10 1061@asce0733-94451991117@61573

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

ANALYSIS O F F R A M E S WITH NONPRISMATIC

MEMBERS
By Nasreddin El-Mezaini,1 Can Balkaya,2 and Ergin Citipitioglu,3
Member, ASCE
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

ABSTRACT: The linear elastic behavior of frames with nonprismatic members is


investigated by using isoparametric plane stress finite elements. It is determined
that the conventional methods of analysis for these types of structures lead to er-
roneous results. Comparison of the fixed end moments, stiffness, and carry-over
factors of nonprismatic members available in the literature with those computed
by finite element analysis reveals large discrepancies. Based on an extensive study,
sources and magnitudes of errors are presented. Recommendations for proper mod-
eling by using conventional frame analysis computer programs are made.

INTRODUCTION

Frequently, beams are deepened by haunches near the supports to increase


the support moment, which results in a considerable reduction of the span
moment. Consequently, midspan depth can be reduced in order to obtain
more clearance and/or less structural height. Nonprismatic columns are also
commonly used in engineering practice, e.g., those supporting crane girders
in industrial buildings. Fig. 1 shows different types of nonprismatic mem-
bers.
The analysis of frames having nonprismatic members is covered in many
classical textbooks on structural analysis. Cross (1958) presented stiffness,
carry-over factors, and fixed end moments of nonprismatic members for uti-
lization of the moment distribution method. More extensive tables and for-
mulas are given in a booklet published by Portland Cement Association (1958)
for the computation of beam constants necessary in the analysis of frames
by the moment distribution and the slope deflection methods.
Recently, Raymond and Wang (1988) reported a numerical technique for
the calculation of the stiffness matrix of nonprismatic members. Resende and
Doyle (1981) and Mumuni (1983) separately developed finite element models
for nonprismatic beams. Fertis and Keene (1990) presented a method for the
analysis of nonprismatic members.
In all of the references cited, analysis is based on formulas that were de-
rived for prismatic members in which the centroidal axis is a straight line,
while in nonprismatic members either the axis or its slope is discontinuous,
as is shown in Fig. 1. Thus, it would be futile to strive for accurate analysis
by using sophisticated mathematical tools when the fundamental assumptions
are only approximate.
The purpose of this paper is to investigate the behavior of beams and
frames having members of variable depth. Based on a finite element anal-
'Res. Asst., Dept. of Giv. Engrg., Middle East Tech. Univ., 06531 Ankara, Tur-
key.
2
Project Engr., Ener. Engrg., Consulting & Contracting Corp., Ankara, Turkey.
3
Prof. of Civ. Engrg., Middle East Tech. Univ., 06531 Ankara, Turkey.
Note. Discussion open until November 1, 1991. To extend the closing date one
month, a written request must be filed with the ASCE Manager of Journals. The
manuscript for this paper was submitted for review and possible publication on April
10, 1990. This paper is part of the Journal of Structural Engineering, Vol. 117,
No. 6, June, 1991. ©ASCE, ISSN 0733-9445/91/0006-1573/$1.00 + $.15 per page.
Paper No. 25874.

1573

J. Struct. Eng. 1991.117:1573-1592.


(a)
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

(c)

FIG. 1. Types of Nonprismatic Members: (a) Stepped Members; (to) Members with
Straight Haunches; (c) Members with Parabolic Haunches

ysis, sources and magnitudes of errors in the conventional analysis are dis-
cussed. Recommendations for proper modeling of this type of structures by
using conventional frame analysis computer programs are made.

FINITE ELEMENT ANALYSIS

The plane stress finite element solution is adopted for the two-dimensional
elastic analysis of the structures under investigation.
A special purpose computer program utilizing quadratic and cubic iso-
parametric finite elements was developed. Automatic element and node gen-
eration for structures of arbitrary shape was incorporated. The number and
type of elements used can easily be changed for the purpose of mesh re-
finement.
In order to make sure that there is sufficient accuracy in the finite element
analysis (FEA), convergence studies were performed for each case by re-
fining the mesh and solving the same model by using first quadratic and
then cubic elements. As a result, near-optimum meshes were determined that
prove an accuracy better than 0.18% in stresses.
The previously mentioned computer program was used for investigating
the behavior of different types of nonprismatic members. Stiffness, carry-
over factors, and fixed end moments for such members were computed. Full
frames with nonprismatic members were also analyzed. The results obtained
are compared with those of conventional analysis.

