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

net/publication/245304745

Modeling for Postyield Buckling of Reinforcement

Article  in  Journal of Structural Engineering · September 2002

DOI: 10.1061/(ASCE)0733-9445(2002)128:9(1139) · Source: OAI

CITATIONS READS
177 1,070

2 authors:

Rajesh P Dhakal Koichi Maekawa

University of Canterbury Yokohama National University
339 PUBLICATIONS   2,876 CITATIONS    473 PUBLICATIONS   5,906 CITATIONS   

SEE PROFILE SEE PROFILE

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

Seismic Collapse Probability Prediction of RC Frame Buildings View project

long term performance of prestressed concrete structures subjected to aggressive environment and repeated loading View project

All content following this page was uploaded by Rajesh P Dhakal on 28 May 2014.

The user has requested enhancement of the downloaded file.

 MODELING FOR POST-YIELD BUCKLING OF REINFORCEMENT

By Rajesh Prasad Dhakal1 and Koichi Maekawa2

ABSTRACT: Finite element analysis using fiber technique was carried out to study the post-yield buckling
mechanism of reinforcing bars. It was found that the softening of compressive stress takes place due to the
geometrical nonlinearity associated with the lateral deformation of the compressed bars, especially after the absolute
strain exceeds the yielding strain. It was clarified that the post-buckling average stress-strain relationship over the
analysis domain can be specified in terms of the product of square root of yield strength and slenderness ratio of the
reinforcing bar. Moreover, a unique relationship between the average stress and average strain of reinforcing bars
including the effect of buckling is established through various parametric analyses. The comparison of analytical
results and proposed model with experimental data showed good agreement.

Key words: buckling, lateral deformation, reinforcing bar, average compressive response, point wise stress-strain
relationship, geometrical nonlinearity

INTRODUCTION

In reinforced concrete members, the reinforcing bars might undergo high compressive strain, which induces
large lateral deformation of reinforcing bars, hereafter referred to as buckling. Because of geometrical nonlinearity,
average compressive stress carried by reinforcing bars decreases in the post-buckling range. However in tension,
geometrical nonlinearity does not prevail as no lateral deformation is induced even after yielding. Consequently, the
average tensile stress-strain relationship over the specified control volume is exactly same as the point wise tensile
stress-strain relationship. Therefore, using similar average stress-strain behavior for reinforcing bars in tension and
compression seems unconvincing.
Of course, the point wise stress-strain behavior of reinforcing bars in compression is also the same as that in
tension because the point wise relationships are not influenced by the change in geometry (Dodd and Restrepo-
Posada 1995). Hence, using the point wise stress-strain relationship can account for the lateral deformation of
compressed longitudinal reinforcing bars provided that the control volume, over which the average stress-strain
relationship is computed, is very small. However, in the finite element structural analysis of larger scale, such small
element size is not feasible because of the long computation time and large memory required. For finite element
analysis of reinforced concrete structures with substantial element size, an average stress-strain relationship of
reinforcing bars that takes into account the effect of buckling is needed. Hence, this research aims to formulate a
versatile average stress-strain relationship that can be applied to reinforcing bars with any geometrical and
mechanical properties.
To generate the data for model formulation, a parametric study based either on experiment or analysis is
necessary. There are some experimental studies performed in the past to clarify the buckling mechanism of bare bar
(Monti and Nuti 1992, Rodriguez et al. 1999). Tests are usually conducted within some fixed ranges of material
properties, and in some cases we have to consider the boundary conditions, which is hardly reproduced in the tests.
For obtaining widely applicable constitutive models, the experiments should consist of test specimens that
systematically cover wide range of geometrical as well as mechanical properties. Hence, analytical parametric study
is preferred ahead of extensive experimental study, and some available experimental results are used to justify the
analytical method and also to verify the proposed constitutive equations.

ANALYTICAL SIMULATION OF BARE BAR BUCKLING

Fiber Technique for FEM Microanalysis

A three-dimensional nonlinear finite element analysis program called COM3 [Concrete Model in 3D] (Hauke
and Maekawa 1999) is used for the analytical parametric study. Nonlinear space frame elements analyzed by fiber
technique (Menegotto and Pinto 1973) are used to model reinforcing bars. In fiber technique, each element is
represented using a single line coinciding with the centerline of the member. The member cross section is divided
into many cells or sub-elements. The strain of each cell is calculated based on the Euler-Kirchoff’s hypothesis, i.e.
plane section remains plane after bending. For each fiber strain along the axis of finite element, response is
calculated using the material constitutive models representing the local behavior. As is well known, the overall
1
Research Fellow, School of Civil and Structural Engineering, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798,
E-mail: [email protected] (Formerly: Graduate Student, The University of Tokyo)
2
Professor, School of Civil Engineering, The University of Tokyo, Hongo 7-3-1, Bunkyo-Ku, Tokyo 113, Japan, E-mail:
[email protected] 1
response of each element is the integrated response of these fibers and the overall response of the member comprises
of all the element responses. The implementation of fiber technique is briefly illustrated in Fig. 1.
In this study, isoparametric elements with three-nodes are used. In each node, six degrees of freedom are
considered, i.e. the displacements (u, v and w) and rotations (θx, θy and θz) in three mutually perpendicular directions
(local axes Xl, Yl and Zl in Fig. 1). Displacement field for an element is calculated from the nodal displacements and
the assumed shape functions. Quadratic shape functions are applied for displacement field. Next, strain-displacement
relationships are used to compute the generalized strains, which include an axial component (ε0), two shear
components (γzy, γzx) and three curvatures (φx, φy and φz). Based on these generalized strains, the average axial strain
of each fiber εf is computed. The strain distribution across the cross section is assumed to be linear, which comes
from the assumption that plane section remains plane even after bending. In case of bending, the average axial strain
of each fiber depends on its distance from the neutral axis. The last two terms (second order derivative terms) in the
computation of average axial fiber strain includes the change in geometry due to large displacement of the
reinforcing bars during loading. In small strain problems, where the geometry of the structure basically remains
unchanged, these higher order terms can be neglected. However, large lateral displacement of reinforcing bars cannot
be captured without giving due consideration to geometrical nonlinearity. This fact is analytically proved afterwards.
Based on this axial strain, fiber axial stress is calculated from the point wise stress-strain relationship. These fiber
stresses across the cross section are integrated to obtain the stress carried by the element.

