You are on page 1of 8

Space Structures 1 (1985) 33-40

Analysis of Double Layer Grids with


Material Non-linearities!IIW--
a Practical Approach

A. Hanaor
Department of Civil and Environmental Engineering, Rutgers University, PO Box 909, Piscataway,
New Jersey 08854, USA

(Received: 7 November, 1984)

Abstract: A practical approach is presented, for the analysis of space trusses


involving non-linear member behaviour, with particular attention to buckling. In its
application to double layer grids, the method takes advantage of the well-defined
yield lines formed by yielding or buckling members. The method, which is 'exact',
employs a piecewise linearized strut characteristic, which may be obtained either
experimentally or by established analytical procedures. The flexibility based
algorithm is extremely efficient and easy to implement through modifications to any
existing linear analysis program.

Notation * Modified as defined in text


o Initial or unit values
A Geometric (kinematic) matrix
a Cross-section area
B, b Force influence matrix and vector, respectively 1. Introduction
C, c Flexibility matrices and vectors, respectively
D, d Nodal displacements The high degree of redundancy in double layer grids
E Elastic modulus and certain analogies with plate behaviour, have led
e Member elongations to attempts to apply limit state approaches to their
F Member forces design (e.g. ref. 1). The formation of well-defined
I Identity matrix 'yield lines' by yielding or buckling members, in
K Stiffness matrix particular ref. 2, suggests such an approach. Limit
L Member length state analysis for computing ultimate load capacity is
P Nodal applied loads only appropriate, however, if struts behave elastic-
Q Fictitious member forces plastically. In practice, compression members often
A. Load factor display what Schmidt' has termed 'brittle-type'
buckling and Smith" 'chordal snap through' buckling
Subscripts and superscripts (Fig. 1). Furthermore, in the majority of cases, strut
buckling is the governing failure mechanism," which
c Central implies that truss ultimate load capacity is usually
F Due to redundant forces represented by load at first buckling, rather than at
I Of linear substructure full formation of yield lines.
P Due to external load Experimental work 2.5.6 has further demonstrated
r Of redundant members that due to the effects of imperfections--members'
T Transpose lack of fit, in particular-even theoretical capacity at
33
Space Structures 0266-3511/85/$03·30©Elsevier Applied Science Publishers Ltd, England, 1985.Printed in Great Britain
34 A. Hanaor

F
(a) The strut model used-i.e. the approximation
.L
made to the actual strut axial load-deflection
d
F
c T curve, which is derived either experimentally,
or analytically (e.g. refs 3,7);
(b) The truss analysis algorithm.
Schmidt':" employs an experimental strut character-
istic, approximated by an elastic-brittle-plastic
model (Fig. 2). His analysis thus employs a 'member
/ removal' technique in a similar manner to elastic-
, /

L-_-+ ,
L/. -
/
d
plastic behaviour but with a reduced level of the

F
Fig. 1. Typical 'brittle-type' strut buckling.

first buckling is often unattainable, with reductions in


load capacity of up to 40% reported. Ideal ized
It thus becomes necessary to develop theoretical -,
models capable of accounting for actual non-linear "~ Experimental
<,
strut behaviour in order to predict truss ultimate load <,
' .....

behaviour and susceptibility to the effects of imper-


fections. Geometrical non-linearities are usually not
an issue for double layer grids which possess consider-
able overall stiffness. This feature has been taken
'--------------d
advantage of by many analytical models, a sample of Fig. 2. Strut model according to Schmidt. 8
which will be briefly reviewed hereunder. Following
this review the paper goes on to present the proposed
model, illustrate its application and assess its merits residual plastic stress level. The method uses a linear
(compared to other available models) and limitations. analysis program in which 'yielded' members are
removed, at each step, and replaced with their
2. Surveyof somemodels reported in the residual force applied externally. It therefore
literature involves, in effect, modification of the global stiffness
matrix at each step but the algorithm is simple to
Some typical methods are reviewed, which deal speci- implement using readily available linear analysis
fically with the problem of brittle-type strut buckling programs. The main limitations of the method are the
and its effect on double layer space truss behaviour. rough approximation made in the strut model and the
Each method is briefly outlined and evaluated. A ambiguity in determining the level of residual force.
summary, which also includes the proposed method, No allowance is made for elastic unloading of
is presented in Table 1. Two main features char- previously buckled members.
acterize each of the reviewed methods: Supple and Collins":" employ a piecewise linearized

Table 1. Comparison of methods

Method Schmidt2 ,8 Schmidt-Gregg 11 Supple-Collins 9 ,l o Smith 3 Proposed

Strut curve" E-B-P P.L. (limited) P.L. (limited) S.L. P.L.


