Proc. lnstn Civ.

Adaptive space
frameanalysis Engrs Structs &
Bldgs, 1993.99,
Part 11: a distributed plasticity approach Aug.317-326

B. A . Izzuddin, MSC,PhD and A . S. Elnashai, MSC,PhD Structural Board

Paper 10151

Written discussion
closes 15 October 1993
m In the companion paper (Part I: a plastic Symbols
hinge approach), adaptive elasto-plastic A matrix of areas of monitoring points
analysis of space frames was discussed in d matrix of distances of monitoring points to the
the context of a plastic hinge approach. centroidal axis
e matrix of material strains
The present Paper extends the concept of ,E matrix of material tangent moduli
automatic mesh refinement to deal with f elementbasiclocalforces
accurate elasto-plastic large displacement <M,,, M , , , M2,, M2=, F,MT)
analysis accounting for thespread of yield $ quarticelementlocalforces
within structural members. Thisis <M,,, M,*, M2,, Mzz, F, MT. T,,T.)
achieved through the use of the quartic f matrlx of generalized stresses for cubic element
formulation in the elastic range, and per- G elastic shear modulus
forming automatic subdivision ofquartic J StVenant'storsionconstant
elements into elasto-plastic cubic ele- .k element local tangent stiffness matrix
ments which account for the spread of L elementlengthbeforedeformation
plasticity across the section depth and M, section moment in the local y direction
along the element length. The main
M. section moment in the local z direction
S matrix of material stresses for cubic element
advantage of this approach is the ability .T matrix for transformation of generalized
to startfull elasto-plastic large displace- stresses to local forces
ment analysis using only oneelement per p elementbasiclocaldisplacements
member, with automatic mesh refinement <a,,, a,*, e,,.e,.,~~ a,>
performed only where and when necessary ,U matrix of generalized strains of cubic element
within the structureand during the u(x) centroidal axial displacement
analysis. The Paper presents a description u(x) centroidal displacement in thelocal y direction
of the elasto-plastic cubic formulation, uyx) imperfection shape inlocal y direction for
discusses the concept ofautomatic mesh quartic element
w(x) centroidal displacement in the local z direction B. A. Izzuddin,
refinement in the context of elasto-plastic Lecturer in
wyx) imperfection shape inlocal y direction for
analysis, and provides verification exam- quartic element Computing,
ples, using the computer program ADAPTIC, X referenceabscissaalongtheelementchord Department of
to demonstrate the accuracy and extreme y local y coordinates of monitoringpoints Civil Engineering,
effkiency of the proposed non-linear i local z coordinates of monitoring points Imperial College
analysis procedure. dx) element twist of Science,
E material strain Technology and
c. centroidalaxial strain Medicine, London
Notation C rate of twist
Generic symbolsof matrices and vectorsare given in X abscissae of Gausssections
bold, with left-sidesubscripts or superscripts. This K, curvature in the localy direction
rule also applies to three-dimensional matrices. K. curvature in the local z direction
Subscripts and superscripts to the right side of the o material stress
generic symbol indicate the term of the vectoror Introduction
matrix under consideration. Two main approaches for the large displace-
ment elasto-plastic analysisof framed struc-
Operators tures were identified in the companion paper,
c right-sidesuperscript,denotescurrentvalues namely the plastic hinge and the distributed
during an incremental step
o right-sidesuperscript,denotedinitialvalues plasticity approaches.' While the computa-
during an incrementalstep tional advantageof plastic hinge analysis is
significant, such an approach is suitable only A. S. Elnashai,
T right-sidesuperscript,transposesign Reader in
6 incrementaloperatorforvariables,vectorsand for preliminary approximate analysis,as it is
unable to deal with the spreadof plasticity. Earthquake
matrices Engineering.
a partial differentiation Moreover, the modelling of realistic stress- Department of
c summationoverrangevariable (i) strain lawsin the contextof the plastic hinge Civil Engineering,
I I encloses terms of a matrix approach is sufficiently complex, which leads Imperial College
< ) encloses termsof a row vector
-
often to the simplifying assumption of elastic- of Science,
la I magnitude of vector a perfectly plastic material.The companion Technology and
lai! absolute value of term a , paper' discussed plastic hinge analysisin the Medicine, London
317
Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
 IZZUDDINAND
ELNASHAI

