Inverse Analysis Tailored for UHPFRC

Svatopluk Dobrusky 1 and Gilles Chanvillard 2

Abstract: Nonlinear behavior of concrete reinforced by rebars and/or fibers has been studied over decades. In this paper, classical models for
flexural behavior are discussed and an improved model is proposed. The model is based on a modified force-based fiber-beam formulation
where progressive loading is driven by a curvature at its nonlinear hinge. Such a procedure allows capturing both softening and hardening
behavior without any further adjustments. The curvature outside of the nonlinear hinge decreases during a deflection-softening phase, and
therefore a damaged constitutive law should be introduced. Rather than using the classical damage model proposed by Mazars, a macroscopic
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.

damage model at the moment-curvature level is proposed. The moment-curvature damage model reduces the whole computation of the beam
equilibrium to only one numerical loop. Consequently, time efficiency of the proposed model is significantly improved. Hypotheses of the
new model such as damage modeling, localization, and shear deflection are discussed and finally practical applications are presented to show
the model benefits. DOI: 10.1061/(ASCE)EM.1943-7889.0001335. © 2017 American Society of Civil Engineers.

Introduction
Identification of nonlinear properties of ordinary concrete and spe-
cially fiber-reinforced concrete (FRC) or ultra-high performance
fiber-reinforced concrete (UHPFRC) is a subject of many research
projects. Uniaxial tensile and compressive tests are the most appro-
priate methods as they directly measure the desired properties.
However, the uniaxial tensile test is complicated to perform as well
as to analyze because of a localization phenomenon. Two kinds of
nonlinear descriptions are used because of the localization: strain,
and crack-opening displacement (COD). While the strain descrip-
tion is natural continuation of elastic properties, the COD descrip-
tion can be only used after initiation of a macroscopic crack.
Conversely, the COD can be measured easily with a notched speci-
men tested in uniaxial tension, whereas the strain measurement has
to be deduced from the measured displacement by assuming a char-
acteristic length. Because of a lack of sufficient information on the
characteristic length and complexity of the uniaxial tensile tests, the
complete strain description is usually deduced from indirect mea-
surements. The indirect measurements are also preferable for the
COD because the overall configuration is significantly simpler then
the uniaxial tensile tests.
Three-point and four-point bending tests are recommended
for fiber-reinforced concrete by many standards and recommenda-
tions [fib MC2010 (fib 2010); AFGC 2013; RILEM TC 162-TDF
(RILEM 2002)]. However, such indirect measurements require ad-
ditional treatment of the experimental results to evaluate the tensile
properties. The additional treatment, also called back-analysis, is di-
vided into three groups: a standard method used inversely, a simpli-
fied method used inversely, and a straightforward inverse method.
The first possible method for the posttreatment of the experi-
mental results is the standard method used inversely. It deduces
1
Dept. of UHPC, LafargeHolcim Centre de Recherche, 95, rue du Mon-
tmurier BP 15, 38291 Saint Quentin Fallavier, France (corresponding
author). E-mail: svatopluk.dobrusky@lafargeholcim.com
2
Scientific Director, LafargeHolcim Centre de Recherche, 95, rue du
Montmurier BP 15, 38291 Saint Quentin Fallavier, France.
Note. This manuscript was submitted on May 10, 2016; approved on
April 13, 2017; published online on June 30, 2017. Discussion period open
until November 30, 2017; separate discussions must be submitted for
individual papers. This paper is part of the Journal of Engineering
Mechanics, © ASCE, ISSN 0733-9399.
the material properties from an iterative procedure where the de-
sired properties are firstly estimated and then iteratively corrected
based on differences between the experimental and numerical
results. The standard method used inversely simulates the experi-
mental results with models based on finite elements [Fig. 1(e),
Bretschneider et al. 2013] or fiber-beam elements [Fig. 1(d),
Taucer et al. 1991]. The finite-element models and the fiber-beam
element models have been studied by researchers for many decades
and three primary formulations can be distinguished: a displacement-
based, a force-based, and a mixed-based formulation. Although, the
finite-element models represent a very powerful tool for versatile
modeling, they also represent the most time-consuming approach.
Because many FRC or UHPFRC applications can be simplified
into less complex structures, the fiber-beam element models are
more appropriate. The fiber-beam element model, originally devel-
oped by Taucer et al. (1991), has been recognized as one of the
most promising models for static and dynamic analyses of rein-
forced concrete frame structures since the late 1980s (Hsu and
Mo 2010). A comprehensive study on the fiber-beam element mod-
els can be found in Valipour and Foster (2007). They summarized
all the important milestones and models in this field since the very
beginning. Force-based or mixed-based formulations are generally
considered more suitable and they were used in much research
(Taucer et al. 1991; Spacone et al. 1996a; Neuenhofer and
Filippou 1997). Although the force-based or the mixed-based fiber-
beam element models are generally faster than the finite-element
models, they are still time-consuming for simply-supported struc-
tures. Therefore, simplified methods used inversely and straightfor-
ward inverse methods are more popular because of their higher time
efficiency.
The simplified method used inversely employs the same ap-
proach as the standard methods used inversely. However, because
of its simplicity, each iterative loop is shorter, and consequently, the
total computation time is significantly reduced. The simplest model
lumps the entire nonlinear deformation into one hinge and the rest
is considered to behave as a rigid body [Fig. 1(a)]. The complete
force-deflection relationship with a postpeak section can be derived
because the entire nonlinearity is lumped into one nonlinear hinge.
The replacement of the rigid body by a pure elastic behavior im-
proves predictive capacities [Fig. 1(a), Stang and Olesen 2000].
Additional improvement is obtained by adding a shear contribution
to the total deflection (Ulfkjaer et al. 1995). The latter model

© ASCE 04017102-1 J. Eng. Mech.