GENERAL BEHAVIOR OF NONPRISMATIC MEMBERS

The behavior of nonprismatic memberes differs from that of prismatic ones


due to the variation of cross section along the member and the discontinuity
of the centroidal axis or its slope. The effects of axis geometry and cross
section variation on the behavior and on the distribution of stresses in non-
prismatic members are discussed in the following sections.
1574

J. Struct. Eng. 1991.117:1573-1592.


(a) (c)

M = P.e
_r
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

(b) (d)

FIG. 2. Axis Discontinuity

Discontinuity of Centroidal Axis


From the mathematical point of view, the centroidal axis in a stepped
member or its derivative in a haunched member is "discontinuous." There-
fore, the differential equation of beam bending is not valid at the points of
discontinuity.
To investigate the effect of centroidal axis discontinuity, two stepped
members having identical dimensions but different axes as shown in Fig.
2(a and b) are considered. A uniform temperature change produces only
axial force in the member shown in Fig. 2(b), whereas in the member shown
in Fig. 2(a), bending moment also develops in addition. Furthermore, if the
same two members are subjected to vertical loading or end rotation, member
(b) stays free of axial force, whereas in member (a) axial force develops.
The totally different behavior exhibited by the two members is attributed to
the difference in the geometry of their axes.
Similar behavior also exists if the variation of the depth is at one end only
as shown in Fig. 2(c and d). It is seen that an axial force in one segment
produces bending moment in the other due to the eccentricity of the centroi-
dal axis.
It is concluded that in these types of members, there exists coupling be-
tween bending, shear, and axial forces. The coupling effect becomes more
significant with an increase in depth of the end segments.
It should be noted that in the conventional analysis, both members (a) and
(b) shown in Fig. 2 are assumed to be identical in terms of stiffness, carry-
over factors, and fixed end moments.

Stress Distribution over Cross Section


Under the same internal forces, the stress distribution over a cross section
of a nonprismatic member is different than that of a prismatic one having
the same dimensions. Timoshenko (1963) showed that shear stress distri-
bution over a cross section of a bar, in the form of a cantilever wedge, is
in direct opposition to that of prismatic bars (see Fig. 3).
Timoshenko stated that: "In many cases the practical shear stresses are of
no great importance and only the normal bending stresses are considered.
Then the formula for maximum bending stress, derived for prismatic beams,
can be used with sufficient accuracy for bars of variable cross section, pro-
vided the variation of the cross section is not too rapid." Apparently, it is
because of this statement that prismatic beam formulas are usually used for
nonprismatic beams and the effect of shear stresses is neglected.
1575

J. Struct. Eng. 1991.117:1573-1592.


Normal Shear Normal Shear
(a) (b)
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

FIG. 3. Stress Distribution: (a) Prismatic Member; (b) Cantilever Wedge

The distribution of stresses at different locations of stepped and haunched


cantilevers with rectangular cross sections obtained from FEA are shown in
Fig. 4. In this figure, normal stresses are due to pure bending, whereas shear
stresses are due to constant shear force.
From Fig. 4, it can be seen that the distribution of stresses differs in each
member depending on the manner of variation of the cross section. It is also
noticed that the neutral axis is above the centroidal axis in the haunch zone
and near the step of the haunched and stepped members, respectively. Shift-
ing of the neutral axis is attributed to the asymmetrical shear stress distri-
bution about the centroidal axis.
Stress Flow in Nonprismatic Members
The points of discontinuities in the member geometry cause disturbances
in the stress flow. In stepped members, due to the abrupt change of the cross

HEUTRAL AXIS NEUTRAL AXIS

(a)

(b)

FIG. 4. Stress Distribution in Nonprismatic Members: (a) Normal Stresses Due


to Pure Bending Moment; (b) Shear Stresses Due to Constant Shear Force

_--4
1'

Null Area Kink

(b)

FIG. 5. Principal Stress Contours

1576

J. Struct. Eng. 1991.117:1573-1592.


section, parts of the member are subjected to significantly small stresses,
denoted as "null areas" in Fig. 5(a) [this may also be observed in Fig. 4(a)].
Due to their small contribution, the existence of the null areas results in a
reduction of the member stiffness.
In haunched members, there are "kinks" on the centroidal axis that exhibit
slope discontinuities as shown in Fig. 5(b). The overall behavior of haunched
members is similar to that of stepped members, except that the reduction in
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

stiffness due to stress flow in end segments is less significant.

