at
very young ages. For overlays on concrete substrates, ρX(t) becomes 40 days, whereas for composite substrates the index stays constant at
increasingly negative between 7 days and 100 days, and subsequently approximately 0.1 after 40 days.
stays approximately constant at − 0.6. For composite steel substrates, ρX Fig. 6d shows that the basic creep factor c1 does not have an effect
(t) becomes increasingly negative as t approaches 100 days, reaching a on the variability of εm2. The effect of c1 on εm2 is discussed in Fig. 7.
minimum value of − 0.75, and subsequently approaches a constant Fig. 6e shows that the variance of t0 contributes to the variance of
value of approximately − 0.2. The average correlation coefficient for the mechanical strains only during the first 10 days. At later ages its
the compressive strength of the overlay is –0.45 for a concrete substrate contribution is not significant.
and –0.31 for a composite/steel substrate. Therefore larger overlay Finally, Fig. 6f shows that the variability of the environmental hu-
compressive strengths, achieved by decreasing the water content in the midity affects the variability of the mechanical strains for drying per-
mix and so decreasing the water-to-cementing materials ratio, is pre- iods larger than 100 days. The composite substrates are more sensitive
ferred to decrease the magnitude of the mechanical strain demands. It is to the environmental humidity given that the maximum sensitivity
not cost-effective to increase the cement content to increase fc28. index of 0.75 is larger than that of 0.45 for concrete substrates. This is a
Fig. 5c shows that the correlation coefficient for the overlay thick- result of the substrate thickness. Composite systems have thin slabs that
ness is negative throughout the time of drying with a minimum mag- makes the overlay lose moisture rapidly and so the environmental hu-
nitude of − 0.1 and − 0.2 for periods between 4 day and 20 days for midity becomes an important variable, similar to creep strains.
concrete and steel substrates, respectively. At later ages the average Fig. 7 shows the variability εm2 with respect to c1 at four different
correlation coefficient is roughly − 0.3 for composite substrates and times of drying. The variation is not linear and therefore supports the
− 0.5 for concrete substrates. Overall, composite systems are slightly results of ρX(t) shown in Fig. 5d. Fig. 7 also implies that the effect of c1
less sensitive to ℓny. Thick overlays on concrete substrates or full-depth on εm2 is of high order, i.e. it is the interaction of c1 with the other

7
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

Fig. 5. Correlation coefficient ρX(t) for input variables: (a) Ultimate shrinkage coefficient; (b) concrete strength; (c) overlay thickness; (d) basic creep coefficient; (e)
curing time; and (f) environmental humidity.

factors that create the variability of εm2. The mechanical strain εm2 for concrete overlays in regions with low environmental humidity is
varies from 0 to approximately 300 μm/m and thus the creep coefficient greater.
is a significant variable. The sum of the sensitivity index SX is on average 0.64, indicating
The most important variables are the compressive strength of the that only 64% of the variance of the mechanical strain demands is ex-
overlay, the overlay thickness, and the time of curing. It is re- plained by the variance of each input variable. The remainder is due to
commended to use strong concretes (with 28-day strengths greater than interactions and non-monotonic behavior of the input variables [26].
or equal to 30 MPa) for the overlay, achieving such strengths by de-
creasing the water content and so the water-to-cementing materials 6. Summary and conclusions
ratio while maintaining a minimum cementing material content of
350 kg/m3 as recommended by the Cement Association of Canada [18]. This investigation presents a parametric and sensitivity analysis of
Thick overlays are less prone to cracking and that controlling thickness cracking of concrete overlays of various thicknesses cast on concrete
variability is effective in reducing the variability of the mechanical and composite steel substrates subjected to restrained shrinkage strains.
strain demands, particularly for concrete substrates. Although the A parametric study is performed that quantifies the importance of
overlay thickness is controllable it may not be cost-effective to use thick substrate materials, overlay material, construction practices, shrinkage
overlays as this may require otherwise unnecessary removal of sound parameters, creep parameters, environmental humidity and modeling
substrate material. Composite systems are less sensitive to the varia- variables using a range index to identify the critical variables for in-
bility of the overlay thickness and more sensitive to the environmental vestigation in the sensitivity analysis. The initial list of 21 input vari-
humidity. ables is reduced to six critical variables: the ultimate shrinkage strain
The environmental humidity is the most important uncontrollable parameter, the overlay concrete compressive strength, the overlay
variable, suggesting that the variability of the mechanical strains pre- thickness, the tension creep constant, the duration of the curing, and
dicted using THAM is greatly influenced by the variability of the en- the environmental humidity. The results of the sensitivity analysis,
vironmental humidity. Thus the uncertainty of the mechanical strains quantified using the correlation coefficient between each input variable