large displacement domain, and outlined in that Kinematics

context a procedure for automatic mesh refine- 5. Six local degrees of freedom are
ment deemed essential for modelling elasto- employed for the3D cubic formulation, as
plastic beam-column buckling. shown in Fig. 1
2. The present Paper focuses on elasto-
plastic analysis of framed structures, account-
ing for the spreadof plasticity across the
section depth and along the member length.
cf =
= (ely,elzt e,,, 4W T
M , , M z y MZz F, m ~ > ~
7 t p
1 (1)

Moreover, primary consideration is given to the No variables for imperfections are includedin
attainment of accurate results with aminimum this formulation, as it is intended to represent
of computational effort. The two main factors short lengths of members, and can hence be
involved are the significant computational assumed straight. The movement of a point
demand posed by finite-element formulations along the element reference axis can
be
accounting for the spreadof plasticity, and the described by four displacement values (a, U, v,
W ) , a s shown in Fig. 2, which are defined by
need to use a numberof such elements within
an elasto-plastic member for accuratemodel- interpolation functions given by

1
ling. Within a traditional approach, elasto-
plastic analysis of frames requires the useof a
fine mesh of elements for all membersof the
structure from the start of analysis, since the
locations of plasticity are not known a priori.
The thesisof this Paper is that the use of auto-
matic mesh refinement in the contextof elasto-
plastic analysis not only provides significant
computational savings, but also maintains the
original accuracy of an initially refined mesh.
3. The Paper proceeds with a description of
a fibre-type elasto-plastic cubic formulation,
which accounts for the spread of plasticity
within the element, and models the effect of
large displacements in 2D and 3D space. The No interpolation function is needed for the axial
process of automatic mesh refinement is then displacement U ,since the constant centroidal
discussed, and its computational and modelling axial strain criterion is used, as discussed here-
advantages are highlighted. Finally,a number after.
of verification examples are presented, and
comparisons are made with other approaches toGeneralized strains
demonstrate the computational savings and 6. In the derivation of the cubic formula-
accuracy of the proposed adaptive non-linear tion, the following assumptions are made.
analysis method.
Warping strains due to non-uniform torsion
are negligible.
Elasto-plastic cubic formulation
4. As for the plastic hinge quartic formula-
Plane sections remain plane after bending
tion,’ the elasto-plastic cubic formulation is deformation.
derived in a convected (Eulerian) system, where Shear strains due to flexure are negligible.
the element local displacements are referred to The section centroid and shear centre are
coincident, and theirloci represent the
the element chord in its deflected state. The
effect of large displacements is thus included in element reference axis.
a similar manner,a s discussed by Izzuddin and This allows the strain state within a cross-
Elnashai.’ The cubic formulation is capableof section tobe determined solely by a setof four
modelling material non-linearity effects generalized strains: centroidal axial strain; rate
Fig. l . Local through a detailed considerationof the section of twist; and curvature strains about the two
freedoms of cubic shape and the material stress-strain law,as principal axes.
formulation discussed hereafter. 7. The derivation of the centroidal axial
strain includes the effect of bowing on stretch-
ing, and hence the cubic formulation is capable
of modelling the beam-column effect. The cen-
troidal axial strain is assumed constant along
the element length; therefore, an interpolation
function for the axial displacement U is not
required. Although in the presence of material
plasticity, this condition does not necessarily

- 318
L L
imply a constant axial force along the element
length, the variation of the axial force is insig-

Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
 ADAPTIVESPACEFRAME
ANALYSIS:PART I1

nificant if the element is used to model short

lengths of elasto-plastic members. The remain-
ing three generalized strains are obtained from
the assumption of small local deformation, a s
given below

where, E,, [, K~ and K , represent the centroidal

axial strain, the rateof twist and the curvatures Fig. 2. Movement of
about the two principal axes respectively. The Before deformation a reference point
combination of equations (2) and (3) leads to the Afterdeformation along element length
relationship between the generalized strains
and the element freedoms

Fig. 3. Location of
Gauss
integration two
Gauss
integration
sections

8. As the relationship between the gener- Monitoring area: A,

alized stresses and strains cannot be estab- yt--+ /
lished explicitly in the presence of material [0 Monitoringpoint m- I
plasticity, the integration of the virtual work
equation to obtain the element forces is per-
formed numerically. Only two Gauss integra-
tion sections are employed along the length, (a)
again because the element can represent only Monitoring area: A,
short lengthsof members. The positionof the
Gauss sections, shown in Fig.3, is chosen such
that exact integrationof the virtual work equa- Ym 0 Monitoring
point m
tion is obtained for the elastic element

Fig. 4 . Distribution
of monitoring points
for different
1. Monitoring
point m I CYOSS-Sections:
(a)rectangular solid
The generalized strains at both Gauss sections section; (b)circular
are hence represented by a matrix
p (c section; hollow
(c) I-section

neglected, only direct strainsneed to be estab-

lished at the monitoring points
which can be determined explicitly from equa-
tions (4) and (5). e-. g = izl
4

d=,i s u i . #

where e-, Iis the direct strainof monitoring

Generalized stresses point m at Gausssection g
9. Each Gauss section is divided into a

l
4 . 1 =1
number of areas across which strains and
stresses are monitored,a s shown in Fig. 4 for dm, 2 = -Ym
certain cross-sectional configurations.If the dm. 3 -=m

effect of shear strains on material plasticity is dm,* =0

319
Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
 IZZUDDIN AND
ELNASHAI

and 9uis the generalized strain matrixdefined and

in equation (6).
10. A uniaxial stress-strain relationship is
employed to establish the direct stresses at the
monitoring points for j = 1, 3, and h = 1, 3 (15)
am. g = de,. g) (9) sk4.4.g = GJ

in which 0 can represent any stress-strain law, All other terms are zero
a s demonstrated by Elnashai and I z z ~ d d i n . ~ where tE is defined in equation (10).
11. For the element tangent stiffness calcu-
lation needed for the iterative solutionpro- Global analysis
cedure, a material tangent modulus is required 15. Global structural analysis, including
geometric non-linearityeffects, is performed for
the cubic formulation ina manner identical to
that for the plastic hinge quartic formulation.'
In the calculation of direct stresses at themoni-
12. The bending and axial generalized toring points according to equation (g), strains
stresses at a Gauss section are determined from must be considered incrementally from the last
the material direct stresses,while the relation equilibrium state. This requires storageof
between the torsion generalized strain and material variables (stresses, strains,etc.) at the
stress is based on the elastic rigidity constant start of the load increment and for the current
n I iteration, with the variablesof the former
updated only after global equilibriumis
achieved.