FIXED END MOMENT, STIFFNESS, AND CARRY-OVER FACTORS

By definition, the bending stiffness A" of a member is the moment required


to produce a unit rotation at one end while all other degrees of freedom are
set to zero. The carry-over factor C is the ratio of the moment developed at
the other end to the applied moment.
In FEA, end rotation is represented by a set of prescribed nodal displace-
ments. K- and C-factors are computed by normalizing the computed end
moments for a unit rotation. Fig. 6(a) shows a typical near-optimum mesh
configuration with fixed boundary, denoted as model 1. This model is also
used for the fixed end moment (FEM), which is calculated about the centroid
of the end section. Due to axial deformation restraints in model 1, significant
horizontal forces are produced at the boundaries that affect the results, as
will be shown in the following section.
To eliminate the effect of axial forces, model 2 [Fig. 6(b)] is developed.
In this model, one end is totally fixed. The other end is supported at the
center node only by a roller support so that no axial restraint exists. End
moment is represented by a set of prescribed nodal forces. K- and C-factors
are computed by normalizing the applied end moment to the obtained end
rotation. To ensure linear end rotations, a rigid block is attached to the beam
end.
Tables 1-3 and Fig. 7 show FEM, K-, and C-factors for stepped and
haunched members, respectively. In these tables, values obtained from model

End Rotation

<")

Rigid block
A T
^
fex
End Moment
f- AL L - 2AL AL
--+ -\
(b)

FIG. 6. Finite Element Models: (a) Model 1 (Fixed End Boundary); (b) Model 2
(Roller End Support)

1577

J. Struct. Eng. 1991.117:1573-1592.


TABLE 1. Constants for Stepped Members
FEM wL2 K-EI/L C-Factor
R Method F,-wL FEM D%a K D% a C D%"
(1) (2) (3) (4) (5) (6) (7) (8) 0)
0.0 PCA NA 0.083 4.00 — 0.500 —
FE1 0.00 0.083 00 3.95 01 0.495 01
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

FE2 NA NA — 3.95 01 0.495 01


0.4 PCA NA 0.097 — 7.32 — 0.634 —
FE1 0.188 0.093 04 6.98 05 0.530 16
FE2 NA NA — 7.11 03 0.573 10
0.6 PCA NA 0.101 — 8.80 — 0.674 —
FE1 0.197 0.094 07 8.39 05 0.469 30
FE2 NA NA — 7.43 16 0.633 06
1.0 PCA NA 0.105 — 11.09 — 0.723 —
FE1 0.202 0.092 11 11.30 -02 0.264 63
FE2 NA NA — 8.45 24 0.666 08
1.5 PCA NA 0.107 — 12.87 — 0.752 —
FE1 0.170 0.091 15 15.18 -18 -0.021 103
FE2 NA NA — 9.00 30 0.685 09
2.0 PCA NA 0.108 — 13.87 — 0.765 —
FE1 0.151 0.088 19 20.59 -48 -0.260 134
FE2 NA NA — 9.20 34 0.691 10
"Deviation relative to PCA; D% = [(PCA - value)/PCA] • 100.

1 are denoted as FE1 and those from model 2 as FE2. Portland Cement
Association (PCA) values for the same members are also listed.
The members are stepped or haunched at both ends with variable depth
ratio R. Step or haunch length ratio A = 0.2L. Depth-to-span ratio H/L is
not specified in PCA tables since shear and axial deformations are not con-
sidered. In FEA, however, different H/L ratios result in slightly different
K- and C-coefficients due to shear and axial deformations. Therefore, H/L
= 0.1 and Poisson's ratio = 0 are chosen as practical values for the purpose
of comparison with PCA beam constants.
FEMs and the developed fixed-end horizontal forces Fx are due to uni-
formly distributed vertical load W applied at the top surface of the members.
Plots for all constants listed in Tables 1-3 are provided in Appendix I.

W
FEM
H t
RH J

^••••0 . 2LT|«- 0 . 6L Jf—0 . 2 L J

FIG. 7. Dimensions, Loading and End Forces Utilized in Tables 1-3

1578

J. Struct. Eng. 1991.117:1573-1592.


TABLE 2. Constants for Members with Straight Haunches
FEM wL2 K-EI/L C-Factor
a a
R Method Fx-wL FEM D% K £>% C D%"
(1) (2) (3) (4) (5) (6) (7) (8) (9)
0.0 PCA NA 0.083 — 4.00 '— 0.500 —
FE1 0.00 0.083 00 3.95 ' 01 0.495 01
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

FE2 NA NA — 3.95 01 0.495 01


0.4 PCA NA 0.093 — 5.75 — 0.588 —
FE1 0.146 0.090 03 5.95 -03 0.471 20
FE2 NA NA — 5.55 03 0.574 02
0.6 PCA NA 0.095 — 6.51 — 0.618 —
FE1 0.170 0.090 05 7.12 -09 0.391 37
FE2 NA NA — 6.18 05 0.600 30
1.0 PCA NA 0.099 — 7.81 — 0.659 —
FE1 0.182 0.089 10 9.91 -27 0.175 73
FE2 NA NA — 7.12 09 0.634 04
1.5 PCA NA 0.102 — 9.08 — 0.691 —
FE1 0.180 0.087 15 14.42 -59 -0.097 114
FE2 NA NA — 7.88 13 0.657 05
2.0 PCA NA 0.104 — 10.05 — 0.711 —
FE1 0.171 0.085 18 20.25 -101 -0.312 144
FE2 NA NA — 8.37 17 0.670 06
'Deviation relative to PCA; D% = [(PCA - value)/PCA] • 100.