Approximation Approximate Exact, no unloading Exact, elastic Exact, elastic Exact, modified
unloading unloading elastic
unloading
Equilibrium iteration No No No No No
Updating of stiffness matrix Yes Yes Yes No(?) No
Scanning of members All All All All Partial
Adaptation of linear programs Easy Difficult Difficult No information Relatively easy

8E-B-P, Elastic-brittle-plastic; P.L., piecewise linearized; (limited) Iimited number of segments can be accommodated; S.L., stepwise
linearized.
Practical approach to double layer grid analysis 35

F through and involve updating of the stiffness matrix,


re-solution and scanning of all truss members for the
next stiffness change. The scanning procedure is
somewhat simplified at the expense of some accuracy
in that no allowance is made for elastic unloading of
buckled members.
Smith" employs a 'stepwise linearized' strut curve
(Fig. 5). The advantage of this approximation to the
strut curve (which is derived analytically, but an
experimental curve can also be used) is that at any
load step any member can possess one of only two
'-----+---------.-d
stiffnesses--the original elastic stiffness or zero.
Fig. 3. Strut model according to Supple and Collins. 9.10
F

strut model (Fig. 3). Their program follows, step by


step, the sequence of strut buckling or stiffness change ,4
and at each step the truss stiffness matrix is updated. to
allow for changed member stiffness. At each step all
truss members are scanned to find the next member(s)
changing stiffness (including possible elastic
unloading) and a new solution is obtained using the
updated stiffness matrix. This time-consuming -- ----
procedure puts a practical limit on the number of
segments that can be accommodated in the strut curve '---------------d
linearization but, in principle at least, it is an 'exact'
Fig. 5. Strut model according to Smith. 3
procedure (Le. it converges to the exact solution).
Schmidt and Gregg 11 adopt a piecewise linearized
strut model (Fig. 4). The truss analysis employs the Furthermore, updating of the stiffness matrix is
concept of 'dual load' , in which a set of 'residual forces' avoided through the use of 'fictitious' forces. The
(R; in Fig. 4) is added to the actual applied loads as an algorithm is essentially a mesh formulation in which
additional load case. These residual loads make each the member forces at any step are given by:
step independent of previous loading steps so that
loading can proceed from any known configuration. {F} = {Fa} + {b}A+[[BJ-[I]]{Q}
In practice, however, all steps are usually gone
where {F} are member forces, {Fa} initial forces, {Q}
the fictitious forces, iI.. the applied load factor, [B] and
F
{ b } are the corresponding influence matrices and [I] the
R \ identity matrix. The forces {Q} are not actually com-
2
\
\ puted and only the submatrix of [B] corresponding to
\
\
members with zero stiffness is updated. The algorithm
\
\
does involve scanning of all members at each step,
\
\
however, for a change in regime and also a complex
\ system of 'book-keeping' associated with re-
structuring of matrix [B]. The stepwise linearization
does not afford the accuracy of piecewise linearization
unless a large number of segments is used. A truly
brittle-type buckling is also precluded although
reducing the extent of yield plateau would improve
the approximation.

3. Outlineof the proposed model


'-----,---;----+-------r-- d
The proposed model takes full advantage of prior
Fig. 4. Strut model according to Schmidt and Gregg. 11 information available on the collapse mechanism of
36 A. Hanaor

