Integrative Sensitivity Analysis Applied to Semi-Integral

Concrete Bridges
Bastian Jung1; Guido Morgenthal2; and Dong Xu3

Abstract: In numerical simulations of engineering structures, several physical phenomena are represented by partial models, and the selection
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

of the important partial models is usually based on engineering judgment. The quantitative assessment of how these partial models influence
various response quantities is possible using variance-based sensitivity analysis. However, this provides information only on local positions in
the structure. The present paper extends the sensitivity analysis at local positions to the integrative sensitivity analysis of the entire structural
load-bearing behavior. In addition to the assessment of the partial model’s sensitivity, the method includes the local response significance factor,
which relates the sensitivity to the response significance at each position in the structure. The integrative sensitivity analysis is applied to the
numerical simulation of semi-integral concrete bridges to evaluate the partial model’s influence with respect to the entire integral structure. The
quantitative assessment of the integrative sensitivity for varying pier heights and different loading conditions in the serviceability limit state and
the ultimate limit state allows conclusions with practical engineering relevance. DOI: 10.1061/(ASCE)BE.1943-5592.0000588. © 2014
American Society of Civil Engineers.
Author keywords: Sensitivity analysis; Model quality assessment; Semi-integral concrete bridge; Global model; Partial model.

Introduction increased? The qualitative model selection is not capable to rec-

ommend detailed answer to such questions.
The load-bearing behavior of engineering structures is mainly in- In contrast, the quantitative model evaluation assists the model se-
fluenced by geometric conditions, material properties, environmen- lection process in numerical simulations to reduce the modeling un-
tal, and loading conditions. In numerical simulation, the analyses certainty. The analysis of the phenomena‘s sensitivity on the structural
of global structural models GM induces the coupling of different load-bearing behavior illustrates where the engineer should use more
phenomena, represented by partial models PM. The description of accurate (high sensitivity) or simplified models (low sensitivity). This
material behavior, creep, and shrinkage of concrete and pile foun- analysis can significantly reduce the uncertainty in the model pre-
dation stiffness are some examples of PM. In several project phases, diction if the engineer selects partial models with high pre-
the engineer has to decide which phenomena should be considered diction quality for sensitive phenomena. Hence, this clear insight in
in the structural model. Subsequently, the engineer chooses a certain the structural behavior and the subsequently quantitative model se-
partial model for each phenomenon. That decision-making process is lection can achieve a more reliable approach for the structural design.
often based on engineering knowledge and experience (qualitative The designs of engineering structures based on models with high
model selection). In general, a large set of other models exists where- prediction quality are more adequate in describing the load-bearing
by the complexity and accuracy vary considerably for representing the behavior. Therefore, the probability of occurrence of structural
several phenomena. damages during a lifetime, which generally lead to high mainte-
Consequently, some questions arise. What is the influence of a nance, repair, and refurbishment costs, can be clearly reduced. In
certain phenomenon in comparison with other phenomena accord- addition, the quantitative comparison of a few global structural
ing to the entire structural load-bearing behavior? For which phe- models can increase the confidence in the computed results and fi-
nomenon should the engineer take into account more complex and nally in the entire structural design.
accurate models in the global structural model? When can the en- In a global model with subordinate partial models, the prediction
gineer choose simplified models? Is it possible to neglect a certain quality of each of the partial models MQPM has an influence on the
phenomenon? How can the uncertainty in the model prediction overall prediction quality of the global model MQYGM . Variance-
be reduced? How can the reliability of the structural design be based sensitivity analysis (Sobol 1993; Homma and Saltelli 1996)
allows for the quantification of the importance of each partial model
with respect to the structural response value Y at a certain position in
1
Ph.D. Student, Research Training Group 1462 “Model Quality,” the structure. Keitel et al. (2011) introduces a methodology for the
Bauhaus-Universität Weimar, 99423 Weimar, Germany (corresponding evaluation of coupled numerical partial models in structural engi-
author). E-mail: [email protected] neering. However, the focus was the development of the evaluation
2
Professor, Dept. of Modelling and Simulation of Structures, Bauhaus- methodology and the application to local structural response values.
Universität Weimar, 99421 Weimar, Germany. E-mail: guido.morgenthal@ This global model quality assessment at certain positions in
uni-weimar.de the structure is particularly valuable for the design of engineering
3
Professor, Dept. of Bridge Engineering, Tongji Univ., Shanghai 200092,
structures with a linear-elastic computation of the section forces and
China. E-mail: [email protected]
Note. This manuscript was submitted on July 17, 2013; approved on subsequent dimensioning of critical cross sections. However, the
November 25, 2013; published online on January 16, 2014. Discussion material nonlinear simulation of structures does not include the
period open until June 16, 2014; separate discussions must be submitted for design of certain cross sections. The design of the entire structure is
individual papers. This paper is part of the Journal of Bridge Engineering, always checked implicitly by the constitutive models at all positions.
© ASCE, ISSN 1084-0702/04014014(13)/$25.00. Instead of the local safety check of cross sections, a global safety

© ASCE 04014014-1 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 check for nonlinear simulations should be performed on a structural serviceability limit state (SLS) and the ultimate limit state (ULS) are
level rather than in local sections (Cervenka 2013; Allaix et al. taken into account in the structural behavior assessment using the
2013). Following the safety level concept for nonlinear simulations, proposed integrative sensitivity analysis.
the model quality assessment cannot be exclusively referred to the
partial model’s sensitivity at local positions in the structure, par-
ticularly for the material nonlinear simulations. Therefore, it is
Evaluation of Global Numerical Models
necessary to establish a sensitivity analysis that is capable of the
evaluation of the entire structure, taking into account all positions
On the geometric level, the numerical simulation of structures in-
in the structure.
troduces a subdivision of the entire structure into structural com-
Furthermore, because of complicated and interactive conditions,
ponents. On a conceptual level, however, these parts are then
it might be difficult to clearly identify the positions to be assessed for
modeled by suitable partial models, which mathematically describe
the structural load-bearing evaluation. The false position identifi-
different phenomena such as the time-dependent behavior of con-
cation will result in a model evaluation at positions in the structure
crete or the stress-strain relationship of the surrounding soil. The
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

