You are on page 1of 11

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL.

15, 237-247 (1980)

A METHOD FOR SPACE TRUSS ANALYSIS IN THE


POST-BUCKLING RANGE

LEWIS C. SCHMIDTt AND BERNHARD M. GREGGt


Department of Civil Engineering, University of Melbourne, Parkville, Victoria, Australia

SUMMARY
A dual load method of truss analysis is presented that allows highly nonlinear member behaviour to be
followed. In particular, a brittle type of strut behaviour is considered. The method assumes that the
inelastic strut response is known and that piecewise linearization of the behaviour is acceptable. An
example is given to illustrate the type of problem that can be handled. The example involves negative
stiffness of the members and consequently attention must be given to the equation solving procedure for
the structure. A comparison is made with some observed test results.

INTRODUCTION
The stiffness method of structural analysis is widely used for the elastic analysis of space trusses.
By scaling the a.pplied load it is possible to find the load at which the truss first becomes inelastic
in behaviour. Several tests,' however, have shown that in the case where compressive member
collapse of a brittle type can occur the load capacity of the truss may not reach the theoretical
load at first yield. Under these conditions it is recognized that initial member imperfections,
initial force systems due to fabrication and erection procedures, and the natural variations in the
peak load capacities of the struts, have a significant influence on the truss behaviour so that the
idealized theoretical initial yield load may not be reached.
It has been shown' that the post-elastic range of behaviour of the truss isof importance, and it
may provide a better indication of truss ultimate load capacity than the load at initial yield. In
order to investigate this range, the member post-ultimate behaviour must be considered.
Various simplified compressive member load (P)versus axial deflection (6) behaviour models
have been used,'-3 as shown in Figure l(a) and (b). These models of behaviour are highly
idealized, although it is thought that they furnish reasonable results for overall truss response.
In many patented space truss systems, the method of connection of members at the joints
approaches the pin-ended condition so that there is little moment interaction between adjacent
members. The form of the real load versus axial deformation relationship for the compressive
members will therefore be known ab initio, whether by experiment or detailed analysis.
For a transition-length strut the relationship between load (P)and axial deformation (6) will
be of the form shown in Figure 2. An idealized strut model of the form of Figure l(a) has been
used' as an approximation to this type of real behaviour. It is necessary, however, to improve the
idealized model as an approximation to the real curve and to carry out comparative studiesof the
effects of the refinement.

t Reader.
$Research student.

0029-598 1/80/02 15-0237$0 1.OO Received 5 September 1978


@ 1980 by John Wiley & Sons, Ltd.
231
238 L. C. SCHMIDT AND B. M. GREGG

Figure 1. Idealized compressive member models in inelastic range: (a) transition length strut; (b) stocky strut behaviour
with plateau

This paper is primarily concerned with a treatment of the post-buckling behaviour of


compressive members; it therefore covers general nonlinear behaviour of members.

BRIEF REVIEW OF ‘INITIAL STRESS’ METHOD


In order to set the proposed method in context it is worth while recalling briefly the initial stress
method proposed by Zienkeiwicz et and subsequently renamed the ‘residual force
method.”
In terms of a one-dimensional state of stress, the method is illustrated in Figure 3.
Following standard techniques the initial stiffness matrix K is calculated, and the nodal
displacements are found by solving
A=K-’Q (1)
where A is the vector of nodal displacements, and Q is the vector of applied nodal loads.
The member (element) strains E are found from
E=BA
where B is the strain-deflection transformation matrix.
From these strains a vector of stresses (T can be found,
u=DE
where D defines the constitutive relationship for the initial linear range for the material.