OBSERVATIONS

From Tables 1-3 it can be seen that there are large discrepancies between
the rescults obtained from the three methods of analysis (FE1, FE2, and
PCA). The reason for these discrepancies and the factors affecting the be-
havior of such members are discussed in the following sections.

Arching Action
In the analysis of model 1 (FE1), in which axial deformation is restrained,
it is observed that under vertical loading, horizontal thrust Fx is produced at
the boundaries. The magnitude of Fx increases rapidly with R up to a certain
limit, then it starts to decrease slightly. This is attributed to the existence of
a built-in arch formed due to the eccentricity of the centroidal axis as shown
in Fig. 8(a).
In beams with a small R, the built-in arch is very shallow so that with a
slight axial deformation, it will become flat, producing a small axial force.
The beam then continues to deform under bending as a prismatic one. As
R increases, the arch depth increases, requiring more axial deformation to
become flat; therefore, more horizontal thrust is produced. When the arch
is deep enough so that it cannot be completely flattened with axial defor-
mations, horizontal thrust starts to decrease as arch height increases, simi-
larly to regular arches.
Arching action helps the beam to resist the rotation near the supports which,
1579

J. Struct. Eng. 1991.117:1573-1592.


TABLE 3. Constants for Members with Parabolic Haunches
FEM wL K-EI/L C-Factor
a
R Method F,-wL FEM D% K £>%" C D%
(1) (2) (3) (4) (5) (6) (7) (8) (9)
0.0 PCA NA 0.083 — 4.00 — 0.500 —
FEl 0.00 0.083 00 3.95 01 0.495 01
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

FE2 NA NA — 3.95 01 0.495 01


0.4 PCA NA 0.090 — 5.16 — 0.565 —
FEl 0.116 0.088 02 5.40 -05 0.433 23
FE2 NA NA — 4.98 04 0.553 02
0.6 PCA NA 0.093 — 5.63 — 0.587 —
FEl 0.130 0.088 05 6.33 -12 0.328 44
FE2 NA NA — 5.35 05 0.572 03
1.0 PCA NA 0.096 — 6.40 — 0.619 —
FEl 0.149 0.087 09 8.73 -36 0.076 88
FE2 NA NA — 5.88 08 0.595 04
1.5 PCA NA 0.098 — 7.17 — 0.645 —
FEl 0.157 0.085 13 12.93 -80 -0.210 132
FE2 NA NA — 6.33 12 0.613 05
2.0 PCA NA 0.099 — 7.77 — 0.662 —
FEl 0.158 0.082 16 18.59 -140 -0.418 163
FE2 NA NA — 6.66 14 0.625 06
"Deviation relative to PCA; D% = [(PCA - value)/PCA] • 100.

in turn, increases K and decreases the FEMs. Therefore, in FEl, the FEMs
are less and K-factors are greater than those given in the PCA (1958) pub-
lication for both stepped and haunched members. The deviation increases
with R.
The axial forces developed due to arching action tend to counteract the
carried-over moment M„ [see Fig. 8(c)], which causes reduction in the C-
factors in FEl. Eccentricity e increases with R. Consequently, the moment
due to axial force P • e increases, causing larger reduction in the C-factor.
This explains the large deviation in C-factors, which even becomes negative
when P • e is greater than M0, as can be seen in the cases of deep haunches.

Effect of Null Areas


In model 2 (FE2), due to the absence of axial restraints, the effect of axial
force, arching action, is eliminated. Therefore, better agreement between

# P.e

£F=FC=\ ^
Built-in arch
(a) (b) (c)