with low significance for the design of the entire structure. More-
coupling between the PM integrates the phenomena of the structu-
over, the sensitivity is strongly dependent on the position in the
ral components into the global model of the entire structure. To
structure (Jung et al. 2013; Keitel et al. 2011). Therefore, the pre-
quantify the global model prediction quality (Keitel et al. 2011), the
diction quality for the same partial model combination varies
first evaluation step is the quantification of whether a phenomenon
similarly from position to position. The authors believe that the
has an influence on a certain structural response value or the phe-
quality assessment for each partial model combination should take
nomenon can be neglected in the numerical simulation, respectively
into account the entire structure, especially for practical engineering
modeled with simplified partial models.
problems. Based on this overall assessment, a global structural
Therefore, variance-based global sensitivity estimators allow for
model with an adequate prediction quality can be selected to obtain
the quantitative assessment of the phenomenon’s importance in
reliable prognoses and a safe design of the entire structure.
relation to the structural response quantities Y (e.g., displacements,
The requirement of the partial model’s sensitivity on a structural
stresses). In the evaluation of the global model quality, the sensi-
level for nonlinear simulations and an overall assessment and the
tivity indexes are used for the quantification of the phenomenon‘s
possible false position identification in the structure justify the
importance. By using discrete random variables for activating or
necessity to advance the existing sensitivity analysis. The present
deactivating a certain PM, the sensitivity analysis in this case is not
paper extends the sensitivity analysis at local positions in the
an estimation of the uncertainty in a model output. The first-order
structure to the integrative sensitivity analysis of the entire structure.
sensitivity index Si (Sobol 1993) can be used to compute the ex-
This enhancement considers the assessment of the partial model’s
clusive influence of a certain PM on the structural response of the
importance in relation to the overall structural load-bearing be-
global model
havior. Therefore, the integrative sensitivity analysis is proposed in
this paper to assess the global model quality with respect to the entire
V½EðY j Xi Þ
structure. Hence, the decision making in selecting an adequate Si ¼ (1)
partial model combination for the entire structure is assisted by the VðYÞ
proposed integrative sensitivity analysis.
Many global structural models are already available for different where V½EðY j Xi Þ 5 variance of the expected value of Y when
types of engineering structures. The coupling between numerous conditioning with respect to Xi and VðYÞ is the unconditional
partial models is found to be more important for structures that variance of Y. Because of interactions in complex engineering
cannot be decoupled into the structural components for analysis problems, higher-order sensitivity indexes are needed. The total
purposes. For instance, the decoupling of the structural components effects sensitivity index STi (Homma and Saltelli 1996) calculates
is not an appropriate assumption for the numerical simulation of the overall influence of a specific partial model while also consider-
integral bridges, where numerous phenomena and their interactions ing the interaction with the other PM in the global model
have to be considered (Marx and Seidl 2011; Zordan et al. 2011).
The interplay between the stiffness of the structural components, V½EðY j X ∼i Þ
STi ¼ 1 2 (2)
such as the superstructure, the piers, the foundation, and the soil, VðYÞ
highly influences the design, especially in the case of restraint-
sensitive integral bridges (Chacòn et al. 2013; Faraji et al. 2001; where V½EðY j X ∼i Þ 5 variance of the expected value of Y when
Krizek 2011). Remarks referring to the numerical analysis of integral conditioning with respect to all parameters except for Xi . In general,
abutment bridges have been studied by Bloodworth et al. (2012), the total effects sensitivity index STi should be used for the quan-
Dicleli and Erhan (2010), Huang et al. (2008), Kim and Laman titative evaluation of the importance of various phenomena in en-
(2010), and Ooi et al. (2010). A common result is that integral gineering structures. A finite number of possible model class (partial
bridges cannot be decoupled into the structural components because models) combinations NP 5 2M are necessary for the assessment of
of the interaction between the partial models. The quantification the sensitivity indexes, whereby the partial models representing the
of each partial model’s influence on the overall structural behavior same physical phenomenon are related to the equivalent model class.
has not yet been studied. The value M is defined as the number of model classes in the global
Therefore, the integrative sensitivity analysis is applied to semi- model. For each model class, a partial model with an adequate
integral concrete bridges to emphasize the applicability of the method complexity and accuracy has to be used for the sensitivity evaluation
for structures with interactive structural components. In the present in the first step. High values of the sensitivity indexes show a sig-
paper, piers with varying cross-sectional stiffness and heights are nificant influence of the model class on the global model response at
considered to analyze the interaction between the structural compo- a certain position in the structure. Model classes with sensitivity
nents for different geometric conditions. It should be noted that the indexes smaller than a given threshold (e.g., SM Ti , 0:03) shall be
load condition also has a significant influence on the response of neglected for the next evaluation method step (Keitel et al. 2011).
the structure and the partial model’s sensitivity. Therefore, the In other words, no further investigations about their partial model

© ASCE 04014014-2 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 quality are necessary. The unimportant model classes are excluded Integrative Sensitivity Analysis
for the subsequent evaluation steps of the global model quality
assessment. The total effects sensitivity indexes corresponding to model class
The second step quantifies the model choice influence of certain SM MC
Ti and model choice STi can be assessed at each nodal coordinate
partial models for the same phenomenon in relation to the other or in each element in the numerical model of the entire structure. The
phenomena. The model choice assessment SMC Ti is similarly com- importance of the model class and the influence of the model choice
puted using the variance based sensitivity analysis. In comparison on the phenomenon of interest are strongly dependent on the posi-
with the first step, the phenomena are not constrained to be activated tion and conditions within the structure (Jung et al. 2013). These
or deactivated; they are analyzed in respect to certain models in the variance-based sensitivity values are independent of the structural
second evaluation step. For example, in the case that the creep of response magnitude at each position. Hence, the significance of
the concrete is an important phenomenon (in the first step: STi,creep each position in the global numerical model of the structure is not
$ 0:03), the second step quantifies the influence of selecting a considered in the variance-based sensitivity analysis. In this context,
certain creep model. This is in relation to the model choices of all the the integrative sensitivity analysis is proposed in this paper. This
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

other sensitive phenomena (e.g., shrinkage, material models, pile quantitative method describes the influence that model class or model
foundation stiffness). For the creep and shrinkage phenomenon of choice has on the overall structural load-bearing behavior. Conse-
concrete, the models according to Model Code 2010 (International quently, the assessment of coupled numerical partial models does
Federation for Structural Concrete 2012), Gardner and Lockman not refer to a local response quantity at a particular position in the
(2001), Bazant and Bajewa (1995), and the American Concrete structure anymore, but rather the integrative sensitivity analysis
Institute (ACI 1992) can be used as the partial models to quantify assesses the quality in relation to the entire load-bearing behavior of
the influence of the model choice. the structure.
The total effects sensitivity index quantifies the effect that the The sensitivity indexes at local positions in the structure are re-
model choice has on the global model’s response at a particular lated to various response quantities Y (e.g., displacements, stresses,
position in the structure. On one hand, low values express that the and section forces). The numerical integration of the sensitivity
choice of different partial models representing the same phenom- indexes (SM MC
Ti , STi ) at each position over a chosen region lj , e.g.,
enon predict a similar partial model output. On the other hand, the the length of structural components (e.g., lsuperstructure , lsubstructure ,
choice of a PM in this model class does not significantly affect the lstructure ), leads to an unweighted overall assessment of the sensitiv-
structural load-bearing behavior. The sensitivity indexes are used as ity indexes. The importance of the response values at each node is
weighting factors for the partial model quality MQPM in the as- assumed to be equally important for the evaluation of the coupled
sessment of the numerical global model. The global model quality numerical partial models. For example, the prediction of the vertical
MQYGM of coupled partial models is quantified by a path through displacement of a simply supported beam is much less important
a graph (Kaveh 2004), with the vertex as the quality of the partial near the supports than near the center of the span. In contrast, the
model MQPM and the edges as the coupling quantities. MQPM is unweighted numerical integration assumes an equal importance in
assessed from the evaluation of the partial model itself, using un- the prediction of both displacements.
certainty, complexity, or robustness criteria (Bazant and Li 2008; To consider the significance of the position in the structure, the
Most 2011; Keitel 2013). Assuming a perfect data coupling between unweighted numerical integration of the sensitivity indexes over
the partial models, the model quality of a global structural model is a certain length is advanced with the help of the proposed integrative
then defined as (Keitel et al. 2011) sensitivity analysis. The integrative total effects sensitivity index
M=MC
STi can be assessed for specific structural components or the
NPP,red S
MC × MQ entire structure including all of its components and elements.
Ti PMj
MQYGM ¼ PNP,red MC (3) Therefore, the importance of the sensitivity indexes at the various
i51 i51 STi positions is expressed by the local response significance factor S Y .
For the overall structural assessment, the integrative total effects
where MQPMj 5 partial model quality of a certain partial model j sensitivity index considers the required sensitivity and significance
in the model class M. The variable NP,red , determined by the first information and is defined as
evaluation step, is the number of considerable model classes
influencing the global response quantity, and SMC ðj
x5l
Ti is the total effects M=MC,Y M=MC,Y
sensitivity index for the model choice assessment. It is obvious STi,Xi ¼ STi,Xi ðxÞ × S Y ðxÞdx
that the sensitivity indexes strongly influence the structural model x50
quality. PN
ðnÞ × 1 × ðxnþ1 2 xn21 Þ × S Y ðnÞ
M=MC,Y
Because of the interaction between the structural components, n51 STi,Xi 2
the sensitivity indexes are strongly varying with respect to the po- ¼ PN 1 (4)
sition in the structure. Consequently, the global model quality for n51 2 × ðxnþ1 2 xn21 Þ × S Y ðnÞ
the same partial model combination also changes from position to
position. Therefore, the existing sensitivity analysis at local posi- where S Y ðnÞ 5 local response significance factor at position n in
M=MC,Y
tions in the structure is enhanced to the integrative sensitivity the structure; N 5 total number of all positions; and STi,Xi 5 total
analysis. This improvement allows the overall assessment of the effects sensitivity indexes for each partial model Xi with respect to
phenomena‘s importance in relation to the entire structural load- the structural response quantity Y. The integrative sensitivity in-
bearing behavior taking into account the response significance at dexes have to be separately quantified in the model class M (first
each position in the structure. Finally, for each partial model step) or model choice MC (second step) assessment. The sensitivity
combination, a global prediction quality can be assessed that analysis quantifies which model causes the variance in the structural
quantitatively analyzes the entire structure. Therefore, the quality model response. In addition, the local response significance factor
assessment is no longer dependent on the position to be evaluated, S Y considers the importance of the prediction at each node or
because all positions can be considered simultaneously in the element in comparison with the overall structural behavior. Thus, it
proposed integrative sensitivity analysis. is a parameter that requires a definition according to engineering