Effect of Buckling in Average Behavior

Following the aforementioned computation scheme, overall average response would become similar to the local
response if only one element is used to model the bare bar length. To capture the associated geometrical nonlinearity,
large number of finer elements should be used. A schematic diagram explaining the effect of reinforcement buckling
on the average response is depicted in Fig. 2. Here, the length between two clamped boundaries L is defined as
referential length, over which the average stress-strain relationship is to be generated. Before the lateral deformation
is induced, the stress and strain distribution along the length as well as across the cross-section is uniform. Therefore,
overall average behavior is the same as local behavior. However, once the reinforcement deforms laterally under
high compression, the stress and strain distributions along the length and across the bar section change significantly.
As shown in Fig. 2, laterally deformed part of the referential length experiences compression hardening, and the rest
is subjected to unloading. Moreover, curvature exists due to the lateral deformation, and the strain across the bar
section is not constant but varies linearly. Obviously, the value of curvature depends on the position of the concerned
section along the length. For example, at mid-height where maximum lateral deformation is experienced, the inner
fibers are in compression hardening state whereas the outer fibers are in unloading state. Because the unloading
stiffness is significantly higher compared to the hardening stiffness, a small decrease in fiber strain causes substantial
decrease in the fiber stress compared to the increase in fiber stress due to an equivalent increase in fiber strain.
Consequently, the average stress of the overall referential length becomes smaller than the local response.

Preliminary Microanalysis

As shown in Fig. 3, the reinforcing bar is modeled as a vertical column of length L and diameter D. The total
length L is discretized into several small elements. Monotonic downward displacement ∆ is applied at the top of the
reinforcement. To know the effect of end condition, a pair of analyses is carried out with considering fixed ends and
hinged ends, respectively. An elastic-perfectly plastic stress-strain relationship without hardening is used as the local
material model, and the local yield stress and Young modulus of steel are assumed to be 400 MPa and 200 GPa,
respectively. The average load-displacement relationships computed with these two boundary conditions are
compared in Fig. 3. It is found that allowing rotations at the extreme ends causes severe softening of the average
response. In order to simulate the bare bar compression tests (Mander et al. 1984; Monti and Nuti 1992), the two
ends of the reinforcing bars are represented by fixed nodes, which coincides with the experimental loading condition;
i.e. both displacement and rotation are not allowed to take place at these nodes.
As mentioned earlier, the lateral deformation of reinforcing bars can be predicted only if the geometrical
nonlinearity is given due consideration in the analysis. In order to verify this statement, a pair of analyses with and
without considering geometrical nonlinearity is performed, and the average response are compared in Fig. 4(a). If
geometrical nonlinearity is not considered; i.e. the second order terms are neglected in the computation of fiber
strain, the post-yield buckling phenomenon cannot be predicted, and the average response becomes the same as the
local response. Next, mesh sensitivity analysis is conducted to obtain the influence of mesh discretization in average
response. A reinforcing bar with L = 16 cm, D = 16 mm, fy = 400 MPa is analyzed with varying number of elements.

2
The average load-displacement curves are compared in Fig. 4(b). Using only one element rendered the average bar
response exactly follow the assumed local behavior, and could not predict the softening of the average compression
in spite of considering geometrical nonlinearity. As expected, the average responses obtained using finer elements
exhibit increased softening, and the mesh dependency nearly vanishes for more than 4 elements. Hence, at least four
or more elements are used for further analyses.
Next, the lateral deformation profile and longitudinal as well as transverse distribution of fiber strain and stress
are explained. Three instants denoted by points A, B and C in Fig. 4(b) are selected for comparison. Point A refers to
the instant just before the average stress starts decreasing, and points B and C lie in the softening phase. The lateral
displacement profiles plotted in Fig. 4(c) indicate that the bar deforms laterally only after point A, and this lateral
deformation is maximum at the mid-height. Similarly, the average strain (strain at the center of the cross-section)
profiles plotted in Fig. 4(d) reveal that the bar experiences uniform compression until point A, i.e. the starting point
of buckling. This uniformity is lost afterwards and some portions, especially near the fixed support and around the
mid-height, experience continuous increase in compressive strain while strains in other parts decrease.
In Fig. 5, the fiber strain and stress distributions across the bar section at the mid-height during phases B and C
are plotted. The bar cross-section lies in the center of the plotted rectangular domain in the figure, and the corners
that are extrapolated from the data inside the bar section should be disregarded. Both the strain and stress are
constant throughout the bar section at phase A, and hence are not shown in the figure. However, curvature is induced
afterwards due to the lateral deformation. Consequently, the strain linearly changes across the bar section, and the
stress distribution becomes highly irregular due to different post-yield loading and unloading stiffness. Due to the
elastic-perfectly plastic material model used in the analysis, the compressive stress in the fibers does not exceed yield
stress in spite of large fiber strain. In contrast, the compressive stress in the unloading fibers is less than the yield
stress, and some extreme fibers even experience tensile stress. As claimed earlier, this complex mechanism leads to
the softening of average compressive stress with increase in lateral deformation.