where A, is the areaof monitoring point m, Adaptive mesh refinement

dm, is defined in equation (8), and n is the
number of monitoring points at a Gauss
section.
Local forces and stiffness
13. After the generalized stresses are
obtained at the two Gauss sections, the
local forces can be established from numerical
integration of the virtual work equation4
4 2
cfi = C C c T i , j . g sfj. g
j=l g = l
where cT is a 6 X 4 X 2 matrix representing
weighted first derivatives of generalized strains the use of a number of cubic elements, since a
with respect to local displacements, and which
is given in Appendix 1.
14. The local tangent stiffness ck is
obtained through differentiation of equation
(12), and can be expressed as4
/Q\ 4 2 4 required for all members of the structure, since
cTi, j. U s k , h. g cTk, h. g
in which ,k and ,k are determined according to impractical for most realistic structures.
the following
16. The elasto-plastic cubic formulation,
discussed in the previous section, is capableof
modelling the spreadof plasticity across the
section and along the element, by virtue of the
use of monitoring areasfor stresses and strains
at the two Gauss sections. The computational
demand of the cubic formulation is significant,
element
especially for complicated stress-strain
relationships, as the stress-strain response has
to be monitored for an often large number of
monitoring areas to achieve accurate represen-
(12) tation of theplasticityspreadwithinthe .
section. Moreover, the accurate modelling of the
spread of plasticity along themember requires
cubic shape function for the transverse dis-
placements and the constant centroidal axial
strain criterion donot qualify the formulation
to represent awhole elasto-plastic member. In
the context of a traditional modelling approach,
a fine mesh of elasto-plastic elements is
the locations of plasticity within the structure
are not known before analysis. This leads to a
(13) considerable computational demand, rendering
the use of the distributed plasticity approach
17. The concept of automatic mesh refine-
ment, discussed in the companion paper' for

1
2FL
fkl. 1 = fk2.2 = fk3.3 = fk4.4 = 15 the plastic hinge approach, is extended herein
to deal with the above inefficiencies pertinent
to traditional distributed plasticitymodelling.
fkl, 2 = fk2.1 = fk3.4 = fk4.3 = - - The proposed procedure relies on the accuracy
30
of the quartic formulation4in representing

320
are
F is the axial force, and all other
terms zero J imperfect elastic beam-columns using onlyone
element per member, and the abilityof the

Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
 ADAPTIVE SPACE FRAME
ANALYSIS: PART I1

cubic formulation tomodel the spread of plasti- X Sections

for
yield
check
city within elasto-plastic members using a H Potentialelasto-plasticcubicelements
number of elements. At the start of the HOriginalelasticquarticelement
analysis, a member ismodelled by one elastic
quartic element. In the courseof load
incrementation, yield is checked at the extreme 1 I
I
X 12 Fig. 5. Example
L12 L12 configuration
for
fibres of selected sections along the element \I
F -

length. If yield is detected, elasto-plastic cubic
elements are inserted onlyin the zones of plas-
ticity, while the restof the element is remod-
elled with elastic quartic elementsbefore the
analysis is continued.
18. The main advantage of the automatic
mesh refinement process is the significant com-
putational savings, since the computer-
intensive elasto-plastic cubic elements are
introduced only where required within the
structure, and when necessary during analysis.
Moreover, the modelling advantages are con-
siderable, a s the meshing of the structure at the
start of analysis can be performed using only
one element per member.The main components
of the automatic mesh refinement procedure are
described hereafter.
Yield check
19. The check for yield along the length of
the quartic element is performedfor each load
step atselected sections. These sections cor-
respond to the Gauss integration positions of
the cubic elements tobe inserted if yield is
detected, with the configurationof the refined
mesh specified before analysis,as shown in
Fig. 5.
20. The biaxial moments at a section are
calculated in the same manner a s for the plastic
hinge approach, discussed in the companion
paper.' If the combination of the axial force and
bending moments at a section resultsin yield-
ing at the extreme fibres, then a cubic element
is inserted at the corresponding location, while
the elastic parts are kept as quartic elements. If
none of the sections has yielded, thenmesh
refinement is not performed.
21. The current load step must be reapplied
if any one of the structure quartic elements has
been subjected to mesh refinement; otherwise,
the analysis proceeds to the nextload step.
22. Elastic quartic elements which are the
result of an earlier mesh refinement process are
checked for yield at sections corresponding to
the remaining elastic partsof the original
refined mesh
plastic cubic elements andnew elastic quartic
elements, as shownin Fig. 6.
24. As with the plastic hinge approach,' the
global displacements of the new nodes are
established at the last equilibrium configu-
ration from the deflected shapeof the original
element and the global displacementsof its end
nodes.
25. For the new elements, variables corres-
ponding to the initial and last equilibrium con-
figurations must be determined. For the cubic
elements, these include initial direction cosines,
orientation of the principal axes at both ends,
local displacements, stresses and strains at
monitoring points, and local forces; while for
the quartic elements, they include initial direc-
tion cosines, initial imperfections, orientation of
the principal axes at both ends, local displace-
ments, local forces, and mesh refinement con-
figuration. Similar to the plastic hinge
approach, allowance must be made for the non-
linear variation of the axial displacement along
the original element length, which is due to the
effect of bowing on axial stretching.
Verification examples
26. The methodology for accurate elasto-
plastic large displacement analysisof space
frames, discussed previously, hasbeen imple-
mented in ADAPTIC,' in coexistence with the
plastic hinge approach. Three examples are pre-
sented herein to demonstrate the accuracyof
the proposed distributed plasticity approach
and
the
significant
computational
savings Fig. 6.Mesh
achievedusingtheconcept of automatic mesh refinement of quartic
refinement in the
context of elasto-plastic element
Yieldino at extreme fibres