© ASCE 04014014-3 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 considerations, which may be different for different response design recommendations then can be assessed in a quantitative
quantities. In this paper, two different definitions of S Y are proposed. manner by applying the proposed integrative sensitivity analysis for
On one hand, the factor can be defined as the ratio between the various structures and conditions.
maximum/minimum response value at a particular node (or element)
Y min=max ðnÞ of the numerical model and the maximum/minimum Semi-Integral Concrete Bridges and
response value of all nodes (or elements) over a chosen length of the Model Assumptions
min=max
structural components Ylj [Eq. (5)]. On the other hand, the
maximum or minimum response value at a particular node (or el-
ement) can also be related to a permissible value Yperm , such as Geometry and Material Properties
permissible material stresses defined in design codes or guidelines, The bridge considered here consists of three spans with corre-
rendering it similar to a utilization ratio [Eq. (6)] sponding span lengths of L1 5 L3 5 40 m (side spans) and
Y min=max ðnÞ L2 5 48 m (midspan). The geometry of the prestressed concrete
S Y,rel ðnÞ ¼ (5) bridge and the tendon profile are illustrated in Fig. 1. Two different
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

min=max
Ylj pier heights of H 1 5 5 m and H 2 5 10 m are considered in com-
bination with unmodified geometrical conditions of the super-
structure. For the pier foundations, single rows of piles in the
Y min=max ðnÞ
S Y,perm ðnÞ ¼ (6) transverse direction of the bridge are chosen according to the pre-
Yperm dimensioning. Each pile row consists of four piles with a length of
Lp 5 25:0 m, a diameter of Dp 5 1:0 m, and a clear spacing of
The material nonlinear response of structures and the probably s 5 1:0 m. This small clear spacing induce a high interaction be-
complex conditions in structures cause the necessity of the quanti- tween the piles in the group. The pile partial models representing the
fication of the partial model’s influence with respect to the entire foundation stiffness consider these interactions. The cross section of
structural behavior. The integrative sensitivity analysis connects the the superstructure is a prestressed single box girder with a width of
sensitivity indexes at each position in the structure with the response 14.50 m and a depth of 2.35 m (Fig. 1). The width is a standard cross
significance at the corresponding position. The numerical inte- section of a three-lane road with an additional lane for one direction
gration of both quantitative information over a chosen length, for (2 1 1 system). The slenderness ratio l 5 L=h of the prestressed
example, the length of the entire structure, enables the necessary concrete box girder is l1 5 l3 5 14 and, respectively, l2 5 13
overall assessment of the partial model’s sensitivity. In principle, the (medium slenderness).
integrative sensitivity analysis enables the categorization of engi- The cross-sectional geometry of the piers with the different pier
neering structures according to the importance of various phenomena, heights is predesigned using the pier (respectively column) slen-
which should be taken into account in the numerical simulation. The derness ratio lcol 5 l0 =i, with l0 5 Hpier , b as the effective column

Fig. 1. (a) Longitudinal geometry of semi-integral concrete bridges and tendon eccentricity with respect to box girder center line; (b) isometric view
of semi-integral concrete bridge and cross-sectional geometry of the box girder and the piers

© ASCE 04014014-4 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 pffiffiffiffiffiffiffiffi
length, and i 5 I=A as the radius of gyration of the uncracked decompression is checked in the quasi-permanent load combination
concrete cross section. In the range of 22 , lcol , 100, the geomet- [EC 0 2010, European Committee for Standardization (CEN) 2010a].
ric second-order effect should be considered (Wang 2000). More de- To fulfill the decompression requirement, the tensile stresses of the
tails on slenderness limits for rectangular RC columns can be found in concrete are controlled to having a zero tensile stress at the extreme
Marí and Hellesland (2005). For the bridges piers of the semi-integral fibers of the cross sections. The result of the predesign is 11 pre-
concrete bridges, the slenderness of lcol 5 50 is chosen. The geometric stressing tendons in the side spans (Ap1 5 Ap3 5 247:5 cm2 ) and 13
properties of the piers in relation to the varying pier heights are shown in prestressing tendons at the midspan (Ap2 5 292:5 cm2 ).
Fig. 1. The material properties of the concrete, the reinforcing steel, and
the prestressing steel for the single box girder and the rectangular piers
are listed in Table 1. Limit States and Material Modeling
The check of decompression in the serviceability limit sate is
used for the predesign of the prestressing tendons according to the The SLS and the ULS are considered in the quantification of the
provisions of German national annex of Eurocode 2 [EC 2-2 NA sensitivity indexes to analyze the structural load-bearing behavior
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