J. Eng. Mech., 2017, 143(9): 04017102


(a)
(b)
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.
(c)
(d)
(e)
Fig. 1. Various complexity of numerical models used for inverse ana-
lyses: (a–c) simplified models; (d) fiber beam element model; (e) finite
element model
provides very good results for materials with deflection-softening
behavior (Naaman 2008). The previous models are no longer
valid for materials with deflection-hardening behavior because
the assumption of the rigid or linear body underestimates the total
deflection. Chanvillard and Corvez (2013) proposed a new model
to avoid the underestimation of deflection, which takes into account
the nonlinear behavior of the entire body [Fig. 1(b)] with an
arbitrary deflection-hardening material. This model uses the force-
based approach and explicitly evaluates the force-deflection rela-
tionship up to the peak load. Soranakom and Mobasher (2007)
and Qian and Li (2008) proposed similar approaches assuming a
predefined stress-strain relationship.
The straightforward inverse method [fib MC2010 (fib 2010);
AFGC 2013; RILEM TC 162-TDF (RILEM 2002); Baby et al.
2013] directly connects the experimental results in bending to the
uniaxial constitutive law (stress-strain or stress-COD relationships).
The straightforward inverse methods differ in complexity: certain
methods can evaluate an arbitrary constitutive law (Baby et al.
2013), whereas others only evaluate a predefined constitutive
law [fib MC2010 (fib 2010); AFGC 2013; RILEM TC 162-TDF
(RILEM 2002)]. The primary benefit of the straightforward inverse
method is its straightforwardness, which eliminates any iterative
loops (estimation and correction of the material properties). Be-
cause of the one-way approach, the determined uniaxial properties
have to be later transferred to different modeling software to predict
the structural behavior of the intended application. This transfer
represents its weakest point because different models could be
based on different assumptions. This problem is avoided by using
the same model for the inverse analysis and the simulation.
© ASCE 04017102-2
J. Eng. Mech., 2017, 143(9): 04017102
In this paper, a method [Fig. 1(c)] that governs all the models in
the simplified method group [Fig. 1(a and b)] and gives results with
the similar accuracy as the fiber-beam models (standard methods) is
presented. Besides the similar accuracy, the method is undeniably
time efficient. The proposed model uses the theory of fiber-beam
elements with a modified force-based formulation. The modified
force-based (M–FB) formulation can overcome difficulties with
deflection-softening material (localization) and it also includes
deflection-hardening material. The M–FB model is designed for
statically determinate structures. The model also takes into account
a shear contribution to the total deflection. Because of the frame-
work of the fiber-beam elements, the M–FB model has a versatile
range of use and it can also analyze concrete beams with rebars.
The M–FB model is optimized with respect to its time efficiency,
on account of being used inversely or for reliability-based code cal-
ibrations based on Monte Carlo simulations.
Model Description
In statically determinate structures, the internal forces are known
and consequently the force-based formulation is ideal for such
structures. The primary shortcoming of the force-based formulation
is its inability to describe the deflection behavior after the onset of
the softening, unless special and hence time-consuming approaches
are used.
Fig. 2 illustrates the proposed model where the modified force-
based formulation is described. The progressive loading is driven
by the curvature of the nonlinear hinge instead of the force. This
technique has two primary advantages; the progressive loading
driven by curvature helps overcome the peak load and continues
on the softening section; secondly it can capture a snap-back effect
in the force-displacement relationship. The complete model con-
sists of three major steps that are mathematically described as
follows. Fig. 2 shows the major steps in the titled boxes. In Cross-
section equilibrium the equilibrium is solved and the corresponding
moment in the nonlinear hinge is computed. As shown in Structural
equilibrium, once the moment in the nonlinear hinge is known (see
1 and 2), the classical force-based formulation is used for the
known distribution of the moment on the rest of the beam (3 and
4). In Force displacement equilibrium the deflection of the beam is
derived from the curvature, moment, and shear force distributions.
The notable benefit of the proposed model comes from simpli-
fication of the structural equilibrium. Each determination of the
structure state has only one iterative loop at the cross section level.
This simplification is valid for an arbitrary combination of materi-
als within the cross section. One iterative loop at the cross section
level is notably faster than two nested loops (Taucer et al. 1991;
Spacone et al. 1996a) in which the innermost loop carries out
the iteration at the element level, to fulfill the equilibrium, and the
outermost loop provides the iteration at the structural level to sat-
isfy the compatibility condition (Valipour and Foster 2007). One
iterative loop at the cross section level should also be, from the
essence of the algorithm, faster than the modified nested iterative
algorithm (Neuenhofer and Filippou 1997) that bypasses the iter-
ations required to satisfy the structural equilibrium condition for
each element (Valipour and Foster 2007).
Notation of the Computation Script
The actual loading step is denoted by superscript t, e.g., ðX t Þ, the
previous loading steps are denoted by superscript τ , e.g., ðX τ Þ, the
nonlinear hinge element is denoted by subscript cr, e.g., ðX cr Þ,
the position of the elements along the span for the moment, the
curvature, and the damage variable is denoted by subscript j,
J. Eng. Mech.
 where ϵx = axial strain at the integration point (layer); ϵr = axial
strain at the top surface; and χ = curvature of the cross section.
And the uniaxial stress-strain relationship of the nonlinear
material is represented by
σx ¼ fðϵx Þ ð4Þ