Figure 2. Real behaviour of pin-ended transition length strut


SPACE TRUSS ANALYSIS 239

00,rl i A
E

Figure 3. One-dimensional 'initial stress' process

The strains and the equation defining the general constitutive relationship of the member or
element behaviour furnish another set of stresses:
u=f(.?) (4)
The difference between the stresses in (3)and (4)represents a set of out-of-balance stresses.
Therefore, by applying the reverse of these stresses as a set of initial stresses, the origin of the
element stresses for the elastic reanalysis is shifted to point A, as shown in Figure 3. The
reanalysis is carried out with the same stiffness matrix, which furnishes an answer closer to an
equilibrium position. Several iterative cycles may be necessary until some criterion of con-
vergency is satisfied.
An advantage of this approach is that the partial inversion of the original stiffness matrix can
be retained for the successive cycles without the need for reassembly of a new K matrix for
either each cycle of iteration or each external load increment. However, the procedure is
suitable only for mildly nonlinear problems.
For any situation where the elements suffer significant strain softening the usual iterative
procedure may lead to convergence problems. The likelihood is therefore remote of then
following post-ultimate load paths for structures of the type considered herein, especially if the
stress-strain curve were of the general shape of Figure 2.

DUAL LOAD METHOD


The method to be discussed here has some relationship to the initial stress method described
above.
240 L. C. SCHMIDT AND B. M. GREGG

6, 6, 63 6, 62 63

(a) compressive member (b) tensile member

Figure 4. Piecewise linearization of member behaviour: (a) compressive member; (b) tensile member

Figure 4(a) shows the type of piecewise linearization of the compressive member behaviour
that is proposed for the curve of Figure 2. Figure 4(b) shows a yielding tensile member with
strain hardening.
A typical segment of Figure 4(a) is shown in Figure 5 as line AB, where SiPrand Si define the
deformation limits of validity of the segment, Rirepresents the force intercept on the force scale
when S = 0, and the slope is a measure of the member stiffness in the range AB.

Figure 5. Residual force determination for segment of typical compressive member behaviour
SPACE TRUSS ANALYSIS 24 1

The line AB can be shifted to position AB’, as shown in Figure 5, so that when it is
extrapolated it passes through the origin. The line AB’ then represents the same stiffness as
before, but the residual force R, must be applied to thestructural system. The use of this second
set of applied loads Ri, or residual forces, together with the unit external load applied to the
structure with revised member stiffness, has led to the name ‘dual load method.’
In the post-buckling range of behaviour of some of the members of the structural system, the
stiffness matrix for the structure will eventually possess negative elements on the leading
diagonal. An equation solving procedure must therefore be incorporated into a computer
program to handle this situation. In this instance the decomposition technique suggested by
Melosh and Bamford6 has been used.
The essential points of the solution procedure can be summarized as follows.
Depending on the axial deformation for each member, an appropriate stiffness and residual
force is known from the assumed piecewise linear relationships. A first-order linear analysis for
two separate loading conditions is then carried out on the structure whose members possess
these newly determined stiffnesses.
The two loading conditions are the unit external load and the residual forces for each member.
Introducing an additional subscript to denote the jth member, the resultant force, F,, for this
jth member is obtained by superposition of the values from the two load cases and the residual
force R,, for the member, so that

F,, = AFc, + F r j + Ri,


Here FeIis the force in the jth member due to unit external load, A is a load factor whose value
is to be determined, F, is the force in the jth member due to the current set of residual forces. F,,
can be taken as the ith limit value under increasing deformation for the current linear range of
the jth member.
The two values of A that are of interest are those which satisfy

-
sgn (Ei-F,j Rii)= sgn (Fej)
which ensures A 0,
and min A :

denoted as A,
and max A :
sgn (el)
f sgn We,)

denoted as A h (5)
where the search is carried out over all n members of the structure (j= 1, . . . , n ) .
If both of these A values exist, the value chosen is that which furnishes a consistent solution
(i.e. the member deflections are within the current limits), and is closer to the previous value of A.
If neither As nor A,, furnishes a consistent solution, a check is made to see whether unloading
has occurred in the associated members or whether the next linear range of behaviour of the
members needs to be used.
The overall deflections can be found from
h= A A, -t Al (6)
242 L. C. SCHMIDT AND B. M. GREGG

