EVALUATION OF NONLINEAR FRAME FINITE-ELEMENT MODELS
By Ansgar Neuenhofer! and Filip C. Filippou
Apstaact: In recent years nonlinear dynamic analysis of three-dimensional structural models is used more
and more in the assessment of existing structures in zones of high seismic risk and in the development of
‘appropriate retrofit strategies, In this framework beam finite-element models of various degrees of sophistication
fre used inthe description of the hysteretic behavior of structural components under a predominantly uniaxial
State of stain and stress. These models are commonly derived with the displacement method of analysis, but
recent studies have highlighted the benefits of frame models that are based on force interpolation functions
(exibility approach). ‘These benefits derive from the fact that models with force interpolation functions that
reproduce the variation of intemal element forces in a strict sense yield the exact solution of the governing
equations in the absence of geometric nonlinearity. While the numerical implementation of force-based models
fa first appears cumbersome, simple examples of nonlinear analysis in this paper offer conclusive proof of the
‘numerical and computational superiority of these models on account ofthe smaller number of model degrees
Of freedom for the same degree of accuracy inthe global and local response. A numerical implementation that
‘bypasses the iterative nature ofthe element state determination i recent force-based elements is also introduced,
thus further expanding the benefits of flexibility-based nonlinear frame models.
INTRODUCTION
Recent earthquakes have increased the interest in the de-
velopment of reliable analytical methods of assessing. the
slobal and local dynamic response of existing structures and,
if necessary, developing effective retrofit strategies. In three-
dimensional models of building and bridges, beam finite-ele-
‘ment models are important in describing the hysteretic behav-
jor of structural components under # predominantly uniaxial
stress state. Since the nonlinear dynamic analysis and damage
assessment of large three-dimensional structural models under
earthquake loading are computationally very costly, the pro-
jection of the three-dimensional structural member into a one-
dimensional line element brings great benefits of economy
without undue compromise of aceuracy for girders and col-
‘The most common approach in the formulation of nonlinear
frame elements to date isthe displacement or stiffness method
that is based on appropriate interpolation functions for the
transverse and axial displacements of the member. Conven-
tional frame elements are based on cubie Hermitian polyno-
‘mials forthe transverse displacement fields and linear Lagran-
ian shape functions for the axial displacement, which
represent the exact solution fora linear elastic, prismatic beam,
‘This formulation results in linear curvature and constant axial
sirain along the element, but has well-known shortcomings in
nonlinear analysis (Weiler 1990; Neuenhofer 1993). These
imitations can be largely overcome with higher-order dis-
placement interpolation functions in connection with internal
element nodes (Weiler 1990; Nevenhofer 1993).
‘The formulation of flexibility-based elements, on the other
hhand, is based on interpolation functions for the internal for-
ces. For geometrically linear structures itis staightforward to
select polynomials that satisfy the element equilibrium in a
Strict sense, such as constant axial force and linearly varying
bending moments in absence of element loads (Zeris and Ma-
hin 1988, 1991). These interpolation functions represent the
"Wining Res Eng, Dept of Cv and Env Engg, Univ of Cale
foria Bercy, CA 94720
aus. Prof Dept of Civ, and Envir Engrg, Uni of Caloris
Berkley. Ca
Tote Awicste Editor: Kevin 2, Truman. Dcusion open unl De
cember 1, 1957, To extend the closing date one mouth, 2 wien reaes
trast be filed wih the ASCE Manages of Jour The manus for
this paper was submited for review and povsbe publication on Septet
ter £3, 1936. Tis paper i pr ofthe Journal of Structural Engineer
ing, Vo. 133, No uly 1987. OASCE, ISSN OTH 94431007
(85 Oboe 0 + 3.50 per pags. Paper No. 14157
(958 / JOURNAL OF STRUCTURAL ENGINEERING / JULY 1097
‘exact solution to the governing equations, irrespective of the
‘geometry and constitutive law of the beam element. A dis-
cretization error, as generally encountered in stffness-based
formulations, does not occur.