Any arbitrary constitutive law combining concrete with fibers

and/or rebars can be used with such a description.
Insertion of Eq. (3) into Eq. (4) and then into Eq. (1) creates an
iterative loop that converges to the position of the neutral axis (yg )
for the given axial force (N x ) and the given curvature (χ). Once the
position of the neutral axis (yg ) is known, the corresponding mo-
ment M z is computed from Eq. (2). The modified Newton-Raphson
method was used to solve Eq. (1). The modified Newton-Raphson
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.

method worked well for all cross section layouts of many different
materials (normal concrete, UHPC, UHFRPC, and steel reinforce-
ment). Nevertheless, other arbitrary methods can be used as well.
The coupled moment and curvature are generated ðχ; MÞ
by solving the cross section equilibrium. The M–FB model is
driven by the curvature of the nonlinear hinge, therefore the cross
section equilibrium is repeated for the increasing curvatures ðχt ¼
χt−1 þ ΔχÞ. The repeated computation progressively generates
the coupled moments and curvatures that are stored in a vector form
ðχ̄ ; M̄Þ.

Structural Equilibrium
Each loading step is described by the curvature ðχtcr ¼ χt−1
cr þ ΔχÞ
and the corresponding moment M tcr at the nonlinear hinge. Con-
sidering ϕðxÞ as a moment interpolation function along the span,
the moment distribution is
Mt ðxÞ ¼ ϕðxÞM tcr ð5Þ

A link between the moment distribution and the curvature dis-

tribution is

χt ðxÞ ¼ ½M t ðxÞ ð6Þ

Fig. 2. Complete algorithm of the modified force-based formulation
e.g., ðX j Þ and the position of the fiber along the height of the cross
section is denoted by subscript k, e.g., ðX k Þ
where ÂðÞ denotes the linking operator. In the previous models
(Taucer et al. 1991; Spacone et al. 1996a; Neuenhofer and
Filippou 1997) the linking operator was the iterative loop at the
structural level. In this model, the linking operator is simplified to
a linear interpolation where the corresponding couple ðχtj ; M tj Þ is
searched in two known vectors ðχ̄ ; M̄Þ

Cross Section Equilibrium M tj ∈ ½M τcr ; M τcrþ1  ð7Þ

The cross section equilibrium in two dimensions is written as
Z M tj − M τcr
χtj ¼ χτcr þ ðχτcrþ1 − χτcr Þ ð8Þ
σx dA − N x ¼ 0 ð1Þ M τcrþ1 − M τcr
A
where fM τcr ; M τcrþ1 ∈ M̄g and fχτcr ; χτcrþ1 ∈ χ̄ g. The accuracy of the
Z linear interpolation at the structural level is improved by small
σx ðy − yg ÞdA − M z ¼ 0 ð2Þ incremental steps at the cross section level.
A When the softening occurs, the curvature outside the nonlinear
hinge decreases. A damaged unloading path must be used to satisfy
where y and yg = distances from the top surface to the integration the cross section equilibrium with the partially damaged layers of
point (layer) and to the neutral axis, respectively; σx = stress at the the material. The standard fiber-beam model would lead to nonlin-
integration point (layer); N x = axial force; and M z = bending ear computation of the cross section equilibrium for each element,
moment. where each fiber would have its own stress-strain law. Uniaxial
The compatibility requirement, for the Navier-Bernoulli theory damage of concrete is often described by two damage parameters
without a bond slip, is obtained as for compression and tension [Fig. 3(a)] as proposed by Mazars
(1984). In the M–FB model, the damage model is applied on
ϵx ¼ ϵr þ yχ ð3Þ the moment-curvature relationship [Fig. 3(b)]. This simplifies

© ASCE 04017102-3 J. Eng. Mech.

J. Eng. Mech., 2017, 143(9): 04017102

 Z L
F~ t Δdt ¼ ~ t ðxÞΔχt ðxÞdx þ ΔU tV
M ð12Þ
0

1
X~ t ¼ ðX t þ X t−1 Þ; ΔX t ¼ X t − X t−1 ð13Þ
2
where F = applied load; d = deflection under the applied load; L =
length of the beam; and ΔU tV = energy associated with the shear
forces along the length of the beam. Eq. (13) explains the notation
of Δ and tilde symbols that are used in Eq. (12) and after. The Δ
(a) (b) symbol refers to a difference of two consecutive loading steps,
whereas the tilde symbol refers to an average of two consecutive
loading steps. The shear contribution to the deflection is discussed
in depth in the following section.
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.

Alternatively, the force-deflection relationship can be derived

from a double integration of the curvature distribution.