General Discussion on FEM Microanalysis

In the mesh for microanalysis, all nodes of all elements were perfectly aligned along the same vertical line and
no initial imperfection was provided to induce lateral deformation. In spite of this fact, the analytical solution could
capture the buckling mechanism. The two possible modes of deformation of a bar under axial compression are axial
mode representing longitudinal shortening and bending mode representing the usual buckling shape. The axial mode
always leads to local stress-strain behavior whereas the bending mode triggers the softening of average stress
provided that the material nonlinearity and path-dependency are considered in the constitutive model. Both of these
modes satisfy the requirements; i.e. the compatibility, equilibrium and the constitutive equations. However, the
energy associated with the bending mode is smaller than that with axial mode. Hence, if geometrical nonlinearity is
considered and the intermediate nodes are free to deform laterally, the finite element solution selects the bending
mode, as it requires less energy during the nonlinear iteration process.
For further understanding, the analysis of a bar under compression was performed with geometrical nonlinearity
but without any path-dependency; i.e. stress reversal during cyclic loading was not considered. In other words, once
the compressive strain exceeds yielding strain, the stress is kept constant at yield stress regardless of the strain path
(unloading or reloading). As shown in Fig. 6, the average stress-strain curve in this case is found to be the same as
local stress-strain relationship in spite of the presence of large lateral deformation. Although high curvature (strain
gradient) across the section could be observed, the stresses at all fibers were equal because of the path-independent
constitutive model. Consequently, in spite of the geometrical nonlinearity, softening of the average stress could not
be observed. This can also explain why the average response is nearly the same as local response until yielding. For
reinforcing bars with normal slenderness ratio, lateral deformation does not start in the elastic range and the bar
deforms axially rendering the average behavior to be similar to the local behavior. Nevertheless in very slender bars,
lateral deformation might take place before yielding. The loading and unloading stiffness in elastic range are both
equal to the Young’s modulus and a small strain gradient due to the lateral deformation does not cause significant
difference between the local and the average stress. However, when the average strain is close to yield strain, even a
small strain gradient might be influential as some fibers experience hardening and the others undergo elastic
unloading, causing average stress to be slightly less than the local stress.
Euler’s theorem is mostly used to calculate the buckling load of axially compressed longitudinal bars. However,
it is based on the assumption that the bar behaves elastically throughout the loading. Hence for short bars, the
buckling load predicted by Euler’s formula is usually higher than the yield load. But, because of the elasto-plastic
nature of the reinforcing bars, the load cannot linearly increase in the post-yield range, and Euler’s theory is no
longer applicable. In this context, the FEM microanalysis predicts that the buckling starts in the plastic region, where

3
the axial stress is slightly higher than or equal to yield stress but the strain is substantially higher than the yielding
strain. For normal strength steel bars (fy ~ 400 MPa) with slenderness ratio less than or equal to 35 and fixed at both
ends, Euler’s formula predicts the critical buckling load to be more than yielding load. Keeping in mind that the tie
spacing in RC members almost never exceeds 35 times longitudinal bar diameter, it can be said that Euler’s theory
cannot be directly used, and FEM simulation serves as a more reliable method. Needless to mention, for bars with
higher slenderness ratio, buckling starts in elastic range and Euler’s formula gives accurate prediction of buckling
load that is smaller than the yield load.
The softening of average stress in post-buckling range is due mainly to the difference between the stiffness for
post-yield loading and unloading paths. The average response depends on both geometrical nonlinearity and the
path-dependency, and the influence of path-dependency becomes crucial only when the geometrical nonlinearity is
taken into account. On the other hand, the lateral deformation can be simulated only with geometrical nonlinearity
but the softening of the post-buckling average stress cannot be captured without considering the path-dependency in
the nonlinear material model. The average compressive stress-strain relationships predicted analytically with
assumed material model might not be exact but is certainly not far from the correct one. For sure, if unloading and
hardening properties are correctly known in advance, the correct average constitutive model of reinforcing bars with
any type of point wise stress-strain relationship can thus be obtained analytically.

VERIFICATION AND PARAMETRIC STUDY

Verification of Microanalysis

For verification of analytical method, the computed results are compared with experimental ones (Mander et al.
1984). The experiment consists of direct compression tests of five short medium strength (fy = 290 MPa) reinforcing
bars with different slenderness ratios (5.5, 6, 6.5, 10 and 15). The schematic test setup, point wise stress-strain
relationship and comparative normalized average compressive stress-strain curves are shown in Fig. 7. Note that the
hardening material model obtained through sample bar test was used in the verification analysis. The experimental
and analytical curves are compared for only three different slenderness ratios, as the results of the first three cases are
found to be very close to each-other in experiment as well as analysis, and only one representative case (L/D = 6)
among these three was chosen. The results of FEM microanalysis are in good agreement with the experimental
curves giving ample proof of the reliability of the analysis.
Next, the monotonic tests of Monti and Nuti (1992) are adopted for the verification of monotonic behavior. The
experiment consists of series of monotonic tests on steel reinforcing bars with different slenderness ratios (5, 8 and
11) and three different bar diameters of 16 mm, 20 mm and 24 mm were used for each slenderness ratio. Here, the
average of the three responses (with different diameters) for each slenderness ratio is used for comparison with FEM
microanalysis. As the details of the material properties were not mentioned, the point wise stress-strain relationship
was fairly assumed to match the average response of the bar with slenderness ratio 5, because it was reported that the
behavior of this shortest bar almost coincides with the material model. The point wise stress-strain relationship used
and the normalized average stress-strain curves are shown in Fig. 8. It can be observed that the results of FEM
microanalysis are in fair agreement with the experimental results for all three cases of monotonic loading. However,
small difference can be observed in the early hardening range of the bar with slenderness ratio 5. This is due to the
fact that the nonlinear hardening of the bar with slenderness ratio 5 was represented by using an average linear
hardening range in the point wise stress-strain relationship.
These verifications give ample evidence that FEM microanalysis with coupled geometrical and material
nonlinearity and path-dependent material models can reliably predict the buckling mechanism and also the average
stress-strain behavior of axially compressed reinforcing bars. In addition, the accuracy is found to be good enough
for engineering purposes. Hereafter, detailed parametric study is conducted using FEM analysis to obtain unified and
versatile equations that can represent the average compressive stress-strain relationship of reinforcing bars.

Effect of Length and Diameter

It is well known that the average behavior of reinforcing bar is very sensitive to its length and diameter. To
elaborate more in this regard, two sets of parametric analyses are conducted with constant diameter and different
lengths as well as constant length and different diameters. The normalized average responses are compared in Fig.
9(a) and (b). The softening of average stress becomes steeper with the increase in reinforcement length provided that
the bar diameter is constant. Similarly, if the reinforcement length is kept constant, the softening becomes milder
with increase in bar diameter. Both of these give similar qualitative relationship; i.e. the more slender the bar is, more

4
severe is the softening of average stress. In order to obtain quantitative interrelationship, two pairs of cases, so that
the slenderness ratio (length to diameter ratio) is the same but the length and diameter are different, are compared in
Fig. 9(c). It is found that irrespective of the separate values of length and diameter, the normalized average response
is similar for the same slenderness ratio. Finally, the normalized average responses for different slenderness ratios,
assuming elastic-perfectly plastic behavior and fy = 400 MPa, are shown in Fig. 9(d). Agreeing with the conclusions
of previous studies (Dhakal and Maekawa 2000; Mau and El-Mabsout 1989; Monti and Nuti 1992), the results
showed that buckling is delayed and the post-buckling degradation of average response decreases with the decrease
in slenderness ratio of reinforcing bars.