W
Elastic quartic elements
element. After a mesh refinement process,
therefore, information must be passed to iden-
tify for the new elastic quartic elements the
sections at which yielding is tobe checked as
the analysis proceeds.
e
..-
Mesh refinement
23. Within the distributed plasticity
approach, mesh refinement involves thecre.
ation of a number of new nodes, new elasto-
r r Before

Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
mesh refinement After

321
mes
 IZZUDDINAND
ELNASHAI

2790 mm analysis. All reported CPU times are for

t-7 ADAPTIC v2.1.2, running on aSilicon Graphics

workstation with 24 Mb of physical memory
and rated at30 mips, 4.2 mflops and 26 Spec-
marks.

K-frame
Transverse beam: 27. In this example, comparison is made
Q 219 X 4.37 mm2 between the prediction of ADAPTIC, using the
E = 210 X 103 N/mrn2 plastic hinge and distributed plasticity
uv = 414 N/mm2
approaches, and the analytical and experimen-
talresultsreported bySoreide et onthe
Diagonal members: K-frame set-up shownin Fig. 7. Although
Q 101.7 X 3.30 mm2
E = 210 X 103N/mm2
Soreide et al. did not include any information
U,. = 335 N/mm2 regarding the level of imperfections, it was
found that ADAPTIC provides best comparison
with the experimental ultimate capacity if para-
bolic imperfections of (L/lOOO) are assumed in
the diagonal members.
Fig. 7. Geometric configuration of K-frame 28. Soreide et al. used the computer
program USFOS, which employs a beam-column
formulation with sinusoidalor hyperbolic
shape functions, and adopts the plastic hinge
approach to model the effect of yielding. The
comparison with experimental results in Fig. 8
shows that the predictionof USFOS overesti-
mates the ultimate capacityby 17%,while
underestimating the response in the post-
buckling range.
29. With ADAPTIC, the plastic hinge
_1
Experiment
approach with automaticmesh refinement is
200 - ._.__
USFOS (Soreide et aL6) used first.' The analysis is started using four
--- Plastic hinge approach plastic hinge quartic elements, whereupon
- Distributed plasticity approach (0.1% hardening) detection of a plastic hinge within eitherof the
diagonal elements, automatic subdivision into
O , . . . ' l . . . ' l . . ' . I ' . . ' I . . . ' ~ . . . . I two elements is performed,as shownin Fig. 9.
0 50 100 150 300
200 250
With this approach, the ultimate capacity is
Displacement: mm
overestimated by 9%, compared with the 17%
Fig. 8. Static response of K-frame overestimate predicted by USFOS (Fig. 8). The
improved accuracy may be attributed to the
occurrence of plastic hinges within the diagonal
members away from mid-length, which
Node
N: ( Plastic
hinge 1 reinforces the argument presentedin the
QP: Quartic plastic-hinge element companion paper regarding the need for
automatic mesh refinement to account for
N3 N4 N5 such a possibility.
30. The results of ADAPTIC are also obtained
using the distributed plasticity approach,
where the analysis is started with four quartic
elastic elements, and automatic mesh refine-
ment into elasto-plastic cubic elements is per-
formed upon detection of yield. By the end of
the analysis, 32 cubic elements, employing the
bilinear kinematic material model with astrain-
hardening parameter of 0.1%, are inserted in
the yielding zones, while the restof the frame is
still modelled by elastic quartic elements,a s
shown in Fig. 10. This approach provides more
favourable comparison with the experimental
4 quarlic plastic-hinge elements 6 quartic plastic-hinge elements results in terms of ultimate capacity (0.2%
overestimate) and post-buckling response pre-