2012, German Institute for Standardization (DIN) 2012]. Here, for different loading conditions. The quasi-permanent load combi-
nation is applied according to the SLS and the permanent load
combination is used for the ULS (EC 0 2010, CEN 2010a). The
Table 1. Material Properties of Concrete, Reinforcing Steel, and timeline of the loading and the restraint conditions for the semi-
Prestressing Steel for Superstructure and Piers integral bridges are shown in Fig. 2, and the characteristic loads are
Superstructure listed in Table 2. The compressive concrete stresses in SLS are
Material property (units in MPa) Piers (units in MPa) controlled to be less than sc # 0:40 × fcm . A linear-elastic material
behavior of the concrete in this compressive stress ratio can be
Concrete C 40/50 C 30/37
generally assumed. Therefore, the concrete is modeled as a linear-
CEM CEM II 52.5 N CEM II 42.5
elastic material with the secant modulus of elasticity Ecm (Model
Ecm 35,000 33,000
Code 2010, International Federation for Structural Concrete 2012).
Ec0 36,750 34,650
The cracking of the concrete on the tension side after reaching the
fcm 48.0 38.0
average concrete tensile strength fctm is modeled using a smeared
fcR 28.9 21.7
crack model. The smeared crack model is suitable for applications,
fctm 3.5 2.9
where many cracks occur in the entire structure (Balázs et al. 2013;
Reinforcing steel B 500 B
Gribniak et al. 2012). In the area effected by the reinforcement hct,eff ,
Es 200,000
the tensile stresses decrease to bt × fctm , allowing for the tension-
fy 500
stiffening effect. The depth of the area is defined as 2.5 times the
fyR 550
distance from the reinforcing steel centroid to the surface of the cross
ftR 594
section (EC 2-2 NA 2012, DIN 2012; Balázs et al. 2013). This model
Prestressing steel Y 1170 S7-16.0A
(Model Code 2010, International Federation for Structural Concrete
Ep 195,000
2012) assumes a constant contribution of stress from the concrete
fp0,1k 1,520
after cracking, expressed by the completeness factor bt .
fp0,1R 1,672
The coefficient bt defines the completeness of the concrete stress
fpR 1,947
distribution over the crack spacing sr,max . The simulation of the load-

Fig. 2. Timeline of inner and outer loading conditions for the limit states

© ASCE 04014014-5 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 Table 2. Loading Conditions for the completeness factor bt . The ULS is a short-term loading
Type of load Symbol Characteristic loads state. Hence, the completeness factor is time independent and has
a constant value of bt 5 0:4 (Model Code 2010, International
Dead load Gk Superstructure gk1 5 162:38 kN=m Federation for Structural Concrete 2012). The reinforcing and
Dead load Gk Pavement gk2 5 45:60 kN=m prestressing steel in the ULS are modeled as bilinear materials with
Prestressing Pk Superstructure sPM0 5 1,292 MPa the calculation values according to EC 2-1 NA 2011 (DIN 2011)
Traffic load Qk Uniformly distributed load qk1 5 57:00 kN=m (Table 1).
Traffic load Qk Tandem load qk2 5 1,200 kN
Temperature Tk Superstructure DTN,con 5 226 K
Temperature Tk DTM,cool 5 25 K Model Classes and Structural Response Values
Temperature Tk Piers DTM 5 15 K The model classes cracking of the concrete caused by tension in the
superstructure and in the piers, the creep and the shrinkage of the
bearing behavior of structures is generally focused on the average concrete, the geometrically nonlinear kinematics, the thermal action,
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

structural behavior (Zilch and Zehetmaier 2010). Instead of the and the foundation stiffness are taken into account for the simulation
maximum crack spacing, the average crack spacing srm  2=3 × sr,max of the semi-integral concrete bridges. The model classes cracking
is used for the description of the concrete strain between the cracks. tension superstructure (X1 ) and cracking tension piers (X2 ) describe
Therefore, the average smeared concrete strain between the cracks the importance of the consideration of the concrete cracking caused
is ɛ c  2=3 × ɛ ct , and in response, bt 5 2=3 × 0:6 5 0:4. The long-term by tension. If these model classes are activated, the cracking is
loading condition in the SLS leads to a time dependency of the considered according to the comments of the previous section. The
completeness factor because of the creep of the bond. Therefore, bt deactivation of both model classes describes a purely linear-elastic
for ribbed reinforcement bars reduces over time from bt 5 0:40ðt 5 0Þ stress-strain relationship of the concrete even in tension. The com-
to bt 5 0:25ðt → ‘Þ (Model Code 2010, International Federation for pressive relationship is not influenced by these model classes.
Structural Concrete 2012). In the other unreinforced concrete layers of The model classes creep (X3 ) and shrinkage (X4 ) of the concrete
the cross section, the tensile stress decreases immediately to zero for are computed using the model published in the Model Code 90-99
strains greater than ɛ c 5 fctm =Ecm . (Comité Euro-International du Beton 1999). This creep model
In the SLS, the reinforcing bars and the prestressing tendons are follows the product approach with a hyperbolic time function. If the
modeled as linear-elastic materials. The ULS defines significantly model class is activated, then the creep strains ɛ c,cr ðtÞ are computed
higher loading levels, and therefore, the response of the structure is according to Model Code 90-99 (Comité Euro-International du
influenced by the nonlinear behavior of the material. In the ULS, the Beton 1999). In contrast, if the creep phenomenon is deactivated,
assumption of the physical linearity of the material models is not ap- the creep strains are ɛ c,cr ðtÞ 5 0. The shrinkage phenomenon in the
propriate. A nonlinear simulation of the entire structure requires a safety sensitivity analysis is considered equivalent to the creep model class.
level, which is comparable with the conventional cross section design Either the shrinkage strains ɛ c,sh ðtÞ are calculated according to
methods using the linear-elastic determination of the section forces and Model Code 1990-99 (Comité Euro-International du Beton 1999)
the subsequent nonlinear design of critical cross sections (Cervenka or ɛ c,sh ðtÞ 5 0 in case of the deactivation of the shrinkage model
2013). The main limit state equation for the nonlinear simulation of class. This shrinkage model explicitly decomposes the shrinkage
structures is defined by the nonlinear safety concept of EC 2-1 NA 2011 process into autogenous shrinkage and drying shrinkage under
(DIN 2011), with the permanent load combination (g R 5 1:3) consideration of the corresponding time-dependent functions.
The nonlinear kinematic relation between the nodal displace-
P ments and the strains in the finite-element analysis is considered in
R
Ed # 0R $ g R  gG,j  EGk,j þ gP  EPk þ gQ,1 the model class geometric kinematic. In the practical design of en-
gR j$1 gineering structures, the second-order theory is commonly used. The
P  nonlinear kinematics of the beam element (X5 ) is considered in the
 EQk,1 þ gQ,i  c0,i  EQk,i (7) simulation of the bridge load-bearing behavior, which results in an
i$1
increase in complexity and in accuracy of the computational model
in comparison with the simplified second-order theory.
The reduced quantile material properties including the material Thermal actions of the superstructure and the piers are applied
safety factors are used for the design of structures with the linear- according to the specifications of Eurocode 1 (EC 1-1-5 2010, CEN
elastic computation of the section forces and the subsequent de- 2010b) and are considered in the model class thermal action. The
sign of the critical cross sections. In contrast, the calculation values thermal loading condition considered for the superstructure is the
(expressed by index R in Table 1) are defined for the material contraction state with the constant part DTN,con 5 226 K and the linear
properties in the case of the nonlinear safety concept. The failure of temperature cooling gradient of the magnitude DTM,cool 5 25 K. The
the system is controlled by the material limit strains. More comments piers of the concrete bridge are subjected to a linear temperature
regarding the nonlinear global safety format of Eurocode 2 with gradient of DTM 5 15 K. In the case of the deactivated model class
various comparative case studies can be found in Allaix et al. (2013), thermal action (X6 ), the thermal strains are defined to ɛ c,t ðtÞ 5 0.
Cervenka (2013), and Schlune et al. (2012). The stiffness prediction of the pile foundation is computed by
The model selection of adequate stress-strain relations for the different pile group models for the separate loading conditions. In
material description is fundamentally important for the nonlinear case of the vertical pile group stiffness, the model from Randolph
simulations. The behavior of the concrete under compression at and Wroth (1979) is applied. Therein, the interaction in the pile
ULS is modeled using the nonlinear stress-strain relationship for the group is dependent on the loaded pile. All other displacements in the
structural load-bearing analysis according to EC 2-1 NA 2011 (DIN pile group are calculated dependent on the loaded single pile ap-
2011) and Model Code 2010 (International Federation for Structural plying the superposition principle. For the calculation of the dis-
Concrete 2012). placement caused by the lateral loading of the pile group, the model
The material behavior of the concrete cracking under tension in according to Randolph (1981) is used. Both models consider the
the ULS is modeled similarly to the serviceability limit state, except assumption of an elastic soil continuum. Taking into account the