Effect of Yield Strength

The authors believe that the strength of reinforcing bar also influences the average behavior in compression. To
explore the interrelationship, two sets of sensitivity analyses are performed with constant slenderness ratio and
different yield strengths. The comparative normalized average stress-strain curves for slenderness ratios 5 and 10 and
yield strengths ranging from 100 MPa to 1600 MPa are presented in Fig. 10(a) and (b) respectively. The softening of
average stress becomes steeper with the increase in yield strength provided that the slenderness ratio is constant.
Agreeing with previous study (Mau and El-Mabsout 1989), the average response of high strength bar is found to
become more brittle than that of lower strength bar. It was reported that the average response becomes nearly the
same as the local response if L/D ratio is equal to or smaller than 5 (Mau 1990; Monti and Nuti 1992). However, the
microanalysis suggested that this is not the reality and this critical slenderness ratio depends on the yield strength of
the reinforcing bar. For example, the average response of bar with slenderness ratio 5 is very close to local response
for fy = 100 MPa but the difference between the local and average responses is very big for fy = 1600 MPa. Hence,
rather than specifying the relationship in terms of slenderness ratio and yield strength separately, a combined term
including both of these parameters seems more suitable.
For this purpose, two pairs of special cases are compared with each other in Fig. 10(c). The first pair includes
two cases, one with fy = 100 MPa; L/D = 10 and the other with fy = 400 MPa; L/D = 5. Similarly, the second pair
includes the cases with fy = 400 MPa; L/D = 10 and with fy = 1600 MPa; L/D = 5. Interestingly, the normalized
average responses of the two cases in each pair were exactly similar to each other, suggesting that there is a unique
interrelationship between the parameter L/D√fy and the average response. Although some studies on modeling of
average compressive response of reinforcing bars have been performed in the past (Mau and El-Mabsout 1989; Mau
1990; Monti and Nuti 1992), this unique interrelationship between yield strength and average compressive response
was not realized. Next, the normalized average responses for different values of L/D√fy, using elastic-perfectly
plastic point wise stress-strain relationship, are shown in Fig. 10(d). As expected, buckling happens earlier and the
post-buckling average stress degrades faster with increase in the value of this parameter.

Reinforcing Bars with Strain Hardening

The earlier discussions were based on the microanalysis of reinforcing bars of elastic-perfectly plastic material.
In contrast, commonly used reinforcing bars exhibit hardening after a yield plateau in the post-yield range, though
the hardening stiffness and range may vary between the bars produced in different countries. To cover all types of
reinforcing bars, the parametric study hereafter is conducted for linear hardening material, which along with elastic-
perfectly plastic material constitute the two opposite extremes of all possible hardening mechanisms. The assumed
point wise stress-strain relationship is shown in Fig. 11(a), and the average compressive stress-strain curves of bars
with slenderness ratio varying between 5 and 15 are shown in Fig. 11(b)-(d) for yield strength equal to 200, 400 and
800 Mpa, respectively. As expected, the softening of average post-buckling compressive stress for the same
slenderness ratio becomes steeper with the increase in yield strength. Moreover, the starting point of negative slope is
delayed due to strain hardening effect, and the absolute value of normalized stress is also higher than that in elastic-
perfectly plastic reinforcement. The FEM microanalysis could successfully capture the average response of high
strength bars (fy = 800 Mpa) that is distinctly more brittle compared to the average response of normal strength bars
with same slenderness ratio.
Note that the negative slope of the average stress-strain curves in the later descending stage is nearly constant
(approximately equal to 2% of Young’s modulus), irrespective of the value of the parameter L/D√fy as in the elasto-
plastic bars. Similarly, the average compressive stress nearly becomes constant after it decreases to 20% of the yield
strength. This information can greatly assist in the modeling of average compressive behavior of reinforcing bars. In
this case too, the average response was found to uniquely depend on the parameter L/D√fy, irrespective of separate
values of L, D and fy. As the response of these two extreme cases (elastic-perfectly plastic and linear hardening)

5
indicate similar interrelationships between the slenderness ratio, yield strength and the average compressive
response, the same should be applicable to reinforcing bars with hardening mechanisms in between these extremes.
Hence, the limiting L/D ratio below which the average behavior is close to the material model can be better
approximated by (100/√fy) rather than 5 regardless of bar strength as recommended in previous studies.

MODELING OF AVERAGE STRESS-STRAIN RELATIONSHIP

Average Model for Monotonic Compression

As proved earlier, the average behavior of reinforcement in compression can be reliably captured if sufficiently
small elements are used in the finite element analysis. Nevertheless, it is impractical and tedious to use such small
elements in the analysis of real size reinforced concrete structures. While using elements of substantial size, the point
wise material models should be replaced with the average stress-strain relationships defined over the specified finite
element. Here, an average stress-strain relationship for reinforcing bars including the effect of buckling is
formulated.
Through the analytical parametric study, various facts regarding the average behavior of reinforcing bars in
compression are revealed. Some of them are: 1) The average compressive stress-strain relationship solely depends on
the product of slenderness ratio and yield strength. 2) The average compressive stress becomes smaller than the local
stress after the initiation of buckling, and the starting point of average stress degradation also depends on the point
wise stress-strain relationship. 3) The stress degradation rate in the later stage of post-buckling average stress-strain
curve is nearly constant with a negative slope approximately equal to 2% of the Young’s modulus. 4) The average
compressive stress becomes constant after it reaches the value equal to 20% of the yield strength.
Guided by these unique interrelationships, an average monotonic compressive stress-strain model is proposed,
the general layout of which is sketched in Fig. 12. An intermediate point (ε*,σ*) is established, after which a
constant negative stiffness equal to 0.02Es is assumed until the average stress becomes equal to 0.2fy. Here, Es and fy
are Young’s modulus and yield strength, respectively. To represent the aforementioned mechanisms, following
equations relating the absolute average compressive stress with the absolute average compressive strain of
reinforcing bar are proposed.