- 322
lnitial modelling

Fig. 9. Modelling of K-frame with plastic hinge approach

Final modelling
diction, as demonstrated in Fig. 8.
31. This example demonstrates the high

Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
 ADAPTIVESPACE FRAME
ANALYSIS: PART I1

levels of accuracy that canbe achieved with the

distributed plasticity approach, although, it
may be argued, that the plastic hinge approach
is sufficiently accurate. The issue of accuracy N3 N4 N5
against efficiency of the two approaches canbe
highlighted by considering theCPU times,
where the distributed plasticity approach
requires 42.3 S while the plastic hinge approach
requires only 8.2 S . However, no general conclu-
sions can be made regarding the relative merits
of the two approaches, since the problem con-
sidered in this example is reasonably small,
and the material stress-strain relationship is
almost elastic-plastic without strain hardening.
For realistic problems, the efficiency of the
plastic hinge approach may not be so pro-
nounced, as shown in the last example, in
which case, accuracy considerations playan
important role in choosing the appropriate 4 quartic elastic elements 6 quartic elastic elements+
32 elasto-plastic cubic elements
approach.
modelling Final modelling Initial
One-bay steel jacket
32. The steel jacket structure, shown in Fig. Fig. 10. Modelling of K-frame with distributed plasticity approach
11, is subjected to a transient seismic signal
having a duration of 4 S (see Fig. 12). The aim
of this example is to demonstrate the consider-
able efficiency of the proposed automatic mesh
refinement process a t no loss in accuracy, and,
therefore, the analysis is performed with Legs: 4~250 X 10 mm2
ADAPTIC, using two distributed plasticity Other members: Q 150 X 5 mm2
approaches. The first approach uses an initially
E. = 210 X 103N/mm2
refined mesh of 400 elasto-plastic cubic ele- U, = 300 N/mm2
ments, with eight elements modelling each of
the braces, and six elements used for the rest of
the members; whereas the second approach
starts with only52 elastic quartic elements and m , = 125 kg
employs the automatic mesh refinement process m2 = 500 kg
mz = 15000 kg
to model the elasto-plastic behaviour, poten-
tially subdividing the original mesh into the
same 400 cubic elements if all the structural
members undergo plastic deformation. (supports: NO rotational restraint)
33. A comparison of the results from the
two approaches in Fig. 13 demonstrates excel-
lent agreement, and verifies that there is no Ground motion
loss of accuracy resultingfrom the use of auto- Fig. l l . Geometric configurationof one-bay steel jacket
matic mesh refinement. However, automatic
mesh refinement provides significant computa- 1
tional savings owing to the restriction of plasti-
city to the jacket braces (see Fig. 14); hence
only 104 elasto-plastic elements are required 0.5
and at a later stagein the analysis.
34. A detailed consideration of CPU con- 0,

sumption in Fig. 15 shows that an initially g

.- W

refined mesh requires 2 hrs 11 min 6 S, while -g o

with automatic mesh refinement only 27 min 8
45 S is needed. A great part of the savings with
2
the latter approach is achievedin the elastic -0.5
range, before signal time of 2 S , where the
structure ismodelled with only 52 elastic
quartic elements a s opposed to 400 elasto- -1
plastic cubic elements.However, more 4 savings 0 3 1 2
are obtained after the occurrenceof plasticity, Time: S

as only partof the structure undergoes plasti- Fig. 12. Transient signal applied to steel jacket ~