FIG. 8. Arching Action

1580

J. Struct. Eng. 1991.117:1573-1592.


FE2 and PCA values is obtained. However, it is noticed that the stiffness
K in FE2 is less than that of PCA for both stepped and haunched members
and that the reduction is greater in the case of a stepped member. This is
attributed to the effect of shear deformation and, in case of stepped mem-
bers, the effect of the null areas, which cause softening of the member stiff-
ness.
In FE2, the deviation in C-factors is less than that in K-factors, which
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

means that in the absence of axial displacement constraints, forces in non-


prismatic members are not as sensitive as displacements to the geometry of
the centroidal axis. This explains the large error in displacement reported by
Resende and Doyle (1981) when they modeled a haunched beam by one
straight-line element as a part of a continuous beam in the absence of axial
displacement constraints. In FE1 on the other hand, the deviation in C-f ac-
tors are greater than that in K-factors, which indicates the influence of arch-
ing action on the forces of the member. These facts are demonstrated in an
example problem later in this paper.
From the preceding discussion, it can be concluded that FEM, K- and C-
factors given in PCA publications are in significant error, especially for
members with deep haunches. The FEMs in PCA are generally overesti-
mated. K- and C-factors of FE2 do not satisfy the definition of bending
stiffness and carry-over factors because axial deformation is not restrained.
Accordingly, FE1 values represent the correct ones. However, if the member
stiffness matrix is to be formed using K- and C-factors obtained from F E 1 ,
.coupling with axial forces must be considered.
Formulation of the stiffness matrix for the different types of nonprismatic
members may not be a practical solution. Alternatively, better results can
be obtained by appropriate modeling of nonprismatic members using the
commonly used two-noded beam elements.
It should also be mentioned that the 1% deviation in K- and C-factors of
the prismatic member, R = 0.0, is due to shear deformations rather than
approximations of the FEA. It was found that this discrepancy diminishes
for very slender members.

SOURCES OF ERROR IN CONVENTIONAL ANALYSIS


In the conventional methods of analysis, nonprismatic members are usu-
ally analyzed using beam idealization, i.e., the member is divided into seg-
ments having one straight-line axis. Using beam theory, the analysis is per-
formed by integrating the energy equation for each segment of the member
separately without considering the discontinuity of the centroidal axis. In this
analysis there are two major sources of error as follows:

1. Neglecting the discontinuity of the centroidal axis results in neglecting the


coupling between axial and other end forces, which becomes a major source of
error.
2. Neglecting the existence of kinks and null areas, especially in stepped
members, results in overestimating the member stiffness.

MODELING OF NONPRISMATIC MEMBERS

The plane stress isoparametric finite elements offer an excellent solution


for most structures with or without nonprismatic members. However, for

1581

J. Struct. Eng. 1991.117:1573-1592.


f— AL -Jf L - 2 AL \ Al -\

(a)
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

|— AeL—| L - 2 AeL 1— AeL—|

(b)

FIG. 9. Modeling of Members with Straight Haunches

practical applications, it is desirable to obtain efficient models for nonpris-


matic members composed of the commonly used two-noded beam elements.
For this purpose, several models for haunched and stepped members are
tested using the SAP80 (1984) computer program. The results are compared
to those obtained from the FEA. Each model is tested for different loading
conditions. As a result the most efficient models are determined as follows.

Members with Straight Haunches


As a first trial, the members are modeled by dividing the haunch zone,
along its centroidal axis, into small segments of beam elements [see Fig.
9(a)]. For each segment, the average area and inertia of the cross section
are considered.
This model yielded stiffer structures than FEA even if a sufficient number
of segments are used in the haunch zone. The computed horizontal reactions
obtained at the supports were also high. The reason is that in these models
all of the haunch zone is considered, neglecting the effect of kinks and null
areas that cause softening of the member stiffness. Therefore, a modified
model is developed by removing a part of the haunch zone [see Fig. 9(b)].
By doing so, it is intended that the member stiffness be reduced. Also, this
reduces the horizontal reaction component Fx, due to the increase in the slope
of the axis.
As a result of several trials, it is found that the use of 75% of the original
haunch length, as an effective length Ae, gives good results for different
haunch dimensions. The effective haunch zone is represented by one beam
element using average cross-sectional area and inertia.

Members with Parabolic Haunches


Following the same procedure used for straight haunches, the effective
length of the parabolic haunch was investigated. The model shown in Fig.
10, in which Ae = 0.5A, was found to give good results. The modified
1582

J. Struct. Eng. 1991.117:1573-1592.


4k.
X"'"l
r en oved a r e a -""

\ LAa f- L -2AeL -f-LA. ^


Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

(b)

FIG. 10. Modeling of Members with Parabolic Haunches

haunch zone is also represented by one beam element for which average
area and inertia are used.

Modeling of Stepped Members


The case of the stepped member was investigated by Balkaya (1987). In
his study, Balkaya tested three alternative models as shown in Figs. l l ( a -
c). Among the three models, the one shown in Fig. 11(Z?) exhibited the best
agreement with FEA. In this model, the lower corner, a null area, is re-
moved at an angle of 45°. For the transition segment, average area and in-
ertia are used.