Discussion

In the previous section, the proposed model was mathematically

described without deeper discussion of three topics that are impor-
(c) tant for the proposed model. In this section, each topic is discussed
in depth: the simplification of the structural equilibrium (moment-
Fig. 3. Damage models for concrete: (a) the standard stress-strain da- curvature damage model); a fiber-beam discretization with the
mage law (data from Mazars 1984); (b) moment-curvature damage law; localization issue; and the shear contribution to the deflection.
(c) combination of moment-curvature damage law and elastoplasticity
for the adapted model
Damage Model Applied on the Moment-Curvature
Relationship
Continuum damage mechanics were first used by Kachanov in the
nonlinear computation of the cross section equilibrium to one
context of creep rupture (Jirásek 2007) and the simplest version of
simple formula [Eq. (11)], and therefore significantly improves
continuum damage mechanics is an uniaxial damage model. In
time efficiency.
the standard fiber-beam model, two unique damage parameters
The damage moment-curvature law is written as
(Mazars 1984) are used for each fiber and then the cross section
M ¼ ð1 − ωÞke χ ð9Þ equilibrium is computed. If the cross section is composed of ma-
terials all obeying the Mazar damage law, and unloading occurs, the
where ω = damage variable; and ke = elastic bending stiffness. The axial cross section equilibrium can be written in the following dis-
growth of the damage variable is directly linked to the maximum creet form:
curvature level at the cross section X
m
  Nt ¼ Ētk ðχt−1 þ Δχt Þ½yk − ðyt−1
g þ Δyg Þ
t
ð14Þ
M j ðτ Þ k¼0
½1 − ωj ðtÞke ¼ min ð10Þ
τ <t χj ðτ Þ

Ētk ¼ Ek ð1 − ωt−1
k Þ ð15Þ
where M j ðτ Þ and χj ðτ Þ = coupled pairs of moment and curvature,
which were previously reached at the cross section. Then the actual where m = total amount of the fibers at the cross section; Ētk =
curvature is apparent stiffness of the material; ωt−1
k = damage parameter; Δχ =
  unknown unloading step; and Δyg = corresponding change of the
Mtj neutral axis. Eq. (15) is derived with the assumptions that each
χtj ¼ min ; ½M t ðxÞ ð11Þ
ð1 − ωtj Þke layer undergoes unloading and the layers where loaded elastically
in compression. Therefore, the damage parameters (ωt−1 k ) do not
The damage parameter is updated after every load step accord- change during the step. Because of the constant axial force during
ing to Eq. (10). The combination of Eqs. (9)–(11) can be explained the loading and unloading process, a difference between the pre-
as the linear damaged unloading of each element, which was found vious and actual loading steps is equal to zero and a new expression
to be similar to the model proposed by Soranakom and Mobasher is obtained
(2008). Nonetheless they proposed an elastic recovery or no cur-
X
m
vature recovery that could entrain inaccuracies in both cases. 0¼ ½Δχt Ētk ðyk − yt−1
g Þ − Ēk ðχ
t−1 þ Δχt ÞΔyt  ð16Þ
g
k¼0

Force-Deflection Relationship With

P the axial force equals to zero (majority of experimental set
Deflection at each loading step is derived in an incremental form ups), nk¼0 Ētk ðyk − yt−1
g Þ is zero, consequently Δyg must be zero
t

from the principle of energy conservation proposed by Timoshenko as well. Therefore, the moment resistance is simplified to a new
(1941) form

© ASCE 04017102-4 J. Eng. Mech.

J. Eng. Mech., 2017, 143(9): 04017102

 the yielding in the rebars and then the damage parameters were kept
unchanged. This approach ensured elastoplasticity-like behavior of
beams with steel rebars. Fig. 3(c) illustrates in a diagram form a
combined mode of the damage law and the elastoplastic law.
The previous paragraphs show a perfect correlation between the
moment-curvature relationship and the stress-strain relationship for
the damage law. Wille et al. (2014) observed the unloading modu-
lus from the peak (Epc ) on average as 8.8% of the elastic modulus
in tension (Epc =E). The direct tensile tests were carried out on sam-
ples made of UHPC matrix reinforced with three different types of
fibers and various fiber dosages. The unloading modulus to the ori-
gin according to the damage law proposed by Mazars in the case of
(a) Wille et al. (2014) would be on average 6.0% of the elastic modulus
in tension. Thus, the moment-curvature damage law represents a
good approximation for UHPFRC.
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.

Conversely, the model proposed by Soranakom and Mobasher

(2008) assumes rigid-perfectly-plastic unloading or elastic
unloading for the moment-curvature relationship. This is not in
accordance with the standard damage modes of concrete. The rigid-
perfectly-plastic unloading (no curvature recovery) and the elastic
unloading (undamaged recovery) lead to curvature locking, and
hence overestimate the postpeak deflection.

Fiber Beam Discretization and Localization Problem

(b) Eq. (12) is straightforward but also problematic because the locali-
zation occurs at the onset of the deflection softening. Numerical ap-
Fig. 4. Difference between the stress-strain damage law (data from proaches use a longitudinal discretization and therefore Eq. (12)
Mazars 1984) and the moment curvature damage law for a rectangular transforms into a discreet form
cross section 40=100 mm: (a) ordinary concrete C30=37 [data from fib X
n
MC2010 (fib 2010)]; (b) UHPFRC (data from Dobruský et al. 2013) F~ t Δdt ¼ ~ tj Δχtj dj þ M
M ~ tcr Δχtcr dcr ð19Þ
j¼0