323
Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
 IZZUDDINAND

0.06
- Automatic
refinement + Initially
refined
city and 44 elastic elements with104 elasto-
plastic elements provide accurate
representation.
35. In general, using the automatic mesh
refinement process can never be less efficient
than using an initiallyrefined mesh, since at
worst all the structuremembers are refined
automatically from the first time-step and an
initially refined situation is achieved.However,
the relative efficiency of the two approaches
may vary depending on the time-history of the
spread of plasticity within the structure. This
is demonstrated in Fig. 16, where automatic
-0.06 I I I mesh refinement is about 20 times more effi-
0 1 2 3 4 cient than an initially refined mesh in the
Time: S
elastic range, with that ratio dropping to about
Fig. 13. Response of steel jacket to transient signal 5, still a considerable saving, after plasticity is
achieved.

3 0 jacket under earthquake loading

36. This example is a repeat of the one pre-
sented in 52-53 of the companion paper,'
except that the distributed plasticity approach
is used insteadof the plastic hinge approach.
[- Plasticityzones:cubicelements
(same on opposite side)
The purpose of this example is to verify the
accuracy of both approaches in the non-linear
analysis of realistic structures, andto provide a
comparison of accuracy and efficiency in that
context.
37. The comparison of results in Fig. l?
demonstrates only minor differences in the pre-
Initial dictions of the two approaches, which are
1
I

Final I 44 I 104 1
I
attributed mainly to the inability of the plastic
hinge approach to accountfor the spread of
plasticity within the cross-section and along
Modefling with automatic mesh refinement the member. Such difference, however, could be
much more significant if strain-hardening
effects are included, or a realistic cyclic model
for steel ise m p l ~ y e dIt
. ~ is also worth noting
that, while the direction of buckling of the
braces is different for the two approaches
(compare Fig.18 herein with Fig.22 in the com-
Fig. 1 4 . Deflected shape of steel jacket panion paper'), the overall responseof the
structure is not significantly affected. This
asserts the bifurcation natureof brace buckling
for the problem under consideration.
38. In terms of efficiency, the distributed
(').e* plasticity approach requires aCPU time of
(2)44 elastic elements ( 7 *' 2 hrs 34 min 40 S, whereas the plastichinge
104 elasto-plastic elements
(3) 400 elasto-plastic elements approach requires 1 hr 18 min 3 S. The compari-
son of CPU time ratios, a ratioof 2 for this
4 example compared with a ratioof 5 for the first

.' .*
example, shows that the relativeefficiency of
^^^^ refined Initially r) the two approaches is problem dependent,
a \ although the plastichinge approach is gener-
0
1 Automatic
refinement 3 e ally more efficient. It also suggests that the dis-
tributed plasticity approach is not necessarily
prohibitively expensive,as may be expected,
and should hence always be used in situations
where accuracy is important and the assump-
tions of the plastic hinge approach impose
Time: S severe simplificationsover the problem in
Fig. 15. CPU time f o r analysis of steel jacket hand. However, it must be noted that the
324
Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
 ADAPTIVESPACE FRAME
ANALYSIS: PART I1

achieved efficiency of the distributed plasticity 25

is attributed mainly to the automatic mesh (Initially refined): (Automatic refinement) 1
refinement process proposed in this Paper, as
shown in the previous example.