σ  σ *  ε − ε y 
= 1 − 1 − *  * ; for εy < ε ≤ ε*
σl  
 σ l  ε − ε y  (1)
σ = σ * − 0.02 E s (ε − ε * ); σ ≥ 0 .2 f y ; for ε > ε*

Here, σl and σl* are the local responses corresponding to ε (current strain) and ε* (strain at intermediate point),
respectively. Similarly, εy and Es are yielding strain and Young’s modulus of the reinforcement. To make the model
applicable to bars with all types of material model, the stresses at and before the intermediate point are normalized
with respect to the stresses computed from the point wise stress-strain relationship at the corresponding strain value.
This normalization technique also allows the shape of the average response before this intermediate point to look like
point wise stress-strain curve, which was distinctly observed in all microanalysis results. The coordinates of the
intermediate point (ε*,σ*) can be calculated by means of following equations.

ε* fy L
= 55 − 2.3 ; ε*/εy ≥ 7 (2)
εy 100 D

σ*  f y L 
= α 1.1 − 0.016 ; σ∗ ≥ 0.2 fy (3)
σ l*  100 D 
 

Note that equations (2) and (3) are derived from the microanalysis results of bars with linear hardening behavior
plotted in Fig. 13. When checked with the results of elastic-perfectly plastic bars, it was found that the normalized
strain at the intermediate point ε* was not much different, and equation (2) was still applicable without any
amendment. Nevertheless, the normalized value of stress at the intermediate point σ* was smaller than that in linear

6
hardening bars. Hence, for bars with more realistic hardening model, ε*/εy is expected to be unaffected but σ*/σl is
expected to be smaller than in linear hardening bars and larger than in elasto-plastic bars. To account for this change,
a coefficient α is included in the formulation of σ*. For the two extreme cases assumed earlier for the parametric
study, the value of α is found to be 1.0 for linear hardening bars, and 0.75 for perfectly elasto-plastic bars. For bars
with limited hardening range, in which most of the industrial products fall, it should be chosen between 0.75 and 1. If
the hardening stiffness is very small and the hardening range in terms of strain is short, the value should be closer to
0.75. On the other hand, if the hardening stiffness is significant and lasts for large strain range, the value should be
closer to 1.0.

Giuffre-Menegotto-Pinto Model for Cyclic Loops

As the framework for unloading and reloading behavior during strain reversals, the authors adopt Giuffre-
Menegotto-Pinto model (CEB 1996). As illustrated in Fig. 14, it uses a smooth transition curve asymptotic to the
tangents at the point of stress reversal and the point of maximum/minimum strain in the loading history. The original
model of Giuffre-Menegotto-Pinto takes on the form of normalized stress-strain relationship, as expressed in (4)-(6).

(1 − b)ε eq
σ eq = bε eq + (4)
R )1 / R
(1 + ε eq
ε − εr σ −σ r
ε eq = ;σ eq = (5)
ε0 − εr σ 0 −σ r
a1ξ
R = R0 − (6)
a2 + ξ

These equations represent a smooth transition from the unloading/reloading stiffness at the starting point to the
hardening stiffness at the minimum/maximum strain point in the loading history. In (4), εeq and σeq are normalized
strain and stress parameters and can be calculated according to (5), where (ε0, σ0) is the intersection point of the two
asymptotes (point A in Fig. 14), and (εr, σr) is the unloading/reloading point (point D in Fig. 14). These points are
updated after each strain reversal. Similarly, R is a parameter that influences the shape of the transition curve, and is
expressed as shown in (6). Here, ξ is the normalized strain difference between the current asymptote intersection
point (point A in Fig. 14) and the maximum/minimum strain ever experienced in the past loading history (point B in
Fig. 14). R0 is the value of the parameter R during first loading and should be experimentally defined along with the
constants, a1 and a2. The larger the positive value of R is, more abrupt the change of stiffness of the loop around the
intersection points of the two tangents becomes. Provided that the resulting value of R is positive, the transition
becomes smoother if the value of a1 increases or that of R0 and a2 decreases. However, making the transition
smoother will increase the difference between the maximum/minimum stress achieved earlier and the stress
represented by the transition curve at the target point. Hence, one should be careful while selecting the values of
these constants so that the value of R remains positive and the stress difference at the target point remains small to
avoid sudden stress jump that might cause convergence problem during FEM iterations. As these constants do not
influence the monotonic curve and small changes in their values marginally affect the transition shapes during strain
reversals, the authors have adopted R0 = 20; a1 = 18.5; and a2 = 0.15 after extensive checking.
In (4), b is the strain-hardening ratio, defined by the ratio between slope at the minimum/maximum point in the
loading history (E1 in Fig. 14) and slope at the unloading/reloading point. The unloading stiffness from tension is
equal to Es; i.e. the Young’s modulus of reinforcement. However, it has been observed that reloading stiffness from
post-buckling compression state is significantly smaller than Es (Suda et al. 1996; Monti and Nuti 1992). Hence, the
reloading stiffness from post-buckling compression state is reduced to E* as expressed in (7).