Comments on Applicability
It should be noted that the models proposed for members with straight and
parabolic haunches are valid only at the fixed or continuous ends, as shown
in Figs. Yl(a and b). If the member end is not continuous, or if it is con-
nected to a column as shown in Fig. 12(c), the best results are obtained by
considering the full haunch zone with the original centroidal axis.
When a nonprismatic member is modeled as part of a structure, the lo-
cations of supports relative to the centroidal axis of the members should be
carefully considered.

T r a n s i t i o n segment Rigid l i n k

d/2
d/2.
±
cir
4
(a) (b) (c)

FIG. 11. Modeling of Stepped Members

n=l t
(a) (b) (c)

FIG. 12. End Conditions of Nonprismatic Members

1583

J. Struct. Eng. 1991.117:1573-1592.


STRUCTURES HAVING NONPRISMATIC MEMBERS

In this section, the results of the analysis of structures having nonprismatic


members by different methods are presented. The accuracy of the conven-
tional methods of analyzing this type of structures is investigated. The va-
lidity of the proposed models for nonprismatic members is demonstrated.
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

Beam Example
The beam shown in Fig. 13 is analyzed for two support conditions: (1)
Support B is a roller (no horizontal reaction); and (2) Support B is a hinge
(with horizontal reaction).
The analyses are as follows:

1. Analysis I: FEA is performed using the near-optimum mesh shown in Fig.


13(a).
2. Analysis II (proposed): The models proposed in the previous section used
six two-noded beam elements as shown in Fig. 13(£>). One additional rigid mem-
ber is used to account for the actual location of support B.
3. Analysis III (conventional): The beam is modeled by using nine beam ele-
ments having straight centroidal axis as shown in Fig. 13(c). For each segment,
average area and inertia are considered. Analysis is performed using the SAP80
program.

The results obtained by the three methods are listed in Table 4 for the two
conditions of support B.
As can be seen from Table 4, the results of Analysis II are in good agree-

P =100 kN w=io kN/m

%M »Fy

^•-2.0
U -"

tn •%--— 3 . 0 B
c

^- 3.0 ra
--^j
——-
B

-—-jf-- J . O n - |
Nr 3 . O m •- -|

, f 1 . 0 ^ 1 . 0^- 3.0m Jf— -)f\ . 0 ^ 1 . 0 ^ 1 . 0 ^ 1 . 0 ^ 1 . O^

(c)

FIG. 13. Beam Example: (a) Finite Element Model (Analysis I); (b) Proposed Model
(Analysis II); (c) Conventional Model (Analysis III)

1584

J. Struct. Eng. 1991.117:1573-1592.


TABLE 4. Reactions and Deflections of Beam Example
Support A Support B Deflection
Moment F„ Fy F, Fy atC at D
Analysis (kN-m) (kN) (kN) (kN) (kN) (mm) (mm)
(1) (2) (3) (4) (5) (6) (7) (8)
(a) Roller Support at B
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

I 380.0 — 133.5 — 96.5 -10.70 9.62


II 379.0 — 133.4 — 96.6 -10.80 9.78
III 389.0 — 134.4 — 95.6 -10.10 9.13
(b) Hinged Support at B
I 219.3 114.3 106.0 -114.3 124.0 -6.00 1.17
II 213.3 118.3 105.0 -118.3 125.0 -6.00 0.98
III 389.0 0.0 134.4 0.0 95.6 -10.10 9.13

ment with those of Analysis I for the two conditions of support B , indicating
the validity of the proposed models. The results of Analysis III, on the other
hand, exhibit large discrepancies, making no distinction for the support con-
dition, which is expected because of the assumed straight-line axis.
For the beam with roller support, the results obtained from Analysis III
are in fairly good agreement with those of Analyses I and II. However, the
discrepancies in displacements are slightly higher than those in forces. This
is consistent with the fact that forces are less sensitive to the geometry of
the centroidal axis in the absence of axial forces. Moment and vertical re-
actions depend primarily on the horizontal distance between the two supports
A and B.
For the hinged support condition, the errors in Analysis III are unaccept-
ably large for both forces and displacements. This shows the effect of cou-
pling with axial forces produced by arching action. It should be pointed out
that in reality, even roller supports produce horizontal reaction, due to fric-
tion, which is critical in this case.

Portland Frame Example


The portal frame with a haunched beam shown in Fig. 14 is considered.
The analyses are as follows:

1. Analysis I: FEA is performed using the near-optimum mesh shown in Fig.


14(a).
2. Analysis II (proposed): The frame is represented by five beam elements
located exactly at the original centroidal axis as shown in Fig. 14(&). For the
elements representing the haunch zones, average area and inertia are used. Anal-
ysis is performed using the SAP80 program.
3. Analysis III (conventional): Moment distribution analysis is performed us-
ing PCA beam constants. Fig. 14(c) shows the model that is commonly used for
such structures.