Conclusions
39. This Paper presented adaptive space
frame analysis in the context of a distributed
plasticity approach. The proposed approach is
based on modelling the structure using one
elastic quartic elementper member at the start
of analysis, and performing a process of auto-
matic mesh refinement in which elasto-plastic
cubic elements are insertedwhere and when 0
needed, within the structure and during 0 1 2 3 4
analysis respectively. The processis therefore Time: S
temporally and spatially adaptive. It wasrecog. Fig. 16. CPU savings with automatic mesh refinement
nized that the accuracyof the proposed method.
ology benefits largely from the abilityof the
quartic formulation tomodel the elastic geo-
metrically non-linear behaviourof an imperfect
Plastic hinge approach
member using only one element.
40. The Paper presented the formulation of
the elasto-plastic cubic element,which employs
Gaussian quadrature at two sections across
which stresses and strains are monitored for a
number of points. The improved accuracyof
the cubic formulationin comparison with
plastic hinge formulations was discussed, and
was attributed to the abilityof this formulation
to model the spread of plasticity a s well as the
effect of realistic material stress-strain laws,
albeit at the expenseof employing a number of
elements per member.
41. The process of automatic mesh refine-
ment within elasto-plastic analysis was out-
lined, pointing out the departurefrom
conventional finite-element modelling. Inthis
context, it was shown that the inefficiencies
Fig. 17. Response of three-dimensional jacket to earthquake signal
associated with the useof a number of elasto-
plastic elements tomodel each and every
member of the structure, since the locationsof
plasticity are unknown a priori, canbe com-
pletely avoided. Also, the benefitsthat auto-
matic mesh refinement avails in terms of ease
of structural modelling, as amesh of only one
element per member is required, were men-
Scale = 10

-
tioned.
42. Finally, the Paper presented a number
of verification examples, where comparisons
Plasticity
were made with the resultsfrom other work
and the plastic hinge approach. The accuracy of
the distributed plasticity approach was shown
to be superior to thatof the plastic hinge
approach, although, in general, it was found
that the latter approach is more efficient.
However, it was observed that the relative effi-
ciency of the two approaches depends on the
particular problem under consideration. It was
therefore suggested that the distributed plasti-
city approach should be employed for cases
where accuracy is important, since efficiency Fig. 18. Deflected shape of three-dimensional jacket: (a)=t 1 . 8 3 S;
differences are generally not very significant. ( b ) t = 5.0S

Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.
 IZZUDDIN AND
ELNASHAI

The importanceof automatic mesh refinement

in that context was demonstrated one
in of the
examples, where an improvement factor of 5 to
20 was achieved by the proposed methodology
in comparison with conventional finite-element
modelling.

Acknowledgements
43. The Authors thank Professor Patrick J.
Dowling for his continuous technical and moral
support of this work. The assistance provided
by the Edmund Davis fund of the Universityof
London is also gratefully acknowledged.

Appendix 1. Matrix =T
44. The matrix required in equation (12) is
given by
References
1. IZZUDDIN B. A. and ELNASHAI A.S. Adaptive
space frame analysis.Part I: a plastic hinge
approach. h o c . Instn Civ. Engrs Structs & Bldgs,
1993,99, Aug., 303-316 (companion paper).
2. IZZUDDIN B. A. and ELNASHNA. S. Eulerian for-
mulation for large displacement analysis of space
frames. J. Engng Mech.Am. Soc. Civ. Engrs, 1993,
119,NO. 3,549-569.
3. ELNASHAI A. S. and IZZUDDINB. A. Modelling of
material nonlinearities in steelstructures sub-
jected to transient dynamic loading.Earthq.
Engng Struct. Dyn., 1993,22, tobe published.
4. IZZUDDIN B. A. Nonlinear dynamic analysis of
framed structures. Thesis submitted for the
degree of Doctor of Philosophy in the University
of London, Departmentof Civil Engineering,
Imperial College, London, 1991.
5. IZZUDDINB. A. and ELNASHN A. S. ADAPTIC: a
program for the adaptive dynamic analysis of
space frames. Imperial College, London, 1989,
Report No. ESEE-89/7.
6. SOREIDE T.H. et al. Collapse analysisof framed
offshore structures. 18th Ann. Offshore Tech-
nology Conf., Houston, Texas, 1986,95-102.

326
Downloaded by [ University of Melbourne] on [24/09/16]. Copyright © ICE Publishing, all rights reserved.