X
m where n = total amount of the elements outside the nonlinear hinge;
M tj ¼ Ētk ðχt−1 þ Δχt Þðyk − yt−1 2
g Þ ð17Þ dj = length of these elements; and dcr = length of the nonlinear hinge.
k¼0 In this example, the rectangular rule for the numerical integration is
used and the shear contribution to the deflection is not considered.
The new form of the moment resistance helps derive the change The higher the number of elements (integration points), the shorter
of the curvature during unloading the length of elements. Consequently, the global response has spu-
ΔM tj rious mesh sensitivity. In the proposed model, an approach from
Δχtj ¼ χt−1
j ð18Þ the crack-band theory is used (Bažant and Oh 1983), where the
M t−1
j localization is lumped into one crack band with a constant effective
element size dcr equal, in this case, to the characteristic bending
Eq. (18) provides the same curvature increment as Eq. (11)
length lfl . The elements outside the crack band (nonlinear hinge)
during unloading, which verifies the moment-curvature damage
do not localize, and consequently the size of the element can be ar-
model. Fig. 4 compares the standard stress strain damage model
bitrary. The length of the crack band (nonlinear hinge) dcr is one of
and the moment-curvature damage model for two different con-
the input parameters.
cretes (C30/37 and UHPFRC). Both loading paths are identical be-
The rectangular rule for numerical integration outside the non-
cause the cross sections are composed of the materials that all obey
linear hinge is sufficiently precise because of the monotonically-
the Mazar damage law.
increasing curvature. However, Fig. 5 shows a possible error if
The new damage model significantly simplifies the whole-beam
the rectangular rule is applied on the nonlinear hinge. Another
analysis. When a beam has a constant cross section (dimensions
drawback of the rectangular rule for the nonlinear hinge is its direct
and materials), each element is represented by only one single-
proportionality to the size of the element. The length of the nonlinear
damage parameter. The damage parameter describes the state of
hinge (size of the element, dcr ) is generally empirically derived, thus
the element and it is continuously updated during the loading.
it does not have a unique value. Consequently, different standards
The common constitutive law for all elements is stored only once
and recommendations lead to different results due to the strong sen-
without any update during the analyses.
sitivity to the size of the element. In the proposed model, these two
The proposed model was originally intended for UHPFRC,
drawbacks were minimized assuming a symmetrical parabolic
which is completely within the scope of the previous paragraphs.
shape of the curvature within the nonlinear hinge as proposed by
In order to extend the scope to beams with steel rebars, the model
Casanova (1995) and used by others (Chanvillard and Roque 1999)
had to be adapted. In most reinforced beams, steel rebars play a
dominant role in its bending response, hence the unloading of such χtcr ðxÞ ¼ ax2 þ bx þ c ð20Þ
beams should follow the elastoplastic law rather than the damage
law. Thus, when the steel rebars were used in the model, the where the constants a, b, and c are derived from the parabolic boun-
moment-curvature damage model was used until the onset of dary conditions

© ASCE 04017102-5 J. Eng. Mech.

J. Eng. Mech., 2017, 143(9): 04017102

 (a)

(a) (b)
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.

(b)

Fig. 6. Shear contribution to deflection, development of a cracked area

(c) during loading; (b) decreasing shear contribution beyond the elastic
limit for a rectangular cross section 40=100 mm made of UHPFRC
Fig. 5. Possible error of the numerical integration on a nonlinear hinge: (data from Dobruský et al. 2013)
(a) rectangular rule; (b) parabolic shape; (c) sensitivity to the size of the
element of two numerical integrations for a simple supported beam
loaded at the middle 40=100=900 mm made of UHPFRC (data from
Dobruský et al. 2013) boundary condition, only the uncracked part resisted the shear
forces [Fig. 6(a)], therefore the cracked part was discarded. This
assumption considered that the cracked area would not have any ag-
gregate locking effect or fiber shear bridging effect. Such a case
d t could represent a good approximation for UHPFRC using very small
χtcr ðL − dcr =2Þ; χtcr ðL=2Þ; χ ðL − dcr =2Þ
dx cr aggregates reinforced with organic fibers. The upper bound condi-
tions assumption split Eq. (21) into a linear and a nonlinear part
Fig. 5 shows differences between the rectangular rule and the Z Z hðxÞ 2
assumed parabolic shape for UHPFRC with two different charac- V 2 Lnl h2 3V 2 L2 2 h ðxÞ − 5y~2
UV ¼ þ2 3
dxdy ð22Þ
teristic bending lengths. 20 GI z 2 G Lnl −hðxÞ
2
4 h ðxÞ

Shear Effect
The effect of shear on the deflection increases as the span-height
ratio decreases. The elastic deflection caused by shear of a simple
supported beam loaded at the middle is 3.0% for the span-height
ratio equal to 10 and 12.5% for the span-height ratio equal to 5. The
second part of Eq. (12) represents the energy associated with the
shear forces from which the shear contribution is derived. The shear
contribution to the total deflection of the beam was studied in depth
by Timoshenko (1941) in his famous book Strength of Materials.
Timoshenko proposed an approximate solution for the simple sup-
ported beam loaded at the middle [Eq. (e), p. 299]
Z LZ h 3  2 2
2 V h
UV ¼ − y~ 2
bdxdy ð21Þ
0
2
−h2 8GI z 4
where V = shear force; G = modulus of elasticity in shear; I z =
moment of inertia with respect to z axes; and y~ = distance from
the neutral axis. The height (h) in Eq. (21) can vary along the span.
Thus, two boundary conditions were studied in this paper to evalu-
ate the shear contribution to the total deflection in the nonlinear
range. In the lower boundary condition, the entire beam had a pure
elastic shear behavior, like a nondamaged beam. In the upper
where hðxÞ varied along the nonlinear section of the span [Fig. 6(a),
L1nl or L2nl ]. Fig. 6(b) shows the differences between the lower and
upper bounds, where a sinusoidal shape of the cracked area was
assumed for the upper bound.
The study of the shear contribution to the total deflection pro-
vided two major conclusions. First, the shear contribution behind
the elastic limit was continuously decreasing as the moment con-
tribution to the total deflection was becoming more and more dom-
inant. Secondly, the difference between the linear and nonlinear
behavior was insignificant. Therefore, for the purposes of simplic-
ity and time efficiency, the exact elastic solution for the simple sup-
ported beam loaded at the middle can be used [Eq. (k), p. 174,
Timoshenko 1941]
  2  3 
Ft L3 h h
dtshear ¼ 2.85 − 0.84 ð23Þ
48EI L L
Data Analysis
Three possible applications of the proposed model are presented in
this section: (1) a fast inverse analysis of unknown material; (2) a
parametric study of façade panels made of a concrete matrix,