2
σ 
E * = E s  min 
 Es ≥ E* (7)
 σ max 

7
Verification of Proposed Computational Model

A complete path-dependent cyclic model relating the average stress with average strain is formulated by
combining the proposed monotonic compressive average stress-strain relationship with tension envelope of the bare
bar and Giuffre-Menegotto-Pinto model for the unloading/reloading loops. For verification, thus generated cyclic
model is compared with the results of FEM microanalysis as well as some experimental results.
First, the performance of the proposed model is compared with FEM microanalysis and experimental results of
the monotonic tests (Monti and Nuti 1992) in Fig. 15(a). As the point wise cyclic stress-strain relationship of the bar
material was not reported, the FEM microanalysis could not be carried out for cyclic tests, and the average cyclic
responses observed in the experiment are compared only with those computed by the proposed model, as illustrated
in Fig. 15(b)-15(d) for slenderness ratios 5, 8 and 11, respectively. As the hardening stiffness in the material model is
significant (0.055Es) but lasts only for 8εy, the coefficient in the proposed model is assumed as α = 0.9 according to
the earlier recommendation. As seen in the figure, the monotonic curves predicted by the model for different
slenderness ratios are close to the FEM microanalysis and experimental results. Moreover, it can also be observed
that the proposed model is in good agreement with the experimental results for all three cases of cyclic loading,
giving ample evidence that the proposed monotonic average compressive stress-strain relationship combined with
Giuffre-Menegotto-Pinto cyclic model can be reliably used.
Nevertheless, the verification for cyclic loading is conducted within the range of 3% tensile and compressive
strains; i.e. around 15 times yielding strain for normal strength bars. Although the 3% compressive strain is difficult
to be reached in the reinforcing bars in actual RC structures, tensile strain induced in reinforcing bars in RC
structures during strong earthquakes might be larger than the verified range. Test data measuring much higher strain
are scarce, which might be due to the difficulty in measuring higher strain with the commonly used strain gauges.
Note that the average compressive stress of the most slender bar (L/D = 11) unloaded from large tensile strain starts
to decrease before reaching the compressive strain zone because of the negative slope at the target point. This
behavior, observed both in experiment and model prediction, is qualitatively supporting the outcome of previous
study (Rodriguez et al. 1999) that the compressive stress degradation point (defined as buckling onset point) depends
on the maximum tensile strain reached before a strain reversal. The effect of maximum tensile strain might be more
pronounced if the strain reversal is from much larger strain, and the proposed model needs to be verified for such
large strain cyclic tests before drawing quantitative conclusions.
Next, the proposed model is compared with FEM microanalysis for some general cases. Figs. 16 and 17 show
the comparison between proposed model and FEM analysis for reinforcing bars with linear hardening and elastic-
perfectly plastic point wise stress-strain relationships, respectively. The comparisons are performed for different
values of the parameter FLD* = L/D√(fy /100). As established earlier, the coefficient α in the proposed model is
assumed as unity for the linear strain hardening case and 0.75 in elastic-perfectly plastic case. Average responses
predicted by the proposed model are very close to those by FEM microanalysis for almost all cases. As the model is
valid for these two extreme cases, the authors believe that it can represent the behavior of bars with any hardening
range. The earlier verification of the models with the available test data of bars with reasonable hardening range also
advocates this claim.

CONCLUSIONS

Microanalyses based on fiber technique using finite elements with small size were conducted to predict the
average behavior of reinforcing bars in compression accompanying geometrically large nonlinearity. It was revealed
that the average compressive behavior depends only on one parameter; i.e. L/D√fy (product of slenderness ratio and
square root of yield strength). It was also found that the conclusion from some previous studies, stating that the
average compressive response of reinforcing bars with slenderness ratio equal to or less than 5 is the same as the
local response, is not always true, especially in case of high-strength bars. Based on the results of parametric study,
an average compressive stress-strain model is proposed that can be applied for reinforcing bars with any geometrical
as well as mechanical properties. Giuffre-Menegotto-Pinto model for cyclic loops is combined with the proposed
model to yield a complete path-dependent cyclic model. The comparison of the microanalysis as well as the
proposed model with some experimental results showed good agreement and verified the reliability of proposed
model for both monotonic and cyclic loading.

8
ACKNOWLEDGEMENTS
The authors gratefully acknowledge TEPCO Research Foundation and Grant-in-aid for scientific research No. 11355021 for
providing financial support to accomplish this research.

APPENDIX. REFERENCES
CEB (1996). RC Elements under Cyclic Loading - State of the Art Report, Thomas Telford.
Dhakal, R. P., and Maekawa, K. (2000). "Post-peak cyclic behavior and ductility of reinforced concrete columns," Modeling of
Inelastic Behavior of RC Structures under Seismic Loads, ASCE, 193-216.
Dodd, L. L. and Restrepo-Posada, J. I. (1995). “Model for predicting cyclic behavior of reinforcing steel.” Journal of Structural
Engineering, 121 (3), 433-445.
Hauke, B., and Maekawa, K. (1999). “Three-dimensional modeling of reinforced concrete with multi-directional cracking.”
Journal of Materials, Concrete Structures and Pavements, JSCE, 634 (45), 349-368.
Mander, J. B., Priestley, M. J. N., and Park, R. (1984). "Seismic design of bridge piers." Research Report 84-2, Department of
Civil Engineering, University of Canterbury, New Zealand.
Mau, S. T. (1990). "Effect of tie spacing on inelastic buckling of reinforcing bars." ACI Structural Journal, 87 (6), 671-678.
Mau, S. T., and El-Mabsout, M. (1989). "Inelastic buckling of reinforcing bars." Journal of Engineering Mechanics, 115 (1), 1-
17.
Menegotto, M., and Pinto, P. E. (1973). “Method of analysis of cyclically loaded RC plane frames including changes in geometry
and non-elastic behavior of elements under normal force and bending.” Preliminary Report IABSE, 13, 15-22.
Monti, G., and Nuti, C. (1992). "Nonlinear cyclic behavior of reinforcing bars including buckling." Journal of Structural
Engineering, 118 (12), 3268-3284.
Rodriguez, M. E., Botero, J. C., and Villa, J. (1999). "Cyclic stress-strain behavior of reinforcing steel including effect of
buckling." Journal of Structural Engineering, 125 (6), 605-612.
Suda, K., Murayama, Y., Ichinomiya, T., and Shimbo, H. (1996). “Buckling behavior of longitudinal reinforcing bars in concrete
column subjected to reverse lateral loading.” Proc., 11th World Conference on Earthquake Engineering, Acapulco, Mexico, CD
ROM Paper No. 1753.