where A, Ae and A, are the vectors of deflections corresponding to the total load, the unit
external load, and the residual forces, respectively.
The overall truss load-deflection behaviour is found essentially by a secant type of approach,
but by following each segment of the piecewise-linear deflection behaviour of the members, the
dependence on prior history of loading can be followed. In the computer program developed so
far, however, elastic unloading has not been incorporated if reversal of deflection is sensed.
The structure load-deflection response can be followed step by step so that as each new
member or group of members is about to change its state an equilibrium point can be found. It
should be noted that this approach requires the stiffness matrix to be recalculated for each
change of state.
A significant advantage of the method of analysis occurs when it is possible to anticipate a
well-defined condition of the structure in the inelastic range. Some discussion' has been given on
this point in cases where lines of collapsed compressive members have been formed. It is then
possible to initiate analysis in this region. Some iteration, however, may be necessary in order to
find the deflections of the failed members that are compatible with the current stiffness used in
the analysis.

COMPARISON OF METHODS
The initial stress and dual load methods are similar in some respects for the type of problem
considered herein. In both cases a shift of origin for the element stress or force is used as
deformation proceeds.
The initial stress method requires equation (4) to be calculated after each equation solution
step, and similarly the dual load method requires the calculation of member deformations in
order to check that the correct stiffnesses have been used for the formation of the stiffness
matrix.
The initial stress method normally follows the structure load-deflection curve by application
of a series of external load increments and analysis of the structure for each of these. The dual
load method follows the structure behaviour by application of the statements given in ( 5 ) . For
the initial stress method the accuracy of the solution will be dependent on the size of the external
load increments, whereas for the dual load method the accuracy will be dependent on the degree
of piecewise linearization of the element behaviour.

APPLICATION
Experimental results are available' on the inelastic behaviour of three nominally identical
simply supported space trusses of the type shown in Figure 6, where the layout, loading and
initial member properties (area times modulus of elasticity, A € ) are given.
The three trusses tested were designed so that compressive chord collapse would occur before
any tensile chord yield. The jointing system adopted also ensured that compressive chord
members would buckle with very little interaction between adjacent panels.
Four control tests on the pin-ended compressive chord members are shown in Figure 7.
Superimposed on these is an idealization of the axial load behaviour (marked d) that is a better
approximation than that used previously,' which was of the form of Figure l(a).
The three experimental results for the overall truss behaviour are shown in Figure 8, where
the truss load is non-dimensionalized by dividing by the theoretical truss peak load derived from
the strut behavior d of Figure 7. The central vertica! deflection A is similarly non-dimen-
sionalized. The theoretical result derived using curve d of Figure 7 is indicated by d in Figure 8.
SPACE TRUSS ANALYSIS 243

0.216 m

Top -
Chord

Bottom -
Chord

W e b Members
n o t shown

L 1_ 1-L 1 , L A -
O L o a d i n g points

Figure 6. Truss layout and member properties: upper chord, AE = 2.28 h4N; web members, A E = 2.52 MN; lower
chord, AE = 2.76 M N

Load
kN
M

2 4 6 8 10 12 14
Deflection mm

Figure 7. Observed pin-ended strut curves with theoretical idealization d


244 L. C. SCHMIDT AND 8 . M. GREGG

Q
G
1.0

0.5

Figure 8. Observed results from three truss tests with theoretical curve d

I I 1 1 I I
1 2 3 4 5 6
6
-
6,
Figure 9. Idealized compressive member load (P/P,)
vs. deflection ( 6 / 6 , )curves
SPACE TRUSS ANALYSIS 24 5

The comparison of results in Figure 8 shows that the theoretical prediction of initial yield is
not good, as an approximate 27 per cent difference exists between the theoretical and
experimental values. Effects, such as initial forces, possible slippage of members at joints leading
to internal force redistributions under load, together with the variation in strut peak load
capacities (as shown in Figure 7) and accentuated by the small amount of interaction of
members, could lead to premature buckling of some struts. These effects could furnish a lower
initial yield strength than the theoretical analysis described herein could predict.
However, the subsequent behaviour is interesting in that the theoretical and experimental
results tend to the same level of load capacity and display an equivalent ductile truss behaviour,
even though the strut buckling of Figure 7 is of a brittle type. It would appear that a probabilistic
approach will need to be developed if greater refinement is required.
In order to illustrate the sensitivity of the inelastic truss response to the form of the strut
buckling curve, several non-dimensionalized compressive member curves, as shown in Figure 9,
were considered. The corresponding non-dimensionalized truss load-deflection curves are
shown in Figure 10. The results show that the member load capacities can suffer a considerable