© ASCE 04017102-6 J. Eng. Mech.

J. Eng. Mech., 2017, 143(9): 04017102

 (a)
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.
(b)
Fig. 7. (a) Inversely-analyzed stress-strain relationship (small figure)
from the experimental results of the unknown plate; (b) simplified super-
position of fibers and matrix contribution to the response of UHPFRC
organic fibers and nonmetallic reinforcement; and (3) a comparison
between the proposed model, a standard fiber-beam model, nonlin-
ear finite-element model, and experimental results of UHPFRC
beams with steel rebars.
Inverse Analysis
Identification of nonlinear concrete parameters represents an im-
portant role in material research. A task of this study was to identify
nonlinear properties of a material from which a set of plates (500/
150/13 mm) were produced. The proposed model was used to
inversely analyze the experimental results obtained by three-points
bending tests. The tests were performed at LafargeHolcim Re-
search and Development Center according to the UHPFRC recom-
mendations (AFGC 2013) where the piston displacement was
continuously increased until the complete failure. The plate dis-
placement was measured at the middle of the span by linear var-
iable displacement transducers (LVDT). After the experimental
testing, the input properties of the material were randomly gener-
ated and the numerical simulations and the experimental results
were compared on the basis of the root-mean square error. The
entire process was automated by a computer script using simulat-
ing annealing and it successfully converged to the real unique
material parameters without any additional intervention over a very
short period of time (a few minutes for the whole convergence and a
few seconds for one force-deflection simulation). Fig. 7(a) shows a
good capacity of the proposed model to capture the real response.
The inversely-analyzed stress-strain relationship had an expected
shape for UHPFRC as it is shown on Fig. 7(b). The tensile resis-
tance of the matrix is relatively low (compared to other UHPC ma-
trices) but the high volume of fibers (σ2 > σ1 ) helps to increase the
overall resistance of the tested plates.
Parametric Study
The parametric study focused on identification of a contribution of
different materials in a composite panel. The studied panels
© ASCE 04017102-7
J. Eng. Mech., 2017, 143(9): 04017102
(a)
(b)
Fig. 8. (a) Numerical simulation of façade panels made of UHPFRC
with nonmetallic reinforcement; (b) numerical and experimental para-
metric study with façade panels made of UHPC, NR: UHPFRC without
reinforcing mesh, NF: UHPC with reinforcing mesh
(500=150=15 mm) were composed of a UHPC matrix, organic fi-
bers and two layers of a symmetrical nonmetallic reinforcing mesh
placed 5 mm from each side. Firstly, the complete composite panels
(concrete matrix, fibers, and reinforcement) were tested in four-
point bending. The tests were conducted at LafargeHolcim
Research and Development Center according to the UHPFRC rec-
ommendations (AFGC 2013) where the piston displacement was
continuously increased until the complete failure. The panel deflec-
tion was measured at the middle of the span by LVDTs. After the
experimental testing, the complete composite panel was inversely-
analyzed with the proposed model. Fig. 8(a) shows the experimen-
tal results and the fitted numerical equivalent of the complete
composite panel. Once the properties of each material were known
(inversely-analyzed), some components of the composite panel
(fibers or reinforcing mesh) were removed to numerically simulate
various responses. After the numerical simulations, auxiliary exper-
imental tests, with the identical testing conditions, were carried
out to confirm the simulations (matrix-fibers and matrix-mesh).
Fig. 8(b) shows very good conformity between the experimental
results and the numerical simulations. The study proved a good
capacity of the proposed model to simulate different behaviors
without additional experimental testing.
Comparison between Numerical Models
The comparison was focused on behavior of UHPFRC beams
reinforced with organic fibers and a steel rebar. At the beginning,
one rectangular beam (40=60=900 mm) was produced and tested in
three-point bending by continuously increasing the piston displace-
ment until failure. The beam deflection was measured at the middle
J. Eng. Mech.