APPENDIX. NOTATION
a1, a2 = experimental constants to represent cyclic loop;
b = strain-hardening factor; b=E1/Es or b=E1/E*;
D = diameter of reinforcing bar;
Es = Young’s modulus of reinforcing bar;
E1 = tangential stiffness at maximum/minimum strain;
Esh1, Esh2 = first and second stage hardening stiffness;
E* = modified reloading stiffness from minimum strain;
fy = yield strength of reinforcing bar;
FLD* = parameter, FLD*= L/D√(fy /100);
L = test length of reinforcing bar;
R = parameter influencing the shape of the transition curve;
R0 = value of R during first loading;
u, v, w = displacements along X, Y and Z axes;
x, y = distance to a fiber along X and Y axes from center;
α = coefficient depending on the local behavior of bar;
ε, σ = average strain and average stress;
ε*, σ* = strain and stress at the intermediate point;
εf = axial strain of a reinforcement fiber;
ε0, σ0 = strain and stress at the intersection of two tangents;
εr, σr = maximum/minimum strain and corresponding stress;
εsh1, εsh2 = first and second stage local hardening strain;
εy = yielding strain of the longitudinal reinforcement;
εeq, σeq = normalized strain and stress,
εeq = (ε - εr)/( ε0 - εr); σeq = (σ - σr)/( σ0 - σr);
φx, φy, φz = curvatures with respect to X, Y and Z axes;
γzy, γzx = in-plane and out-of-plane shear strains;
θx, θy, θz = rotations with respect to X, Y and Z axes;
σmin, σmax = stresses at the minimum and maximum strain;
σl, σl* = local stresses at current strain and ε*, respectively;
ξ = normalized strain difference between the intersection point
of two asymptotes and the maximum/minimum strain.

9
LIST OF FIGURES:

FIG. 1. Fiber technique adopted in the finite element analysis

FIG. 2. Effect of reinforcement buckling in average response
FIG. 3. Simulation of bare bar compression test and effect of end condition
FIG. 4. a) Effect of geometrical nonlinearity
b) Effect of mesh dicretization
c) Lateral displacement profile after buckling
d) Variation of average strain along the length
FIG. 5. Strain and stress distributions across the cross-section after buckling
FIG. 6. Effect of path-dependency of fiber constitutive model on the average behavior
FIG. 7. Comparison with experimental results (Mander et al. 1984)
FIG. 8. Comparison with experimental results (Monti and Nuti 1992)
FIG. 9. Effect of length and diameter on the average response
FIG. 10. Effect of yield strength on the average response
FIG. 11. Average compressive responses of reinforcements with strain hardening
FIG. 12. Schematic representation of proposed model
FIG. 13. Determination of intermediate point based on microanalysis
FIG. 14. Giuffre-Menegotto-Pinto model for cyclic loop
FIG. 15. Comparison with monotonic and cyclic test results (Monti and Nuti 1992)
FIG. 16. Comparison of model with FEM analysis (linear strain-hardening case)
FIG. 17. Comparison of model with FEM analysis (elastic-perfectly plastic case)

10
 v, θy Yl
x u, θx
Fiber Xl

Zl
y w, θz
σ
Tension An element
Axial strain of fiber Strain-displacement relationships
Comp ε
εf = ε0 + φxy - φyx ε0 = δw/δz; γzy = δv/δz+θx; γzx = δu/δz-θy
+1/2θx2+1/2θy2 φx = δθx/δz; φy = δθy/δz; φz = δθz/δz
Fiber stress-strain curve

Fig. 1

11
Before buckling
εy
Strain εy σ Local behavior
ε
fy
A A fy
fy εy εy σ Average
behavior
ε
Stress fy
fy εy ε
σ A-A

Loading and Variation along Variation across Average and

deformation the length the section local behaviors
εy
∆ Strain εy σ Local behavior
ε
fy
A A fy
Strain
Stress

εy σ
ε Average
L behavior
fy Stress fy
εy ε
σ
A-A
ε = ∆/L: Averaged strain;
After buckling
L: Referential length in which average constitutive model is defined

Fig. 2

12
 90
D fy = 400 MPa
80 hinge
L = 16 cm
70
∆ D = 16 mm fixed

Axial load, kN
60
L 50
σ 40
D 30
20
ε 10
0
0 0.1 0.2 0.3 0.4 0.5
Axial disp, cm

Fig. 3

13
 90 90 A
80 80
70 f y=400MPa 70
Axia l load, kN

L=16cm, D=16mm

Axial load, kN
60 60 B
50 50 C
40 40
with ge ome trical 1 ele ment
30 no nlinearity 30 2 el eme nts f y=400MPa
20 without ge ome trical 20 4 el eme nts L=16cm
10 no nlinearity 10 6 el eme nts D=16mm
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
(a) Axial disp, cm (b) Axial disp, cm

16 16
Po int A
14 14 P oint A
Po int B
12 Po int C 12 P oint B
H eig ht, cm

P oint C
Heig ht, c m
10 10
8 8
6 after buckling d) Variation of average
6 strain along the length.
4 4
2 2
0 0
0 0. 1 0.2 0. 3 0. 4 0.5 -2. 5 -2 -1. 5 -1 -0. 5 0
(c) Lateral disp, cm (d) Strain at center, %

Fig. 4

14
 Point B
Strain, %
Stress, MPa

fy = 400 Mpa
Fiber strain Fiber stress
εy = 0.2 %
distribution distribution

Strain, %
Stress, MPa

Point C

Fig. 5

15
σ 90
Unloading/reloading 80
fy 70
fy = 400 MPa

Axial load, kN
60
L = 16 cm
Elastic loading 50
D = 16 mm
40
Compression 30 Path-dependent
20 model
Es
εy ε 10 Path-independent
model
Point wise stress- 0
strain relationship 0 0.05 0.1 0.15 0.2
Axial disp, cm

Fig. 6

16
 P 1.6
Analysis
1.4

Normalized stress ( /fy)

Experiment
L
1.2 Local
D =16 mm
L/D = 6
1
σ
σu 0.8 L/D = 10

fy 0.6
0.022Es
0.4
fy = 290 Mpa L/D = 15
σu = 433 MPa 0.2
Es Es = 200 GPa ε 0
εy 11.7εy 35.1εy 0 10 20 30 40
Point wise stress-strain Normalized strain (ε /ε y)
relationship