space trusses. In particular, only members confined or is a flexibility vector containing influence coefficients
adjacent to the yield lines need to be considered as of the external load on the 'redundant elongations'.
potentially behaving non-linearly (i.e. buckling or Equations (1) and (3) can be combined to yield:
yielding). All other members are assumed a priori to
remain linear-elastic throughout the loading. This [C*]{F}, = - {cH'A.; [C*] = [[C]'-[C]r] (4)
substantially reduces the time used in scanning and
computing at each load step, since load step cycles are Equation (4) is similar to the standard flexibility (force
performed only on potentially non-linear members. method) formulation. Here, however, the released
Thus a truss containing, typically, hundreds or structure is, in general, statically indeterminate and
thousands of members will have perhaps tens of hence the flexibility matrix [C]I(of order nr x nr, if nr
potentially non-linear members resulting in time is the number of redundants) cannot be computed
saving of a number of orders of magnitude. directly from the equations of statics. The matrices
The truss is 'split' or 'partitioned' into two groups of [C]Iand {c} f, which pertain to the linear substructure,
members: the elastic-linear substructure and the however, need to be evaluated only once. This can be
potentially non-linear group of members. The inter- done by employing a standard linear-elastic analysis,
action between the two substructures is formulated as analysing the released truss for the applied unit load
a mesh-flexibility problem, in which the potentially plus unit values of the redundant forces, one at a time,
non-linear members are considered as 'redundants' as separate load cases. Assuming the standard
and the linear substructure as the 'released' program is a stiffness formulation, this can be
structure: expressed as follows:
[e}, = [C]r{F}, = [A]r{D}\ (1)
(5)
where {e} r are member elongations, [C]r is a diagonal
matrix containing redundant members' flexibilities {Po} is the unit applied load vector and [K]I is the
(L / Ea), {F} r are redundant member forces, [A], is the global stiffness matrix of the released truss. The
submatrix of the 'kinematic' or 'geometric' matrix resulting nodal displacement arrays {D} P and [D Y can
containing redundant members' direction cosines and then be used to compute the required flexibility
{D} I are the nodal displacements of the released arrays:
structure. Subscript r denotes entities related to the
(6)
redundant members while I refers to the linear sub-
structure.
Referring to stability, in general, when selecting Equation (4) now provides a relationship between
redundants, attention needs to be given to the stability redundant forces and the load factor in which only the
of the released structure. As shown by Schmidt, 12 diagonal elements of [C], need to be evaluated at each
however, space trusses with a whole chord removed load step. These elements are defined by the slope of
still possess substantial stiffness, so that the risk of the input strut curve(s) (Fig. 6) for each of the
selection of redundants leading to instability is redundant members.
minimal. Some exercise of judgement is, nonetheless, F
required in the selection of redundants.
The vectors {e}r and {D}\ can be split into two
influence groups:

{D}\ = {DH + {D}i; {e}, = {e}; + {e}; (2)

where the superscript P denotes 'due to the external


load' and superscript F denotes 'due to the redundant
forces'.
If redundant forces are considered as external
forces acting on the linear subassemblage, the Extrapolation
flexibility relation can be expressed as:

{e}, = [CMF},+ 'A.{cH = {e}~ + {e}; (3)

where A is the applied load factor and [C]t-a square


symmetric matrix-is the flexibility matrix of the l..-----------?------_d
released structure in the redundant coordinates. {c}f Fig. 6. Proposed algorithm--strut model.
Practical approach to double layer grid analysis 37