(a)
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.
(b)
Fig. 9. (a) Comparison of the modified fiber beam model, the standard
fiber beam element model (data from SOFiSTiK AG) and the 2D plane
stress model (data from Cervenka et al. 2012) of reinforced UHPFRC;
(b) force-displacement and strain-displacement relationship of rein-
forced UHPFRC (80=40=900 mm)
of the span by LVDT. In the next step, five other beam dimensions
(from 80=40=900 to 160=60=900 mm) were produced and tested
with the identical conditions to the first beam. The production and
the tests were carried out by the Université Claude Bernard Lyon 1
(Dobruský et al. 2013). The experimental results from the first
beam were used to identify (inversely-analyze) the UHPFRC
properties. The steel bar properties were taken from the literature
[fib MC2010 (fib 2010)] for the inverse analyses. Then, the
identified material properties were used to simulate the behavior
of five other beam dimensions. The numerical simulations were
performed with three models: a nonlinear finite-element model
(Cervenka et al. 2012), a standard fiber-beam model (SOFiSTiK
AG), and the proposed model. Fig. 9(a) shows a good conformity
between the numerical models. The small differences between the
numerical simulations and the experimental results were attribut-
able to the input material properties that were inversely-analyzed
from the first beam. Nevertheless, the good conformity was ob-
served at the onset of cracking, at the onset of yielding and also
for the stiffness behavior, even though the postpeak behavior
differed for each model (different treatment of the localization
phenomenon).
This study provided another possibility to verify the robustness
and precision of the proposed model. Strain gauges were placed on
the steel bars and the strain was measured during the tests. Fig. 9(b)
shows the striking conformity between the experimental and
numerical results of the strain-displacement and force-displacement
relationship, although only the latter was fitted and the first
observed.
© ASCE 04017102-8
J. Eng. Mech., 2017, 143(9): 04017102
The proposed model was developed for practical purposes and
the key driver was the time-efficiency, specifically for inverse
analyses and reliability computations. The comparison between
the numerical models (Dobruský et al. 2013) proved excellent time
efficiency of the proposed model, when the simulation of one beam
lasted 5–10 s (121 elements with 100 layers) compared to 285 s of
the fiber-beam model (10 elements), and 90 min of the 2D plane
stress model (90 × 10 elements).
Conclusions
Prior work documented the effectiveness of the force based fiber-
beam model for a broad range of concrete structures. For example,
Spacone et al. (1996b) presented an application of the fiber-beam
element model for simulations of the hysteric behavior of rein-
forced concrete beams and columns. Nevertheless, the force-based
fiber-beam model is still superfluously robust and time-consuming
for most experimental set ups, although it is generally faster than
the finite-element models.
Therefore, simplified models are often used to avoid time inef-
ficiency. However, none of the simplified models can compete
with the standard models, namely for the deflection-hardening
materials, even though the models differ in their complexity.
The primary shortcomings of the simplified models are: the
deflection-hardening behavior, the numerical treatment of the
localization and the shear contribution. The latter has not yet been
studied for such materials.
The proposed model surpasses all the simplified models and
gives the results with the similar accuracy as the fiber-beam models
(standard methods) with undeniable time efficiency. The time effi-
ciency is attributable to the fact that the proposed model works in
the moment-curvature space directly, thus avoiding the computa-
tional costs related to section integration that has to be performed
with the fiber-beam models. The modified force-based formulation
driven by its curvature overcomes difficulties for the deflection-
hardening behavior and the subsequent deflection-softening behav-
ior. Moreover, it can capture a snap-back effect directly without
any additional treatment. Possible errors from the numerical inte-
gration are minimized with the assumed shape of the curvature in
the nonlinear hinge. The decreasing tendency of the shear contri-
bution to the total deflection is presented for deflection-hardening
materials.
This paper showed the very good conformity between the pro-
posed model and other numerical models [Fig. 9(a)], the notable
conformity between the numerical simulations and the experimen-
tal results [Fig. 9(b)], and the undeniable time efficiency compared
to other numerical models during the comparative study. The time
efficiency and the accuracy provide a broad range of applications
for experimental and statistical work.
In addition to the mentioned findings, the moment-curvature
damage law was verified and its suitability for UHPFRC was dis-
cussed. The moment-curvature damage law is equivalent to the
standard damage law proposed by Mazars (1984) for the fiber-
beam element models. This finding may significantly accelerate
standard fiber-beam element models.
Acknowledgments
The support of the Grant Agency of the Czech Republic—Grant
Project No. 16-04454S “Uncertainties on the material level affect-
ing those on the structural level” is gratefully acknowledged by the
author.
J. Eng. Mech.