“The main obstacle in the widespread use of this type of
clement is the difficulty of integrating the nonlinear element
state determination in an analysis program that is based on the
direct siffness method. This obstacle is overcome in recent
proposals of flexibilty-based beam finite elements (Ciampi
land Carlesimo 1986; Taucer etal. 1991; Spacone et al. 1996a)
by a state determination procedure that iteratively determines
the clement resisting forces and stifiness matrix while strictly
satisfying element equilibrium and compatibility in each iter-
ation. This procedure is considerably more involved than for
stiffnest-based elements,
‘This paper presents a procedure that bypasses the need for
‘an iteration during the element state determination by aecept-
ing both residual displacements, which violate compatibility
atthe element ends, and unbalanced section forces, which vi-
‘late strict equilibrium along the element during each iteration
ofthe global equilibrium equations. On convergence such vi-
lations are reduced to the specified numerical tolerance. This
‘new element state determination resembles more closely the
Computational steps in a stiffness-based element state deter-
rmination and is computationally more efficient than earlier
proposals
‘The objective ofthis paper isto present the salient features
of stiffness and fexibility-based frame finite elements and then
‘compare the results for a simple example that clearly high-
ights the advantages of flexibility-based clements for nonlin-
‘ear structures, Since a pure flexibility formulation is not feas-
ible in a mul-degree-of-freedom structural mode, this paper
discusses an element state determination process for a fle
bility-based element within a nonlinear analysis program based
‘on the direct stiffness method. This paper then concludes with
8 comparison of the computational effort of the new approach
With common stiffness-based clement formulations.
MODEL ASSUMPTIONS
‘The frame finite-element models in this paper are based on
the Euler-Bernoulli beam theory for geometrically linear be-
havior. The response in torsion is assumed linearly elastic and
‘uncoupled from the axial and flexural response. Consequently,
the displacements and forees associated with torsion are omit
ted in the following discussion for simplicity
‘Under these assumptions, the governing variables are the
displacement fieldsM,(2)peq()
Cov (e).e(2)
wv
M,(x)Kyla)
i! are
FIG. 1. Dlaplacements, Internal Forces, and Deformations,
we = Le WO) 80OTT o
of the element reference axis, which correspond to deforma-
tion fields (Fig. 1)
sear fa 8H
“These represent the axial strain e(x) along the reference axis
and the curvatures «,(x) and x,(1) about two orthogonal axes
‘y and 2, respectively. The deformations are linear functions of
the displacement fields by simple differentiation according 10
(Q). The corresponding stress resultant or internal force fetds
ig.)
Dox) = [Na) Mo) Miao” ®
are the axial force N(x) and the bending moments M(x) and
‘M2). Weis assumed that a section constitutive relation
»,
and its tinearization
®
are known, where
aN aN aw
a
aM, aM, aM,
x-2 = o
is the section stiffness matrix; and n denotes time. For the sake
of simplicity, the field coordinate x was omitted inthe preced-
ing equations. The section constitutive relations in (4)~(6) can
bbe defined as section force-deformation relations or derived
from a fiber discretization of the eross section. In the later
case, the section forces D are determined from the fiber stress
distribution by integration, and the section stiffness is assem-
bled from the fiber stffnesses,
STIFFNESS (DISPLACEMENT) METHOD
In the stiffness method the displacement fields of the ele-
iment are discretized and interpolated in terms of ry gene
alized displacement degrees of freedom q such that,
0) = Nova o
where
No 0 0
Noe] 0 Naw 0 ®
0 0 Nw)
isa 3xny-matrix containing the vectors of interpolation func:
tions N., Na and N,, for the three displacement fields , ,
and v, respectively. The expression for the deformation fields
a) is then
ac) = Beg o
where the strain-displacement transformation matrix B con-
tains first and second derivatives of the displacement shape
functions according to the kinematic relationships in (2). Re=
placing Ad from the incremental version of (9) in the consti-
{utive relation (5), yields the relation
ADG) = kepada) = KUBO io)
for the force field increment AD(z). The principle of viet
displacements leads to the equilibrium condition
ies ;
and its linearization
Ka «a
in the form of a force-displacement relation, where Q = ele-
ment resisting forces and
22. fpr
represents the element stiffness matrix, while Aq and R are
the vectors of displacement increments and residual forces,
respectively; and L denotes the element length
FLEXIBILITY (FORCE) METHOD
In the Mexibility method the force fields are described by
the relation
Dix) = OQ ay
where (2) contains the force interpolation functions, which
relate the generalized nodal forces Q to the internal forces
DG), Replacing AD from the incremental version of (14) in
the inverse form of the constitutive elation (5), namely Ad(s)
ke'ADQ), yields the incremental deformation field
Ades) = RADE) = fb~)AQ as)
where f(x) = K(x) = section flexibility matrix, The principle
of virual forces leads to the compatibility condition
ae [ ved ae i)
and its linearization
Faq an
in the form of a displacement-force relation, where q = ele-
‘ment end displacements; and
ae fy
reat [ weommen ae a
is the element flexibility matrix, while AQ and F are vectors
‘of force increments and residuat displacements, respectively,
[Note that a meaningful expression for the fexibility matrix F
can only be derived for the beam element without rigid-body
‘modes.