The load factor A and the 'central displacement' are Stage II (new program), using the whole of
computed (through eqn (4» for each point at which a matrix [D] as input.
redundant member changes stiffness (including first
Stage II
yield or buckling or unloading-re-loading).
At each load step (change of member stiffness):
The 'central displacement'-the displacement at
coordinate c-is given by: 1. Adjust diagonal element(s) of [C]* to allow for
change in stiffness of 'current critical member'.
de = Ad~ + {D}~{FL (7) If this is the first load step (linear-elastic), just
create C* = C1 - Cr.
where {DeY is row c of [DY. The program terminates 2. Solve eqn (4) for A = 1 (i.e. with A as
when De exceeds a prescribed value. parameter) .
3. Search for 'next critical member'-next member
4. Strut characteristic and solution algorithm to change stiffness yielding lowest change in
absolute value of A. Record current value of A.
The piecewise linearised strut characteristic is 4. Check if any previously buckled member is
presented in Fig. 6. The curve incorporates, as part of unloading (as a result of stiffness change in
the input points, an elastic unloading portion. The previous cycle). Such unloading is indicated by a
curve can be derived experimentally or analytically 'negative' increment of axial deflection
but the derivation does not form part of the algorithm. (elongation) computed from eqn (1). If such an
The elastic unloading-re-loading portion of the post- unloading member is encountered, it becomes
buckling curve has a smaller stiffness than the original, the next critical member.
virgin stiffness, due to the curvature of the buckled 5. Compute member forces and central deflections
member. An allowance for this reduced stiffness is under the updated load factor. Check if de
made in the proposed algorithm by interpolating or exceeds its prescribed limit, if so-terminate.
extrapolating the elastic unloading-re-loading stiff- 6. GOTO (1)
ness between the virgin and the input post-buckled
As an illustration of the efficiencyof the procedure,
values. The elastic portions are reversible (unloading-
a truss with 316 members and 8 potentially non-linear
re-loading); the inelastic post-buckling curve is
struts took a total of 50 s CPU to compute (on a CDC
irreversible.
Cyber 73) out of which 43 s were used by the linear
4.1. Algorithm elastic analysis (Stage I).
Since Stage II is limited to a relatively small number
The solution is performed in two disjoint stages. This of redundant members it can be performed entirely in
can take the form of two separate programs, in which central memory with considerable time saving.
program I computes and stores data for program II, or
of an overlaid program. 5. Two examples
Stage I
The two experimental and analytical examples
1. Carry out linear analysis of the linear sub- illustrate the usefulness as well as the limitations of the
structure for unit applied loads and for unit axial proposed model in predicting truss behaviour.
forces along axes of redundant members, one at The first example is of a truss free of the effect of
a time as a separate load case (redundant imperfections. The subassemblage shown in Fig. 7(a)
members themselves are not included in this is internally statically determinate-member forces
analysis). can be computed given the external load and support
Any standard analysis program can be used reactions using equilibrium equations alone. Effects
for this purpose. A subroutine for computing of support misalignment were mostly eliminated by
axial force components given incidence of means of adjustable hangers. The experimental strut
redundant members can readily be added to curve used in the analysis is presented in Fig. 7(b) and
generate load data. experimental and analytical results in Fig. 7(c).
2. Using output nodal displacements and Agreement between experimental and analytical
redundant members' incidence, compute matrix results is good for the early post-buckling stages. At
[C]l' vector {CH and row vector {De}. Store later stages, geometric non-linearities (large displace-
these data for use in Stage II. These computations ments) start playing a major role. The difference
can be done using a subroutine added to the between theory and experiment at first buckling (less
original program. Alternatively it can be done in than 10%) is well within the expectable scatter in strut
38 A. Hanaor

F/~

R R

R R

{bJ

R R
.5

R R

VyVR (0 1

0 2 3 I. 5 6 7 dft:la
p

.5

I
01 D.
a 2 3 I. 5

Fig. 7. Imperfection free subassemblage: (a) layout, (b) experimental strut characteristic; (c) experimental and analytical results.

buckling force. eliminated. The average experimental strut character-


The second subassemblage (Fig. 8(a)) is a sub- istic is presented in Fig. 8(b) and results in Fig. 8(c).
assemblage with an internal degree of static Some values of 'critical' member forces are also shown
indeterminacy of 3 and overall indeterminacy of 12. It in Fig. 8(c), including initial forces (at zero applied
was, therefore, subject to members' lack of fit as well load). It is apparent that in the presence of imper-
as to some support misalignment which could not be fections, large discrepancies exist between actual and
Practical approach to double layer grid analysis 39

F/£
1 - - _.

(b J

.5

R R

\zyVR Ia )

P o 2 3 4 5 6 7 8

P/~
F/~
1

'"
Go
<J
5
L-
o

L-
01
Ll
-c-;
__ gv,-_ _ _
E
01 (
E
I
__fTli~ _ _I_
I
.5 I


f\
/ •. Zonp of mpmber forces 1-4

/;
. . ·.11 lc I
1/'I·I
/

o
, ..
2 3 4 0100

Fig. 8. Subassemblage with lack of fit: (a) layout; (b) experimental strut characteristic; (c) experimental and analytical results

predicted truss behaviour. The discrepancy is parti- thus serve as a design tool by suggesting adoption of
cularly marked at first buckling with some 25% increased safety or load factors in imperfection-
reduction in load carrying capacity. The highly'brittle' sensitive trusses. Alternatively, measures to increase
nature of the theoretical curve can serve as an ductility (e.g. refs 5,13 and 14) can be adopted.
indication of susceptibility to premature buckling and The discrepancy in load bearing capacity at first
sensitivity to imperfections. The analytical model can buckling between theory and test is fully accounted
40 A. Hanaor

