c0mpu1m & SfnlclweJ Vol 13. pp 55-62
Printed in Great Britam All rights remvcd
0043-7949/81/01055-Qaso2.m~
Copyright (B 1981 pcrgmon Press Ltd
A FAST INCREMENTAL/ITERATIVE SOLUTION
PROCEDURE THAT HANDLES
“SNAP-THROUGH”
M. A. CRISFIELD
Transport and Road Research Laboratory, Department of the Environment,
Crowthornc, Berkshire RGl 1 6AV, England
(Received 25 April 1980)
Abstract- Riks [l) has recently proposed a new solution procedure for overcoming limit points. To this
end, he adds, to the standard equilibrium equations, a constraint equation fixing the length of the incre-
mental load step in load/deflection space. The applied load level becomes an additional variable.
The present paper describes a means of modifying Rik’s approach so that it is suitable for use with the
finite element method. The procedure is applied in conjunction with the modified Newton-Rapbson method
in both its original and accelerated forms. The resulting techniques not only allow limit points to be passed.
but also, improve the convergence characteristics of the unconstrained iterative procedures. Illustrative
examples include the large deflection analysis of shallow elastic shells and the collapse analysis of a stiffened
steel diaphragm from a box-girder bridge.
INTRODUCllON
Snap-through and snap-back buckling phenomena
(Fig. 1) pose some of the most difficult problems in non-
linear struotural analysis. For most practical problems,
it is quite unnecessary to trace such a convulated load/
deflection path as that shown in Fig. 1. Indeed, were
such a path to be traced, most analyses would trace the
static path ABCDEFGHIJ and thus infer the dynamic
“snaps”. Although the analyses are therefore somewhat
artificial, they may be very important.
Dd*n,on. p
Fig. 1. “Snap buckling”
For some problems. all that may appear to be
required is the load level at the first limit point. How-
ever, without analysis techniques that allow the limit
points to be passed. even this information may be un-
available or unreliable. “Collapse loads” are often
associated with a failure to achieve convergence with
the iterative solution procedure. However, it may only
be the iterative solution procedure that has collapsed
(possibly as a consequence of round-off error). For
other problems. the analysis may be performed on an
individual component of a complete structure. In such a
situation, it may be important to obtain information on
the nature of the load shedding following the limit
points, in order to assess the performance of the com-
plete structure.
When analysing relatively simple structures, it is
tempting to try and avoid the full complexities of a
“snap analysis” by applying a simple form of displace-
ment control. Such ‘displacement control” may simul-
ate a physical testing procedure. For instance, referring
to Fig. 1, if the displacement p were to be prescribed,
the limit point B could be passed and the load-shedding
curve BC could be traced. However, a similar procedure
would fail at_ or just before, the limit point G. This fail-
ure might not matter if the analyst could conclude
“Following the (local) maximum at F, there is a very
sharp drop-off in load”. However, the dramatic non-
linear behaviour associated with the limit point G may
induce a failure in the incremental/iterative solution
procedure at point E (Fig. 1). In such a situation, the
analyst is left with no information on the nature of the
failure and may not even be sure that he has a htruc-
tural (rather than numerical) collapse. Consequently,
non-linear finite element computer programs should
be provided with solution procedures that will handle
such %napping phenomena” particularly if this can be
achieved without resorting to many ‘special tech-
niques” [2]. Present procedures generally involve the
use of fictitious springs [3] or adopt some form of
“displacement control” [4, S]. The main disadvantage
of the former method is the trial and error often in-
volved in the selection of the appropriate springs [6].
The disadvantages of the latter method relate to the
selection of the appropriate displacement variable. For
instance, Maewal and Nachbar [7] note the necessity
to change the prescribed displacement variable fol-
lowing slow convergence or divergence of the iterative
solution procedure.
Began et al. [6,8] use the ‘kurrent stiffness param-
eter” [S], to predict the position of the local maximum
or minimum. They then suppress the equilibrium itera-
tions in the ncighbourhood of the extremum (limit)
point and reverse the sign of the load following a change
in the sign of the determinant of the tangent stiffness
matrix. The supression of the equilibrium iterations
dictates the provision of very small load increments
55