Reactions and moment diagrams obtained from FEA, SAP, and PCA are
shown in Fig. 15(a), \5(b), and 15(c), respectively.
As can be seen from Fig. 15, the results of Analysis II are in good agree-

1585

J. Struct. Eng. 1991.117:1573-1592.


1 0 . 0 kN/m

f
1.17m
3ffl
^ 1 . ^ - 1 . 5 +• 6 . 0 IB + 1.5 +1.+
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

(«)

1 1 If- 2 . 0m ^ —

(b)

(c)

FIG. 14. Portal Frame Example: (a) Finite Element Model (Analysis I); (b) Pro-
posed Model (Analysis II); (c) Conventional Model (Analysis III)

ment with those of Analysis I, indicating the validity of the proposed model.
In Analysis III, the shear forces in the columns are underestimated. The
reason is that the assumed straight-line axis for the haunched beam cannot
represent the built-in arch that causes the development of axial thrust in the
beam. Underestimating the shear forces in columns resulted in underesti-
mating the corner moment; consequently, the positive span moment is over-
estimated.
It should be noted that the model used in Analysis III is not the proper
one. It is used for comparison purposes and to point out a common mistake
in engineering practice. In fact, the corner moment calculated in this model
is misleading, since the centroid of the frame corner is located at an elevation
of 3.83 m, not at 4.5 m (see Fig. 14).

SUMMARY AND CONCLUSIONS


The study presented in this paper reveals a large difference between the
behavior of prismatic and nonprismatic members. The discontinuity of the
centroidal axis in nonprismatic members causes a strong coupling between
bending moment, shear and axial forces. In the case of stepped members,
the existence of the null areas causes softening of the member stiffness. If

1586

J. Struct. Eng. 1991.117:1573-1592.


k N -m
6 8 . 0
38.7 kN^rn
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

3 7.0 kN-m
kN
2 7 . 4

SO kN

(a)

kN-m C
6.6

kN-m
68.1
38.5 kN-m

3 7 . 2 kN-m
kN
27. S

(b)

48.6 k N-m

3 8.2 kN-m

(C)

FIG. 15. Reactions and Moments for the Portal Frame Example: (a) Analysis I;
(b) Analysis II; (c) Analysis III

the member is axially restrained, arching action takes place, producing a


significant axial thrust that affects the force distribution in the structure. In
frames, beams are axially restrained to some extent, due to the stiffness of
columns. In continuous beams, axial restraints exist due to friction at sup-
ports.
The following are the main conclusions:
1587

J. Struct. Eng. 1991.117:1573-1592.


Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

0.5 1.0

DEPTH RATIO (R)

FIG. 16. Fixed End Moment (FEM)

••
— ~ :.-^, _
i .s - """i
-<
!
/ U
i/
if, Stepped
~~tfr 3 t r t . H.

0 . S 1.0 1.5

DEPTH RATIO (R)

FIG. 17. Fixed End Horizontal Reaction (/•',)

1. Fixed end moments, stiffness, and carry-over factors for nonprismatic


members available in the literature should not be used if accurate analysis is
desired. This practice involves significant errors, especially for deep haunches.
2. Coupling between bending and shear and axial forces due to arching action
must be considered.
3. Proper modeling that considers the actual centroidal axis is essential.

This study is not intended to correct the tables of nonprismatic beam con-
stants available in the literature. The purpose is to attract attention to the
error involved in the conventional methods of analyzing frames that have
nonprismatic members and to present efficient techniques for proper mod-
eling to be used with conventional frame analysis computer programs. How-
ever, the writers believe that the isoparametric finite element method offers
the best solution for determining the elastic behavior and stress resultants of
frames having nonprismatic members. With the advent of computers and

1588

J. Struct. Eng. 1991.117:1573-1592.


1 T~ T" 1
PCA
,I _ _,—
FEl
' FE2
czJi
S
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

^i * * * * • _ ^ _ _ « i

j ^ * ; i ^ " ^

DEPTH RATIO ( R )

FIG. 18. Stiffness Coefficient K (Stepped Members)

jie
*--^C!
r^rr
'^v

'"'V,
^
"'X
o \.
>* 0.00
"'V.
V
FEl
FE2 -—
.
DEPTH RATIO (R)

FIG. 19. Carryover Factor C (Stepped Members)

software, finite element analysis can become as practical as conventional


frame analysis.