 References
AFGC (Association Francaise de Genie Civil). (2013). “Ultra high perfor-
mance fibre-reinforced concretes—Recommendations.” Paris.
Baby, F., Graybeal, B., Marchand, P., and Toutlemonde, F. (2013).
“UHPFRC tensile behavior characterization: Inverse analysis of four-
point bending test results.” Mater. Struct., 46(8), 1337–1354.
Bažant, Z., and Oh, B. (1983). “Crack band theory for fracture of concrete.”
Mat. Struct. (RILEM, Paris), 16, 155–177.
Bretschneider, N., Villmann, B., and Slowik, V. (2013). “On the characteri-
zation of strain hardening cement-based materials by inverse analysis of
bending tests.” Fracture Mechanics of Concrete and Concrete Struc-
tures: Proc., 8th Int. Conf. on Fracture Mechanics of Concrete and
Concrete Structures (FraMCoS-8), held in Toledo, Spain, 11-14 March
2013, J. van Mier, G. Ruiz, C. Andrade, R. Yu, and X. Zhang, eds.,
CIMNE, International Center for Numerical Methods in Engineering,
Downloaded from ascelibrary.org by Tufts University on 08/01/17. Copyright ASCE. For personal use only; all rights reserved.
Barcelona, Spain, 2056–2065.
Casanova, P. (1995). “Bétons renforcés de fibres métalliques: Du matériau à
la structure.” Ph.D. thesis, l’Ecole Nationale des Ponts et Chaussées,
Champs-sur-Marne, France.
Cervenka, V., Jendele, L., and Cevenka, J. (2012). “ATENA program docu-
mentation. 1: Theory.” 〈http://www.cervenka.cz/assets/files/atena-pdf
/ATENA_Theory.pdf〉 (Mar. 1, 2014).
Chanvillard, G., and Corvez, D. (2013). “Explicit back analysis method for
quick determination of direct tensile strength of plate structural mem-
bers.” PRO 87: UHPFRC 2013, F. Toutlemonde and J. Resplendino,
eds., RILEM Publications, Paris, 659–668.
Chanvillard, G., and Roque, O. (1999). “Behavior of fibre reinforced con-
crete cracked section under sustained load.” PRO 6: 3rd Int. RILEM
Workshop on High Performance Fiber Reinforced Cement Composites,
H. Reinhardt and A. E. Naaman, eds., RILEM Publications, Paris,
239–250.
Dobruský, S., Lussou, P., Chanvillard, G., Perez, F., and Corvez, D. (2013).
“Flexural behaviour of UHPC beams reinforced with organic fibers and
steel bars.” PRO 87: UHPFRC 2013, F. Toutlemonde and J.
Resplendino, eds., RILEM Publications, Paris, 659–668.
fib (International Federation for Structural Concrete). (2010). “Model code
2010—First complete draft, volume 1.” fib MC2010, Lausanne,
Switzerland.
Hsu, T., and Mo, Y. (2010). Unified theory of concrete structures, Wiley,
New York.
Jirásek, M. (2007). “Nonlocal damage mechanics.” Revue Européenne de
Génie Civil, 11(7–8), 993–1021.
Mazars, J. (1984). “Application de la mécanique de l’endommagement au
comportement non linéaire et à la rupture du béton de structure.” Ph.D.
thesis, Université Paris VI, Paris.
© ASCE 04017102-9
J. Eng. Mech., 2017, 143(9): 04017102
Naaman, A. E. (2008). “Chapter 3: High performance fiber reinforced ce-
ment composites.” High-performance construction materials: Science
and applications, C. Shi and Y. Mo, eds., World Scientific, Hackensack,
NJ, 91–153.
Neuenhofer, A., and Filippou, F. (1997). “Evaluation of nonlinear frame
finite-element models.” J. Struct. Eng., 10.1061/(ASCE)0733-9445
(1997)123:7(958), 958–966.
Qian, S., and Li, V. C. (2008). “Simplified inverse method for determining
the tensile properties of SHCC.” J. Adv. Concr. Technol., 6(2),
353–363.
RILEM. (2002). “Design of steel reinforced concrete using the σ-w
method: Principles and applications.” Mater. Struct., 36(262), 560–567.
SOFiSTiK AG 27.11 [Computer software]. 3D FEM Professional, Obers-
chleissheim, Germany.
Soranakom, C., and Mobasher, B. (2007). “Closed-form solutions for
flexural response of fiber-reinforced concrete beams.” J. Eng. Mech., 10
.1061/(ASCE)0733-9399(2007)133:8(933), 933–941.
Soranakom, C., and Mobasher, B. (2008). “Correlation of tensile and flexu-
ral responses of strain softening and strain hardening cement compo-
sites.” Cem. Concr. Compos., 30(6), 465–477.
Spacone, E., Filippou, F. C., and Taucer, F. F. (1996a). “Fibre beam-column
model for non-linear analysis of R/C frames. I.” Earthquake Eng. Struc.
Dyn., 25(7), 711–725.
Spacone, E., Filippou, F. C., and Taucer, F. F. (1996b). “Fibre beam-column
model for non-linear analysis of R/C frames. II.” Earthquake Eng.
Struct. Dyn., 25(7), 727–742.
Stang, H., and Olesen, J. (2000). “A fracture mechanics based design ap-
proach to FRC.” PRO 15: Fibre–Reinfoced Concretes (FRC), P. Rossi
and G. Chanvilard, eds., RILEM Publications, Paris, 315–324.
Taucer, F., Spacone, E., and Filippou, F. (1991). A fiber beam-column
element for seismic response analysis of reinforced concrete structures,
Vol. 91, Pacific Earthquake Engineering Research Center, College of
Engineering, Univ. of California, Berkeley, CA.
Timoshenko, S. (1941). Strength of materials. Part I: Elementary theory
and problems, 2nd Ed., D. Van Nostrand Company, Inc., New York.
Ulfkjaer, J., Krenk, S., and Brincker, R. (1995). “Analytical model for ficti-
tious crack propagation in concrete beams.” J. Eng. Mech., 10.1061
/(ASCE)0733-9399(1995)121:1(7), 7–15.
Valipour, H., and Foster, S. (2007). “A novel flexibility based beam-column
element for nonlinear analysis of reinforced concrete frames.” UNICIV
Rep. R, Univ. of New South Wales, New South Wales, Australia, 117.
Wille, K., El-Tawil, S., and Naaman, A. E. (2014). “Properties of strain
hardening ultra high performance fiber reinforced concrete (UHP-
FRC) under direct tensile loading.” Cem. Concr. Compos., 48, 53–66.
J. Eng. Mech.