COMPARISON OF STIFFNESS AND FLEXIBILITY-
BASED MODELS.
‘To illustrate the fundamental difference between stiffness
and flexibility-based finite-element models, a single-degree-of-
JOURNAL OF STRUCTURAL ENGINEERING / JULY 1997/969freedom, single-clement cantilever with a linearly varying
cross section and unit length shown in Fig. (a) is considered.
‘The system is subjected to an axial load only, such that the
force fields D(x) and deformation fclds d(x) of the beam re-
duce to the scalar quantities N(x) and e(x), respectively, The
stress-strain relationship is assumed to be given by
fe 0S-e, = 095
onesie oss, Sa o>
With o = of€ = 0.95]. For the sake of brevity all units are
consistent and omitted in the following discussion, For the
finite-element solution of this problem, a three-point Gauss
integration
1 [wa Done) @
with weights « = $/18, 8/18, and 5/18 at the three integration
points is used to compute the element quantities, where € =
XIL is a dimensionless element coordinate. The cross-sectional
areas atthe three integration points are selected as A = 3, 2.5,
and 2, respectively.
In the first analysis a flexibility-based formulation is used
jn the iterative solution of the system response under an im-
posed displacement of g* = 0.634 at the free end of the can-
tilever, In the second analysis a stiffness-based formulation is
used in the iterative solution of the system response under an
applied force of P = 1 at the free end of the cantilever. The
force value corresponds to the imposed displacement of g* =
(0.634, as determined from the first analysis of the system,
‘This procedute is selected for illustrative purposes only,
such that an iterative solution strategy is required in both
‘cases. A flexbilty-based analysis of the system under imposed
Toad and a stiffoess-based analysis under imposed displace-
ment would not require an iterative solution, This fact has
important ramifications in the implementation of a frame ele-
ment in a general purpose analysis program, where the element
state determination represents an analysis under imposed ele-
‘ment end displacements, This fact will be addressed later in
this paper
Table 1 shows the flexiility-based analysis for the inter
polation of the axial force with a constant function
Ne
be en
)
FIG. 2. Finito-Element Solution: (a) Sample Structure; (b)
Flexibiity-Based () Sitfness-Based
{960 / JOURNAL OF STRUCTURAL ENGINEERING / JULY 1997,
TABLE 1. Flexibility Besed Solution
TTolel«]ale]ale
olato} |e lelole
+ | $390 oso] ons | 10 | rs | 53¢7 | 5003
1 | S30 aes
3 | eae os
3 | Bio | aras| sors | ose | sano | ees | 1200
2 | 10000, we
3 foes one
3 | tat | 370] one | ores | sore | Osee | oors
3 | soo te
4 | os ous
4 | S38] sue] oom | oor | nano | 8353 | ase
ths ots
TABLE 2, silnese-Based Soliton
Tel elela|e ele] o
ol @lololelololele
[aon o966
1 [3580 [500.00 o400|osao | o400| bss | oso
1 | Zom Sess
2 | 2a es
3 | 15 | 1502000125] oss fosss| oars | o978
3 | Lie
3 |e
3 | Her |r| 0022] o019|oss2| oss2| 0389 | osm
3 | oss
2 tae
£18 | 1119 foom| 001] oss |osss £9 | 1.00
tows 2
accoring 10 (14). Te table iss the iteration counter i the
tangent section flexi f= forthe te nteraion
poins she tangent element Resy
ref rrmaer[roa~e Zen an
sccortng 0 (18), the residual displacement = 9 —
the force incrempnt A= Ag te ol foe @
AQ’, the strain €; for the three sections j, and the element node
diiacement
tee) de = f Ho a~L Soe 2
‘The expressions in (22) and (23) are derived from the principle
of viral forces by three-point Gaussian quadrature ofthe re-
sulting integrals. The strain e} in (23) is obtained from the
inverse form of the constitutive relation in (19).
Table 2 shows the corresponding stiffness-based analysis.
‘The axial displacement field
an Fq 2
is assumed to vary linearly, such that the strain
1
1
is constant along the element. Notice that the preceding wo
‘equations correspond to expressions (7) and (9) of the general
formulation, Table 2 contains the tangent section stiffness kj =