APPENDIX I. PLOTS

Figs. 16-23 show plots of the constants listed in Tables 1-3 for stepped
and haunched members.
Fig. 16 shows a comparison of the FEMs obtained from FEl and those
from PCA for the different types of nonprismatic members. The effect of
arching action in reducing the FEMs can be observed.
Fig. 17 shows the variation of Fx with R, reflecting the development of
the built-in arch.
Figs. 18, 20, and 22 show the variation of A" with R for the different types
of members. The influence of arching action in increasing A"-factors can be
observed in FEl curves. In FE2 curves, the reduction in ^-factors due to
1589

J. Struct. Eng. 1991.117:1573-1592.


" n i • i "' 1
PCA 1 , b..
FEl >
1 ^
FE2 —

./
s
'
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

^^; - • ^n

DEPTH RATIO (R)

FIG. 20. Stiffness Coefficient K (Straight Haunches)

=r=r= r = ? = =rtr=
-——
s—=•:

" • — . . .

'S
*'\.
\.
N
• FEl s
FE2 \
.
0.5 1.0

DEPTH RATIO (R)

FIG. 21. Carryover Factor C (Straight Haunches)

I I I I I I
PCA H

20 . FEl b- \ 2L-f
RH

FE2 + •* - \ 0 .5L

^
S*
''

tf^—•
M

DEPTH RATIO (R)

FIG. 22. Stiffness Coefficient K (Parabolic Haunches)

1590

J. Struct. Eng. 1991.117:1573-1592.


Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

DEPTH RATIO (R)

FIG. 23. Carryover Factor C (Parabolic Haunches)

shear deformation can be seen. The larger reduction in the case of the stepped
member shows the effect of null areas.
Figs. 19, 2 1 , and 23 show the variation of C with R for the different
members. The good agreement observed between PCA and FE2 values in-
dicates that C-factors are not sensitive to the geometry of the centroidal axis
in the absence of axial displacement restraints. FEl values, however, exhibit
large discrepancies due to the moment of the eccentric axial force developed
by arching action [see Fig. 8(b and c)]. For large R, this moment exceeds
the flexural moment, producing negative C-factors.

APPENDIX II. REFERENCES

Balkaya, C. (1987). "Finite element investigation of beams of variable cross sec-


tion," thesis presented to Middle East Technical University, Ankara, Turkey, in
partial fulfillment of the requirements for the degree of Master of Science.
Cross, H., and Morgan, N. (1958). Continuous frames of reinforced concrete. Four-
teenth Ed., John Wiley and Sons, Inc., New York, N.Y., 126-155.
Fertis, D. G., and Keene, M. E. (1990). "Elastic and inelastic analysis of nonpris-
matic members." J. Struct. Engrg., ASCE, 116(2), 475-489.
Funk, R., and Wang, K-T. (1988). "Stiffness of nonprismatic member." J. Struct.
Engrg., ASCE, 114(2), 489-494.
Mumuni, I. (1983). "A finite element model for the analysis and optimal design of
beams and plates with variable flexural rigidity," thesis presented to Vanderbilt
University, at Nashville, Tenn., in partial fulfillment of the requirements for the
degree of Doctor of Philosophy, 23-38.
Portland Cement Association. (1958). "Handbook of frame constants." Beam factors
and moment coefficients for members of variable section. Portland Cement As-
sociation, Chicago, 111.
Resende and Doyle. (1981). "NONPRI—An effective nonprismatic three dimen-
sional beam finite element." Comput. Struct., 14(1-2), 71-77.
Timoshenko, S. (1953). Strength of materials. McGraw-Hill, New York, N.Y., 60-
63.
Wilson, E., and Habibullah, A. (1984). SAP80 structural analysis programs. Com-
puters and Structures Inc., Berkeley, Calif.

1591

J. Struct. Eng. 1991.117:1573-1592.


APPENDIX III. NOTATION

The following symbols are used in this paper:

A = step or haunch length ratio defined in Fig. 6(a and b);


Ae = effective haunch length ratio defined in Figs. 9 and 10;
C = carryover factor;
Downloaded from ascelibrary.org by University of Exeter on 07/15/15. Copyright ASCE. For personal use only; all rights reserved.

E = modulus of elasticity;
e = eccentricity of centroidal axis defined in Fig. 2;
Fx = horizontal reaction;
Fy = vertical reaction;
FEA = finite element analysis;
FEM = fixed end moment;
FE1 = finite element analysis using model 1, Fig. 6(a);
FE2 = finite element analysis using model 2, Fig. 6(b);
H = smaller depth of nonprismatic member;
/ = moment of inertia of cross section;
K = bending stiffness of member;
L = span of member;
Ma = bending moment defined in Fig. 8(c);
P = axial force defined in Figs. 2(c and d) and 8(c); and
R = step or haunch depth ratio defined in Fig. 6(a,b).

1592

J. Struct. Eng. 1991.117:1573-1592.

You might also like