Fig. 7

17
 P 1.6

Normalized stress ( /fy)

1.4
L L/D = 5
1.2
D
1 L/D = 8
σ
0.8
fy
0.055Es 0.6

fy = 480 MPa 0.4 Analysis

Es = 200 GPa Experiment
L/D = 11
Es 0.2
ε Local
0
εy 2εy 10εy
0 5 10 15 20
Point wise stress-
strain relationship
Normalized strain (ε /ε y)

Fig. 8

18
 1 .2 1 .2
f y=400MPa, D=16mm f y=400MPa, L=12cm
1 1

No rmaliz ed stres s
Normaliz ed s tres s

0 .8 0 .8

0 .6 0 .6
L=8cm
0 .4 L= 12cm 0 .4
L= 16cm D=20mm D=16mm
0 .2 0 .2
L= 24cm D=12mm D=10mm
D=8mm
0 0
0 5 10 15 20 0 5 10 15 20 25
(a) Normalize d strain (b) Normalized strain

1 .2 1.2
L=16cm, D=16mm
f y=400MPa f y=400MPa
1 1

N ormal iz ed stre ss
No rmaliz ed stres s

L=12cm, D=12mm
L=24cm, D=16mm 0.8
0 .8
L= 12cm, D=8mm
0 .6 0.6

0 .4 0.4

0.2 L/D=6 L/D=7.5

0 .2
L/D=10 L/D=12
0 L/D=15
0
0 5 10 15 20 25 0 5 10 15 20 25
(c) Normalized strain (d) Norma liz ed strain

Fig. 9

19
 1.2 1.2
L/D=5 L/D=10
1 1

No rmalize d stre ss
Normal iz ed stre ss

0.8 0.8

0.6 0.6
fy=1 00MPa
0.4 fy=2 00MPa 0.4 fy= 100MPa
fy=4 00MPa fy= 200MPa
0.2 fy=8 00MPa 0.2 fy= 400MPa
f y=16 00MPa fy= 800MPa
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25
(a) Norma liz ed strain (b) Normaliz ed strain

1.2 1.2

1 1
Norma lize d stre ss

No rmalize d stre ss
0.8 0.8
fyLD*=5
0.6 0.6 fyLD*=15

fy= 100MPa, L/D=10 fyLD*=20

0.4 0.4
fy= 400MPa, L/D= 5 fyLD*=24
0.2 fy= 400MPa, L/D=10 0.2 fyLD*=30
fy= 1600MPa, L/D=5 fyLD*=(fy /100) 0.5 *L/D fyLD*=10
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 35
(c) Normalized strain (d) Normaliz ed strain

Fig. 10

20
 40εε y 8εε y εy 1.8
Local
1.6

N orma liz ed stress

ε 1.4 fy=200MPa
Es 1.2
1
0.8
0.6
L/D=6 L/D=8
0.03Es fy 0.4
L/D=10 L/D=12
0.2
L/D=15
0
Point wise stress- 0 5 10 15 20 25 30 35
(a) σ
strain relationship (b) Normaliz ed strain

1.6 1.6
Local L/D=5 L/D=6
1.4 1.4
L/D=8 L/D=10
N ormalized stress

fy=400MPa

N ormalized stress
1.2 1.2 Local
L/D=15
1 1
0.8 0.8
0.6 0.6
0.4 L/D=5 L/D=6 0.4
L/D=8 L/D=10 0.2
0.2
L/D=12 fy=800MPa
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 35
(c) Normalized strain (d) Normalized strain

Fig. 11

21
 σ
Local
(ε*,σl*)
fy
Model
(ε*,σ*) 0.02Es

0.2fy
Es Analysis
ε
εy

Fig. 12

22
 40 1.2
35
1
30 y = 55 - 2.3x y = 1.1 – 0.016x
25
y

0.8

l
20
/

/
*

*
15 0.6
10
0.4
5
0 0.2
0 10 20 30 40 50 0 10 20 30 40 50
0.5 0.5
L/D*(fy/100) L/D*(fy/100)

Fig. 13

23
 ξ2εy
(εr1,σr1) B
(ε0 ,σ0 )
2 2
E1
A
Steel stress, σ
(b)

(a) Es

εy Steel strain, ε

(εr2,σr2) C (ε01,σ01)

ξ1εy

Fig. 14

24
 1.6 800
Normalized stress (σ /fy)
f y = 480MPa
L/D = 5 Model
1.4 600

Average st ress, MPa

Ex peri ment
1.2 L/D = 8 400
1 200
0.8 0
0.6 -200
0.4 Analysis -400
0.2 Model f y = 480MPa
L/D = 11 -600
Local L/D = 5
0 -800
0 5 10 15 20 -0.04 -0.02 0 0.02 0.04
(a) Normalized strain (εε /εε y) (b) Average strain

8 00 8 00
Model Model
6 00

Average stress, MPa

Average stress, MPa

Experiment 6 00 Exper iment

4 00 4 00
2 00
2 00
0
0
-2 00
-4 00 -200

-6 00
f y = 480MPa -400
L/D = 8 f y = 480MPa
-8 00 -600 L/D = 11
-0.04 -0.02 0 0.02 0.04 -0.04 -0.02 0 0.02 0.04
(c) Average strain (d) Average strain

Fig. 15

25
 P 1.6
Model FLD* = 10

Normalized stress ( /fy)

1.4
Analysis 12
L 1.2
D 16
1
σ
P 0.8
fy 20
0.02Es 0.6
24
0.4
30
0.2
Es ε FLD* = L/D√(fy/100)
0
εy 8εy
Point wise stress- 0 5 10 15 20 25 30
strain relationship Normalized strain (ε /ε y)

Fig. 16

26
 P 1.2
FLD* = L/D√(fy/100)

Normalized stress ( /fy)

1
L FLD* = 10
D 0.8

σ P 0.6 12
15
fy 0.4 20

Model 24
0.2
Analysis 30
Es ε
0
εy 0 5 10 15 20 25 30
Point wise stress- Normalized strain (ε /ε y)
strain relationship

Fig. 17

27

View publication stats