© ASCE 04014014-6 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 multiplier coefficients of Poulos (1971), the lateral and the rotational girder, the pier height H 1 5 5 m, and the serviceability limit state.
spring stiffness for the pile group are computed at the pile top. The The horizontal axis expresses the position in the longitudinal di-
horizontal, vertical, and rotational stiffnesses are implemented in rection of the superstructure. The vertical axis in the graph shows the
the finite-element model of the bridge by support springs. If the total effects sensitivity index SM,s c2
Ti,Xi , which takes all possible
model class foundation stiffness (X7 ) is deactivated, the support interactions between the partial models into account. The strong
springs are substituted into fixed support conditions at the pier base dependency of the partial model’s sensitivity on the position in the
(u 5 w 5 w 5 0). structure is clearly evident. In general, a significant influence of the
The sensitivity analysis for the model classes considered is partial models shrinkage and creep of the concrete are visible in the
strongly dependent on the structural response values. The impor- ranges of the side spans and the midspan. These internal restraint
tance of the model classes will change with respect to the various effects caused by the statical indeterminacy result in a high influence
response output values, which are typically the horizontal trans- of both partial models on the structural response prediction. At the
lations, the vertical displacements, the concrete stresses, the rein- monolithic connection between the superstructure and the piers, the
forcing steel stresses, and the prestressing steel stresses. For the thermal action and the foundation stiffness have a recognizable
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

assessment of the load-bearing behavior of semi-integral concrete sensitivity.

bridges, the sensitivities of the model classes are therefore quantified To assess the overall quantitative importance in relation to the
using the proposed integrative sensitivity analysis. The displace- entire load-bearing behavior of the structure, the integrative sensi-
ments u and w are evaluated at each node of the finite-element model. tivity analysis computes the influence of the partial models by
The concrete stresses sc1 and sc2 are the stresses at the top and a weighted numerical integration of the total effects sensitivity in-
bottom level of the cross sections of the girder and the piers, re- dexes. Therefore, the local response significance factor S Y has been
spectively. The position of the reinforcement stresses ss1 and ss2 in proposed in this paper. The distribution of S Y for the concrete stress
the cross sections are similarly defined to the top (index 1) and at the bottom level of the box girder S sc2 , for the horizontal
bottom (index 2) concrete stresses. The prestressing steel stresses translations S u , and for the vertical displacements S w is shown in
sp are evaluated at each element of the superstructure. The results Fig. 4 for the SLS and the shorter piers. Because of the fact that the
of the integrative sensitivity analysis for the structural assessment superstructure and the piers are designed with different concrete
of the semi-integral bridges in the serviceability limit state and strength classes, the local response significance factors S sc1 , S sc2 are
ultimate limit state are presented in the following section. related separately to the single box girder and the piers. On one hand,
the order of magnitude of the concrete compressive stress sc2 at
each node (or element) can be related to the minimum value of all
Application of Integrative Sensitivity Analysis nodes (or elements) of the entire single box girder. This de-
termination of the local response significance factor can be com-
puted according to Eq. (8). On the other hand, the significance of the
Position Dependency of Sensitivity Indexes local response of the concrete compressive stress can be related to
The integrative sensitivity analysis is applied to the previously a permissible concrete stress [Eq. (9)]. In the case of the service-
mentioned semi-integral concrete bridges. In this study, the focus ability limit state, Eurocode 2 (EC 2-1 NA 2011, DIN 2011) permits
is the importance evaluation of the physical phenomena represented sc,perm 5 0:45 × fck for the quasi-permanent load combination.
by the partial models. As a first result of the analysis, the assessment
of the partial model’s sensitivity is illustrated in Fig. 3 for the
S sc2 ,rel ðnÞ ¼ smin
c2 ðnÞ=sc2,lsuperstructure
min
(8)
concrete stresses at the bottom level sc2 of the prestressed single box

Ti,Xi for the concrete stresses at the bottom level of the box girder sc2 with respect to the position at the
Fig. 3. Total effects sensitivity index SM,s c2

superstructure, the pier height H 1 5 5 m, and the serviceability limit state

© ASCE 04014014-7 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Local response significance factors S Y for the serviceability limit state and the pier height H 1 5 5 m; concrete stress at bottom level: relative
definition, S sc2 ,rel ; permissible definition, S sc2 ,perm ; horizontal displacements, S u,rel ; vertical displacements, S w,rel

S sc2 ,perm ðnÞ ¼ smin For the application of the integrative sensitivity analysis, the
c2 ðnÞ=sc,perm (9)
local response significance factors are computed by the relative
determination (S rel ), and the structural load-bearing behavior is
S u,rel ðnÞ ¼ umax ðnÞ=umax
lstructure (10) expressed by several response quantities. These are the horizontal
translations u, the vertical displacements w, the concrete stresses
S w,rel ðnÞ ¼ wmax ðnÞ=wmax
lstructure (11) sc1,c2 , the reinforcing steel stresses ss1,s2 , and the prestressing steel
stresses sp .
The determination of the local response significance factor for both
proposals (S sc2 ,rel , S sc2 ,perm ) is shown in Fig. 4. The minimum
SLS
concrete stress at the bottom level of the single box girder occurs at
both side spans and has a magnitude of smin c2 5 27:1 MPa. The On the basis of the integrative sensitivity analysis, the results for
permissible stress is sc,perm 5 0:45 × 2 40 5 218:0 MPa. There- the structural load-bearing assessment of the semi-integral concrete
fore, the local response significance factor for all elements of the bridge with the shorter pier height H 1 5 5 m in the SLS are shown in
superstructure is S sc2 ,perm ðnÞ , 0:40. In contrast, the relative deter- Fig. 5. In case of the longer pier height H 2 5 10 m, the quantification
mination scales the minimum stress at each element to the minimum of the phenomena’s influence is illustrated in Fig. 6. The creep,
stress of the entire box girder. Therefore, the local response sig- shrinkage, and thermal strains induce horizontal displacements.
nificance factor is S sc2 ,rel ðn : 10 , x½m , 20Þ  1:00 in the left side Therefore, the prediction of the horizontal translations u in the
span between Bridge Axis A and B. In the span between Bridge bridges is mainly sensitive to the phenomena creep and shrinkage.
Axis C and D, the local response significance factor is similar: In addition, a considerable influence of the thermal action is rec-
S sc2 ,rel ðn : 105 , x½m , 115Þ  1:00. ognizable. The concrete shrinkage has an integrative sensitivity of
Furthermore, the distribution of the local response significance 0.74 for the shorter bridge piers and an influence of 0.75 for the
factors for the horizontal [S u,rel ; Eq. (10)] and vertical [S w,rel ; longer bridge piers, respectively. In contrast, the concrete creep has
Eq. (11)] displacements is illustrated in Fig. 4. The maximum a sensitivity of 0.18 for the shorter pier height and 0.16 for the longer
horizontal displacements occur at the transition between the box pier height. The sensitivity of the thermal action is 0.08 for both pier
girder and the embankment, respectively, in the sliding bearings. heights. To predict the horizontal translations in the semi-integral
The horizontal translation in the center of the midspan is almost concrete bridges in SLS, the structural model mainly requires ad-
zero for all model combinations. Therefore, the local response equate shrinkage and creep models.
significance factor for the horizontal translations at the end of the In the entire structural model, a high variation in the vertical dis-
side spans is S u,rel ðn: x 5 0 mÞ 5 S u,rel ðn: x 5 128 mÞ 5 1:00 and in placement w prognosis is caused by considering or neglecting the
the center of the midspan is S u,rel ðn: x 5 64 mÞ 5 0:00. The maxi- pile foundation stiffness. This is mainly because of the fact that the
mum vertical displacement occurs in the center of the midspan deactivation (disregard) of the pile foundation stiffness is considered
because of the pile foundation settlement, the creep of the concrete, by fixed support conditions at both pier bases (u 5 w 5 w 5 0), and
and the shrinkage of the concrete. The vertical displacement at the the activation (consideration) is considered by support springs
bridge abutment axes are assumed to be rigid. Thus, the local re- (u  0, w  0, w  0). Hence, the high variation in the prediction of
sponse significance factor for the vertical displacements in the range the vertical displacement is forced by the foundation, which leads to
of the mid span is S w,rel ðn: 50 , x½m , 78  1:00 and in the a corresponding high sensitivity. Furthermore, the maximum ver-
abutment axes is S w,rel ðn: x 5 0 mÞ 5 S w,rel ðn: x 5 128 mÞ 5 0:00. tical displacements occur at the bridge piers, the midspan, and at the