1-5

-
Q

QY

1.0

0.5
/

1 I 1
0
1 2 3
-
A
AY

Figure 10. Truss load (Q/Q,) vs. central deflections (A/A,) corresponding to member behaviour shown in Figure 9
246 L. C. SCHMIDT AND B. M. GREGG

drop below their initial yield load before the truss load capacity falls below the level correspond-
ing to its initial yield load.
The lines of collapsed compressive members derived from the theoretical analyses are
illustrated in Figure 11.The results shown in Figure 10 have been taken only so far as to cause
the imminent completion of these collapse lines, as it has been argued elsewhere' that this
condition furnishes a reasonable estimate of the post-initial-yield reserve of strength and gives
an indication of the likely ductile behaviour of the truss.

Figure 11. Lines of collapsed compressive members

CONCLUSION
A method of analysis has been presented that allows known, nonlinear behaviour of members to
be followed. In particular, the compressive member curves considered can be highly nonlinear,
involving discontinuities and negative stiffness in the post-initial-yield ranges. The method
allows the overall structural response to be determined in a straightforward manner.
The method is essentially a secant type of analysis, but a step-by-step approach can be used
allowing a full load history to be followed if necessary. The method, however, has application to
situations where a well-defined failure condition can be anticipated, so that a relatively direct
analysis can be made of this condition if the effects of prior load reversal in members is unlikely
to be serious.
The application of the analysis technique to some observed behaviour of space trusses has
highlighted the importance of effects not included in the analysis. The influence of the scatter of
peak values of element behaviour when they behave in a brittle type of manner, the influence of
initial force systems caused by fabrication and erection, and the influence of possible slippage at
joints during loading are factors that need further investigation.

ACKNOWLEDGEMENTS

The authors are grateful to the Department of Civil Engineering, University of Melbourne, for
providing facilities for the work to be carried out. L. C. Schmidt was on leave at St. John's
College, Cambridge, during the initial stage of the work; B. M. Gregg held a University of
Melbourne Research Grant.
SPACE TRUSS ANALYSIS 247

REFERENCES
1. L. C. Schmidt, P. R.Morgan and J. A. Clarkson, ‘Space trusses with brittle-type strut buckling’, Proc. A.S.C.E.
102(ST7),1479-1492 (1976).
2. L. C . Schmidt, ‘Member buckling characteristics and space truss behaviour’, I.A.S. W. World Congress Space End.
Concordia Univ.. Montreal, July, pp. 849-857, 1976.
3. L. C. Schmidt, ‘Effects of compression chord buckling on the behaviour of a simply supported space truss’, Fifrh
Ausrrulusiun Conf. on Mech. of Srr. and Murk Melbourne and Monash Univ., Melbourne, August, pp. 457-470,
1975.
4. 0. C. Zienkiewicz, S. Valliappan and I. P. King, ‘Elasto-plastic solutions of engineering problems “initial stress”,
finite element approach’, Inr. J. num. Merh. Engng, 1, 75-100 (1969).
5. G.C. Nayak and 0.C. Zienkiewicz, ‘Elasto-plastic stress analysis. A generalization for various constitutive relations
including strain softening’, Inr. J. num. Merh. Engng, 5, 113-135 (1972).
6. R.J. Melosh and R. M. Bamford, ‘Efficient solution of load deflection equations’, Proc. A.S.C.E.95(ST4),661-676
(1969).
7. J. A.Clarkson, ‘Ultimate load analysis of space trusses’, M. Eng. Sc. Thesis,Univ. of Melbourne (1975).

You might also like