8
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

Fig. 6. Sensitivity index SX(t) for input variables: (a) Ultimate shrinkage coefficient; (b) concrete strength; (c) overlay thickness; (d) basic creep coefficient; (e) curing
time; and (f) environmental humidity.

and the mechanical strain demand, indicate that the overlay compres- curing times.
sive strength and its thickness and the time of curing are significant 4. The overlay thickness is negatively correlated to tensile strains in
variables and the tension creep strain constant is not significant. This is the overlay. Thin overlays are restrained by relative stiffer sub-
a very important finding because these significant variables can be strates and dry quickly, so they are more likely to crack. The use of
controlled to reduce the likelihood of the overlay concrete. thick overlays may not be cost–effective, however, as it typically
The following conclusions are drawn: requires sound substrate material to be removed.
5. The most critical uncontrollable variable is the environmental hu-
1. The most important variables that influence the mechanical strain midity. It has a strong negative correlation coefficient, –0.6, with the
demands are the overlay compressive strength, the duration of maximum overlay tensile strains for overlays cast on concrete sub-
curing and the overlay thickness. strates and –0.78 for overlays cast on steel substrates. Overlay
2. The overlay compressive strength has a negative correlation with cracking due to restrained shrinkage is therefore attributable to low
the mechanical strain demand: higher compressive strengths reduce environmental humidity: greater environmental humidity reduce
the tensile strains in the overlay due to restrained shrinkage and so moisture loss in the overlay and so reduce drastically the magnitude
the likelihood of cracking. The overlay concrete should have a of the mechanical strains that produce cracking.
minimum 28-day compressive strength of 30 MPa, as is re- 6. The tensile strain demands in the overlay are not sensitive to the
commended in current MTO practices. This minimum strength type of substrate system investigated. For example the range index
should be achieved by decreasing the water content and so the of the curing time is the same for all cases of the different substrate
water-to-cementing materials ratio, not increasing the cement con- considered in the present investigation. The average correlation
tent. coefficient for the compressive strength of the overlay is –0.45 for a
3. The duration of curing has a strong positive correlation to the mid- concrete substrate and –0.31 for a composite/steel substrate. The
depth tensile strain in the overlay during the first ten days of drying results of the sensitivity analysis are consistent for both cases con-
only: prolonged curing periods may provide other benefits but they crete and composite/steel substrates.
allow the stiffness of the overlay to increase before any shrinkage
takes place. If cracking does not occur during the first ten days of
drying, later cracking is unlikely to be attributed to prolonged

9
L. Orta and F.M. Bartlett Engineering Structures 210 (2020) 110396

Fig. 7. Mechanical strain εm2 for different values of c1: (a) t − t0 = 101 days; (b) t − t0 = 102 days; (c) t − t0 = 103 days; and (d) t − t0 = 104 days.