© ASCE 04014014-8 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

M,Y
Fig. 5. Integrative total effects sensitivity index STi,Xi for various response quantities, the serviceability limit state, and the pier height H 1 5 5 m

M,Y
Fig. 6. Integrative total effects sensitivity index STi,Xi for various response quantities, the serviceability limit state, and the pier height H 2 5 10 m

range in the side span close to the bridge piers. This is considered in determination of the soil material properties at the construction
the response significance factor (Fig. 4). Hence, the foundation side are fundamentally important for predicting the entire vertical
stiffness has a high sensitivity at these positions where the maximum displacements.
vertical displacements occur. Consequently, the vertical displace- The overall load-bearing behavior for both semi-integral con-
ments prognoses are mainly influenced by the foundation stiffness. crete bridges in the serviceability limit state (Figs. 5 and 6) is not
The pile foundation stiffness has an integrative sensitivity of 0.75 for significantly influenced by the concrete tensile cracking in the
the shorter pier height. In general, the foundation stiffness deter- superstructure and piers. The concrete tensile cracking in the su-
mines the structural behavior more decisively for shorter piers in perstructure and in the piers has a very small overall influence of
comparison with longer piers. In this respect, the integrative sen- ,0:03 because of the SLS for all response quantities. Conse-
sitivity analysis computes a similar interaction between the pile quently, the concrete tensile cracking for both structural compo-
foundation stiffness and the pier height. The integrative sensitivity nents can either be neglected or considered with simplified models
is reduced to the magnitude 0.57 for the longer bridge piers. In in the SLS. For other structures with different prestressing layout,
addition to the foundation stiffness, creep, and shrinkage of concrete prestressing forces, cross-sectional stiffness, or girder span-pier
have a relevant importance for the vertical displacement predic- height ratio, the concrete tensile cracking can be more sensitive
tions. Finally, the selection of the pile foundation model and the even in the SLS.

© ASCE 04014014-9 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 ULS shrinkage has an overall integrative sensitivity of 0.56 and a concrete
tensile cracking of 0.20. The concrete shrinkage influence is reduced
In addition to the SLS, the load-bearing behavior of the semi-integral
to 0.47 for the longer bridge piers. Furthermore, the concrete tensile
concrete bridges in the ULS is similarly assessed based on the pro-
cracking’s influence is similarly reduced to 0.10 for the longer piers.
posed integrative sensitivity analyses for various response quan-
An opposite effect in the stress prediction is visible in the sensitivity
tities. For the shorter pier height, Fig. 7 illustrates the importance of the thermal action. The thermal action has an integrative sensi-
of various phenomena. Moreover, the results for the longer piers are tivity of 0.16 for the shorter piers and a sensitivity of 0.36 for the
illustrated in Fig. 8. The concrete tensile cracking in the box girder longer piers. Because of the higher horizontal and rotational rigidity
and bridge piers has no importance for all structural response values of the shorter piers, the restraint loading conditions cause much
in the SLS. In contrast, in the ULS, the concrete tensile cracking in higher section forces in comparison with the longer piers. Therefore,
the entire structure has a large influence on the overall structural the restraint section forces increase in the range of the midspan and
load-bearing behavior. the connection positions between the superstructure and the piers
The prediction of the concrete stress at the upper cross-sectional for the shorter pier height. These indirect loading conditions cause
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

level sc1 is mainly influenced by concrete shrinkage and tensile more material nonlinear responses in the structure. Hence, the im-
cracking in the superstructure. In the case of the shorter piers, the portance of the concrete cracking caused by tension is higher for

M,Y
Fig. 7. Integrative total effects sensitivity index STi,Xi for various response quantities, the ultimate limit state, and the pier height H 1 5 5 m

M,Y
Fig. 8. Integrative total effects sensitivity index STi,Xi for various response quantities, the ultimate limit state, and the pier height H 2 5 10 m

© ASCE 04014014-10 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 shorter piers compared with longer piers. These results represent model with high quality for predicting the same response quantity at
a clear interaction between the magnitude of restraint effects and the the midspan. The structural engineer needs to have a global model
material nonlinear responses in the structure. The less cross-sectional with an acceptable overall prediction quality. However, what are the
stiffness degradation for the longer piers and the consequently partial model’s influences according to the entire structure? What
greater remaining restraint effects result in a higher sensitivity of the is the global structural prediction quality for the entire structure? These
thermal-induced strains in the ULS. questions cannot be answered by the local position sensitivity as-
The thermal contraction loading state considered in this analysis sessment. Consequently, the integrative sensitivity analysis is pro-
has a much higher influence on the concrete stress at the bottom posed in this paper.
cross-sectional level sc2 . As a result, the integrative sensitivity of the The enhancement to the integrative sensitivity analysis enables
thermal action is increased in comparison with the prediction at an overall assessment of the entire structural load-bearing behavior.
the upper cross-sectional level. The thermal-induced strains have Hence, the global model quality for the same partial model combi-
a sensitivity of 0.34 for shorter piers and 0.53 for longer piers. This nation based on the proposed method is not dependent on the local
again shows the interaction between the restraint thermal effects and position, because all positions are taken into account by the local
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