for in terms of the effect of imperfections, if com- University of Melbourne, Australia, under the
parison is made between theoretical and actual values guidance of Professor L. C. Schmidt.
of the average force in 'critical' members, all of which The paper has been prepared at, and with technical
should nominally have the same force. and clerical assistance from, the Department of Civil
and Environmental Engineering, Rutgers University
6. Concluding remarks of New Jersey, under the chairmanship of Professor
E. G. Nawy.
-A highly efficient analytical model for the post-
buckling analysis of double layer grids has been
References
presented.
-The model takes full advantage of special 1. MARSH, C.; Collapse of point supported space grillage,
conditions prevailing in double layer grids. In ASCE J. Structural Division, 103 (ST9) (September 1977),
particular, prior knowledge of yield line pattern 2. 1703--12.
SCHMIDT, L. c., MORGAN, P. R. and CLARKSON, J. A,
is being utilized, resulting in a practical and Space trusses with brittle-type strut buckling, ASCE J.
efficient procedure with no sacrifice of accuracy. Structural Division, 102 (STI) (July 1976) 1479-92.
SMITH, E. A, Space truss nonlinear analysis, ASCE J.
Phenomena like brittle-type buckling and elastic 3. Structural Engineering, 110 (4) (April 1984), 688-705.
unloading-re-Ioading (at reduced stiffness) are 4. SCHMIDT, L. c., Some aspects of the behaviour of a
all taken into account. yielding space truss, Proceedings 3rd Australasian Conference
-The model is easy to implement through on the Mechanics of Structures and Materials, University of
Auckland, New Zealand, August 1971.
relatively simple additions to existing linear 5. HANAOR, A and SCHMIDT, L. c., Space truss studies
analysis programs. with force limiting devices, ASCE J. Structural Division, 106
-The advantages of the model are fully exploited (STU) (November 1980),231>-29.
6. SCHMIDT, L. c., MORGAN, P. R. and HANAOR, A,
in cases where yield lines or yield zones are well Ultimate load testing of space trusses, ASCE J. Structural
defined and known in advance, but its applic- Division, 108 (ST6) (June 1982), 1324-35.
ability is not limited to such cases. 7. MADI, U. and SMITH, D. L., A finite element model for
determining the constitutive relation of a compression
-Like other analytical models based on idealized member, In: Third International Conference on Space
assumptions, the model fails to account for the Structures, ed. H. Nooshin, Elsevier Applied Science
effects of imperfections, particularly of lack of fit. Publishers, London, 1984, pp. 625-9.
These effects have been shown to cause signifi- 8. SCHMIDT, L. c., Effects of compression chord buckling on
the behaviour of a simply supported space truss, Proceedings
cant reductions in truss load-bearing capacity. 5th Australian Conference on the Mechanics ofStructures and
Reductions of up to 40% have been reported but, Materials, Melbourne, Australia, August 1975, pp. 457-70.
typically, reductions of the order of 25% can be 9. SUPPLE, W. J. and COLLINS, I., Limit state analysis of
double layer grids, In: Analysis, Design and Construction of
anticipated in imperfection-sensitive cases. The Double-Layer Grids, ed. Z. S. Makowski, Applied Science
post-buckling curve can give an indication of Publishers, London, 1981, pp. 93--117.
imperfection sensitivity, through the slope of the 10. COLLINS, I. M., An investigation into the collapse
behaviour of double-layer grids, In: Third International
curve following initial yield, and suggest design Conference on Space Structures, ed. H. Nooshin, Elsevier
measures to prevent premature buckling. Applied Science Publishers, London, 1984, pp. 400-5.
-Imperfections are difficult to model analytically, 11. SCHMIDT, L. C. and GREGG, B. M., A method for space
being a randomly distributed parameter. Further truss analysis in the post-buckling range, Intemat. J.
Numerical Methods in Engineering, 15 (1980),237-47.
more, technical factors such as joint behaviour 12. SCHMIDT, L. C.; Load transfer by shear and torsion in
appear to playa significant role and may render plate-like space trusses, Civil Engineering Transactions, The
an analytical model of full predictive capacity Institution of Engineers, Australia, 1981.
13. SMITH, E. A, Ductility in double layer grid space trusses,
practically unattainable. In: Third International Conference on Space Structures, ed. H.
Nooshin, Elsevier Applied Science Publishers, London, 1984,
Acknowledgements pp.51D-15.
14. MARSH, c., Optimization of space trusses using non-linear
behaviour of eccentric diagonals, In: Third International
Experimental and analytical groundwork was carried Conference on Space Structures, ed. H. Nooshin, Elsevier
out at the Department of Civil Engineering, Applied Science Publishers, London, 1984, pp. 669-71.

You might also like