CRediT authorship contribution statement [11] Khan I, Castel A, Gilbert R. Tensile creep and early-age cracking due to restrained
shrinkage. Constr Build Mater 2017;149:705–15.
[12] American Concrete Institute Committee 209 (2008): Prediction of Creep, Shrinkage
L. Orta: Conceptualization, Methodology, Software, Formal ana- and Temperature Effects in Concrete Structures (ACI 209.2R–08). American
lysis, Investigation, Writing - original draft. F.M. Bartlett: Supervision, Concrete Institute, Farmington Hills, MI, 48 pp.
Writing - review & editing, Project administration, Funding acquisition. [13] Ontario Provincial Standard Specifications (2006): Construction Specification for
Structure Rehabilitation-Concrete Patches Refacing and Overlays OPSS 930.
Ontario Ministry of Transportation and Communications, 22 pp.
Declaration of Competing Interest [14] Canadian Institute of Steel Construction (2004): CAN/CSA S16–01, Handbook of
Steel Construction, Canadian Institute of Steel Construction, Willowdale, Canada.
[15] Bazant Z, Baweja S. Creep and shrinkage prediction model for analysis and design of
The authors declare that they have no known competing financial
concrete structures: model B3. The Adam Neville Symposium: Creep and Shrinkage
interests or personal relationships that could have appeared to influ- – Structural Design Effects. Farmington Hills, MI: ACI Special Publication 194,
ence the work reported in this paper. American Concrete Institute; 2000. p. 1–100.
[16] Comité Euro–International Du Béton. (1993): CEB–fib Model Code 1990 – Design
Code. Thomas Telford, London, Great Britain, 437 pp.
Acknowledgements [17] Bazant Z, Chern J. Concrete Creep at Variable Humidity: Constitutive Law and
Mechanism. Mater Struct 1985;18(1):1–20.
The research reported in this paper was made possible by the fi- [18] Cement Association of Canada (2002). Design and Control of Concrete Mixtures.
Seventh Canadian Edition, Ottawa, Canada, 356pp.
nancial support from the Natural Sciences and Engineering Research [19] American Concrete Institute Committee 318 (2008): Building Code Requirements
Council (NSERC) of Canada and the National Council on Science and for Structural Concrete and Commentary (ACI 318–08). American Concrete
Technology (CONACyT) of Mexico. The authors would like to express Institute, Farmington Hills, MI, 465 pp.
[20] Bazant Z, Najjar L. Drying of Concrete as a Nonlinear Diffusion Problem. Cem Concr
their gratitude to these organizations for their support. Res 1971;1(5):461–73.
[21] Canadian Highway Bridge Design Code (CHBDC) CSA S6–06 (2006). Canadian
References Standards Association, Toronto, Canada, 733 pp.
[22] Bissonnette B, Pigeon M, Vaysburd A. Tensile creep of concrete: study of its sen-
sitivity to basic parameters. ACI Mater J 2007;104(4):360–8.
[1] Wells J, Stark R, Polyzois D. Getting better bond in concrete overlays. Concr Int [23] Bartlett F. Assessment of concrete strength in existing structures PhD. Dissertation
1999;21(3):49–52. Edmonton, Canada: University of Alberta; 1994. p. 297 pp.
[2] Carter P, Gurjar S, Wong J. Debonding of highway bridge deck overlays. Concr Int [24] Kennedy J, Gagnon D, Allen D, MacGregor J. Canadian highway bridge evaluation:
2002;24(7):51–8. load and resistance factors. Can J Civ Eng 1992;19(6):991–1006.
[3] Silfwerbrand J, Paulsson J. Better Bonding of Bridge Deck Overlays. Concr Int [25] Iman R, Conover W. Small sample sensitivity analysis techniques for computer
1998;20(10):56–61. models, with an application to risk assessment. Commun Statist, Theory Methods
[4] Orta L, Bartlett M. Stresses due to restrained shrinkage in concrete deck overlays. 1980;A9(17):1749–842.
ACI Mater J 2014;111(6):701–10. [26] Saltelli A, Ratto M, Andres T, Campolongo F, Cariboni J, Gatelli D, et al. Global
[5] Orta L. Restrained shrinkage in bridge decks with concrete overlays PhD. sensitivity analysis. The primer. West Sussex, Great Britain: John Wiley and Sons
Dissertation The University of Western Ontario; 2009. p. 215 pp. Ltd; 2008. p. 292.
[6] Orta L, Bartlett M. Free shrinkage strains in concrete overlays. ACI Mater J
2014;111(3):263–71.
Luis Orta is Assistant Professor at Tecnologico de Monterrey, Mexico. He received his
[7] Orta L, Bartlett FM. Restrained concrete shrinkage in composite trusses. Can J Civil
Ph.D. from The University of Western Ontario. His research interests include the long-
Eng, Struct Eng 2012;39(7):779–88.
term response of concrete structures.
[8] Silfwerbrand J. Stresses and strains in composite concrete beams subjected to dif-
ferential shrinkage. ACI J 1997;94(4):347–53.
[9] Scollard C, Bartlett M. Rehabilitation criteria for post-tensioned voided-slab bridges. F. Michael Bartlett is Professor Emeritus of Civil and Environmental Engineering,
Can J Civ Eng 2004;31(6):997–1987. University of Western Ontario, London, Canada. His research interests include challenges
[10] Rahman M, Baluch M, Al-Gadhib A. Modeling of shrinkage and creep stresses in encountered when rehabilitating existing bridges.
concrete repair. ACI Mater J 1999;96(5):542–51.

10

You might also like