the concrete cracking on the prediction of the concrete stresses in the response significance factors. Finally, for each partial model, an
entire structure. The concrete tensile cracking in the superstructure, overall sensitivity index and, subsequently for each partial model
shrinkage, and creep additionally influence the prediction of the combination, an entire structural prediction quality can be assessed.
concrete stresses at the bottom layer of the cross sections. Hence, for each partial model combination (here only one scenario
The thermal-induced strains are not completely degraded from is presented), a representative structural prediction quality can be
the concrete tensile cracking, even in the ULS. Therefore, it is evaluated (Table 3; MQsGM,scenario
c2
5 0:81). A quantitative comparison
necessary to use material nonlinear models to analyze the effect of between several global structural models is then clearly feasible on the
the thermal action on the overall load-bearing behavior in an accu- entire structural level. The integrative sensitivity analysis assists the
rate way. The linear-elastic material models are not capable of an- structural engineer in the decision-making process in various project
alyzing the restraint effects in integral bridges and should not be design phases. The engineer can then choose a structural model with
used for the structural design. an adequate prediction quality for the entire structure to obtain more
reliable simulation results and a more safe design.
Global Model Quality Evaluation
Design Recommendations
To illustrate the significance of the proposed integrative sensitivity
analysis, the global model quality assessment in the SLS for the The results of the integrative sensitivity analysis in the SLS and ULS
shorter piers is shown in Table 3. The concrete stress at the bottom present a clear quantification of the phenomena’s importance on the-
level of the cross sections sc2 is the example chosen as the response overall structural load-bearing behavior. The decomposition of integral
value. The creep, shrinkage, thermal action, and foundation stiffness bridges into superstructure and substructure is only valid on a geo-
influences this concrete stress prediction (Figs. 3 and 5). The partial metrical level. In the numerical simulation of semi-integral bridges, the
model qualities are assumed by the following scenario: MQPM,X3 high interaction between the structural parts, such as girder, piers, and
5 0:70, MQPM,X4 5 0:90, MQPM,X6 5 0:50, and MQPM,X7 5 0:60. foundations (soil), has to be considered. Therefore, the direct coupling
The assessment at local positions shows the significant position of all these components in the model is more adequately compared to
influence on the sensitivity indexes (Table 3). Therefore, the global a decoupled simulation. A decoupled design of the integral bridge
model quality for the same partial model combination is different at structure in a separate way for the superstructure and substructure is not
each position because of the changed sensitivity values. Moreover, it appropriate and cannot provide reliable predictions.
is also obvious that some phenomena have no influence at a certain The shrinkage and the creep of the concrete have a significant
position in the structure on the response quantity. In contrast, they influence on the structural load-bearing behavior for the SLS and
can affect the structural load-bearing behavior at other positions. ULS. These phenomena are mainly influenced by the modulus of
Consequently, for three positions in the structure listed in Table 3, elasticity and the concrete compressive strength, which both gen-
the quality assessment quantifies three different global model quali- erally have a high variance. The parameters defined in design codes
ties for the same partial model combination. It is believed that in and guidelines are only imprecise estimates. Therefore, it can be
practical engineering problems, there is no requirement to establish necessary and very useful to perform short-term (modulus of elas-
various global models with the corresponding best prediction quality ticity and compressive strength) and long-term (creep and shrink-
at a local position in the structure for design purposes. For example, it age) experiments to measure more accurate data for the particular
would not be feasible for practical engineering projects to choose concrete mixture. This parameter estimation can significantly reduce
a certain global model with high quality for predicting a response the uncertainty of the model prediction. Finally, the design of the
quantity at the side span and subsequently select a different global entire structure becomes more reliable and safe.

Table 3. Global Model Quality Evaluation Based on Partial Model Quality Scenario in the Serviceability Limit State
X3 Creep X4 Shrinkage X6 Thermal action X7 Foundation stiffness
Position MQPM,X3 5 0:70 MQPM,X4 5 0:90 MQPM,X6 5 0:50 MQPM,X7 5 0:60 MQsGM
c2

Local position sensitivity SM,s c2

Ti,Xi
x 5 14:75 m 0.34 0.64 0.03 0.01 0.82
x 5 39:68 m 0.23 0.18 0.62 0.20 0.61
x 5 64:00 m 0.22 0.77 0.02 0.00 0.85
Integrative sensitivity analysis M,s
STi,Xic2
Entire structure 0.27 0.65 0.08 0.03 0.81

© ASCE 04014014-11 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 In addition, the foundation and soil stiffness is quantified to be illustrate where complex or simplified models can be used for
very sensitive according to the vertical bridge displacements. representing the various physical phenomena in the structural model.
Therefore, the foundation stiffness should also be carefully checked This quantitative model selection assists the structural engineer in
on the construction site. Material properties for the surrounding soil obtaining more reliable numerical simulation results and ensuring
and the piles should not be exclusively related to experience and a safer design. Design recommendations can then be made in a
empirical values. Soil tests or even pile foundation tests on site quantitative manner by applying the proposed integrative sensitivity
should be performed, because such experiments already include the analysis for various structures and conditions.
interaction between the soil and the pile and can additionally reduce
the model prediction uncertainty.
For integral bridge structures with high rigidity of the structural Acknowledgments
components (e.g., shorter bridge piers), the restraint effects caused
by temperature, shrinkage, and creep could lead to a material non- This research is supported by the German Research Foundation
linear response in the structure. The higher rigidity of the shorter (DFG) via research training group “Assessment of Coupled Exper-
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

piers results in larger section forces because of the restraint effects, imental and Numerical Partial Models in Structural Engineering
which result in the concrete cracking. In the application example, (GRK 1462),” which is gratefully acknowledged by B. Jung and
concrete cracking under tension is not sensitive to the structural G. Morgenthal. The close collaboration between Bauhaus-Uni-
behavior for the SLS, but a considerable sensitivity occurs for the versität Weimar and Tongji University is also acknowledged. In
ULS. This relation cannot be generalized, because a lot of factors addition, B. Jung thanks Mr. H. Stutz for the contribution in the pile
such as loading conditions, cross-sectional stiffness/shape, material foundation models.
properties, prestressing forces, and the span-height ratio affect the
magnitude of restraint effects. In general, the structural engineer
should care about the material nonlinear behavior and cannot as- References
sume a simplified linear-elastic material description for integral
bridges. Otherwise, the section forces caused by restraint effects be- Allaix, D. L., Carbone, V. I., and Mancini, G. (2013). “Global safety format
come inaccurately high based on the linear-elastic computation, and for non-linear analysis of reinforced concrete structures.” Struct. Concr.,
thus the amount of reinforcement increases significantly. In contrast, 14(1), 29–42.
concrete cracking and stiffness degradation reduce the section forces American Concrete Institute (ACI). (1992). “Prediction of creep, shrinkage,
because of these indirect loading conditions. Therefore, a nonlinear and temperature effects in concrete structures.” Rep. No. 209, Detroit.
Balázs, G. L., et al. (2013). “Design for SLS according to fib Model Code
simulation leads to a more suitable design approach.
2010.” Struct. Concr., 14(2), 99–123.
Because of the fact that integral bridges are very sensitive to the Bazant, Z. P., and Bajewa, S. (1995). “Creep and shrinkage prediction mod-
material’s and structural component’s stiffness, the structural en- el for analysis and design of concrete structures-Model B3.” Mater. Struct.,
gineer must be careful in model selection and should assess the main 28(6), 357–365.
influencing phenomena. This analysis will illustrate where the global Bazant, Z. P., and Li, G.-H. (2008). “Unbiased statistical comparison of
structural model has to use more accurate partial models and where creep and shrinkage prediction models.” ACI Mater. J., 105(6), 610–621.
even simplified partial models can be used. Therefore, the integrative Bloodworth, A. G., Xu, M., Banks, J. R., and Clayton, C. R. I. (2012).
sensitivity analysis is a powerful tool that can significantly reduce “Predicting the earth pressure on integral bridge abutments.” J. Bridge
the uncertainty in model predictions. The design of engineering Eng., 10.1061/(ASCE)BE.1943-5592.0000263, 371–381.
structures in accordance with the recommendations of codes and Cervenka, V. (2013). “Reliability-based non-linear analysis according to
fib Model Code 2010.” Struct. Concr., 14(1), 19–28.
guidelines should be performed based on the results of a global
Chacòn, R., Mirambell, E., and Real, E. (2013). “Strength and ductility
structural model with a high prediction quality to obtain a more of concrete-filled tabular piers of integral bridges.” Eng. Struct., 46,
reliable and safe design. 234–246.
Comité Euro-International du Beton. (1999). Structural concrete: Textbook
on behavior, design and performance—Updated knowledge of the CEB-
Conclusion FIP Model Code 90, Vol. 1–3, Comité Euro-International du Beton,
Paris.
The integrative sensitivity analysis for the overall estimation of the Dicleli, M., and Erhan, S. (2010). “Effect of soil-bridge interaction on the
partial model’s sensitivity in the entire structure is proposed in this magnitude of internal forces in integral abutment bridge components
paper. The method considers the local response significance and the due to live load effects.” Eng. Struct., 32(1), 129–145.
European Committee for Standardization (CEN). (2010a). “Eurocode: Basis
sensitivity information. If the local response significance is neglec-
of structural design.” EN 1990:2010-12, Brussels, Belgium.
ted, then the sensitivity analysis would assume an equal importance European Committee for Standardization (CEN). (2010b). “Eurocode 1:
at each position in the structure for the overall structural load-bearing Actions on structures-Part 1–5: General actions-Thermal actions.” EN
behavior. However, this is a very simplified assumption for engi- 1991-1-5:2010-12, Brussels, Belgium.
neering structures. Therefore, the magnitude of a desired response Faraji, S., Ting, J. M., Crovo, D. S., and Ernst, H. (2001). “Nonlinear
quantity at each position is taken into account in the computation of analysis of integral bridges: Finite-element model.” J. Geotech. Geo-
the local response significance factors. Permissible material stresses environ. Eng., 10.1061/(ASCE)1090-0241(2001)127:5(454), 454–461.
defined in design codes can be applied as reference values, rendering Gardner, N. J., and Lockman, M. J. (2001). “Design provisions for drying
the local response significance similar to a utilization ratio. An shrinkage and creep of normal-strength concrete.” ACI Mater. J., 98(2),
alternative for the reference value is to use the absolute response 159–167.
German Institute for Standardization (DIN). (2011). “National annex-
maximum at all positions in the numerical structural model.
Eurocode 2: Design of concrete structures-Part 1-1: General rules and
The application of the integrative sensitivity analysis to the nu- rules for buildings.” DIN EN 1992-1-1/NA:2011-01, Beuth, Berlin.
merical simulation of semi-integral concrete bridges clarifies its German Institute for Standardization (DIN). (2012). “National annex-
general applicability for complex engineering structures. Therefore, Eurocode 2: Design of concrete structures-Part 2: Concrete bridges-
the categorization of the phenomena’s importance in the numerical Design and detailing rules.” DIN EN 1992-2/NA:2012-04, Beuth,
simulation of entire structural models can be assessed. The results Berlin.

© ASCE 04014014-12 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014

 Gribniak, V., Kaklauskas, G., Hung Kwan, A. K., Bacinskas, D., and Marí, A. R., and Hellesland, J. (2005). “Lower slenderness limits for
Ulbinas, D. (2012). “Deriving stress-strain relationships for steel fiber rectangular reinforced concrete columns.” J. Struct. Eng., 10.1061/
concrete in tension from tests of beams with ordinary reinforcement.” (ASCE)0733-9445(2005)131:1(85), 85–95.
Eng. Struct., 42, 387–395. Marx, S., and Seidl, G. (2011). “Integral railway bridges in Germany.”
Homma, T., and Saltelli, A. (1996). “Importance measures in global sen- Struct. Eng. Int., 21(3), 332–340.
sitivity analysis of nonlinear models.” Reliab. Eng. Syst. Saf., 52(1), Most, T. (2011). “Assessment of structural simulation models by estimating
1–17. uncertainties due to model selection and model simplification.” Comput.
Huang, J., Shield, C. K., and French, C. E. W. (2008). “Parametric study of Struct., 89(17–18), 1664–1672.
concrete integral abutment bridges.” J. Bridge Eng., 13(5), 511–526. Ooi, P. S. K., Lin, X., and Hamada, H. S. (2010). “Numerical study of an
International Federation for Structural Concrete. (2012). “Model Code 2010: integral abutment bridge supported on drilled shafts.” J. Bridge Eng.,
Final draft.” Bulletin 65 Vol. 1 and Bulletin 66 Vol. 2, Lausanne, 10.1061/(ASCE)BE.1943-5592.0000037, 19–31.
Switzerland. Poulos, H. G. (1971). “Behavior of laterally loaded piles: II. Pile groups.”
Jung, B., Morgenthal, G., and Xu, D. (2013). “Integral bridges: Sensitivity J. Soil Mech. and Found. Div., 97(5), 733–751.
of limit state modelling.” Bautechnik-Special Print Modellqualiäten, Randolph, M. F. (1981). “The response of flexible piles to lateral loading.”
Downloaded from ascelibrary.org by Universidad De Guadalajara on 12/15/23. Copyright ASCE. For personal use only; all rights reserved.

90(4), 32–40. Geotechnique, 31(2), 247–259.

Kaveh, A. (2004). Structural mechanics: Graph and matrix methods, 3rd Randolph, M. F., and Wroth, C. P. (1979). “An analysis of vertical de-
Ed., Research Studies Press, Wiley, Exeter, U.K. formation of pile groups.” Geotechnique, 29(4), 423–439.
Keitel, H. (2013). “Quantifying sources of uncertainty for creep models Schlune, H., Plos, M., and Gylltoft, K. (2012). “Safety formats for non-linear
under varying stresses.” J. Struct. Eng., 10.1061/(ASCE)ST.1943- analysis of concrete structures.” Mag. Concr. Res., 64(7), 563–574.
541X.0000716, 949–956. Sobol, I. (1993). “Sensitivity estimates for nonlinear mathematical models.”
Keitel, H., Karaki, G., Lahmer, T., Nikulla, S., and Zabel, Z. (2011). Math. Model. Comput. Exp., 1, 407–414.
“Evaluation of coupled partial models in structural engineering using Wang, J. (2000). “Piers and columns.” Bridge engineering—Handbook,
graph theory and sensitivity analysis.” Eng. Struct., 33(12), 3726–3736. W.-F. Chen and L. Duan, eds., CRC Press, Boca Raton, FL.
Kim, W. S., and Laman, J. A. (2010). “Numerical analysis method for Zilch, K., and Zehetmaier, G. (2010). “Bemessung im konstruktiven
long-term behavior of integral abutment bridges.” Eng. Struct., 32(8), Betonbau: Nach.” DIN 1045-1 (2008) und EN 1992-1-1 (Eurocode 2),
2247–2257. 2nd Ed., Springer, Berlin.
Krizek, J. (2011). “Soil-structure interaction of integral bridges.” Struct. Zordan, T., Briseghella, B., and Lan, C. (2011). “Parametric and pushover
Eng. Int., 21(2), 169–174. analyses on integral abutment bridge.” Eng. Struct., 33(2), 502–515.

© ASCE 04014014-13 J. Bridge Eng.

J. Bridge Eng., 2014, 19(6): 04014014