J. JBNSSON, S. KRENK
Department of Building Technology and Structural Engineering, University of Aalborg, Sohngaardsholmsvej 57,
Aalborg, Denmark
AND
L. DAMKILDE
Department of Structural Engineering, Technical University of Denmark, Lyngby, Denmark
SUMMARY
In the formulation of the semi-Loof element the rotation of the tangent plane is derived from the
interpolation of the transverse displacement, while the rotation of the normal is interpolated separately
by another set of shape functions. The geometric stiffness matrix can be formulated by use of either of
the two rotation representations. It is demonstrated that the use of the tangent plane representation in
the geometric stiffness matrix is far superior to the common form at present.
INTRODUCTION
In plate theory, translation as well as rotation of the normal to the plate must satisfy continuity
requirements. These requirements complicate the formulation of displacement-based finite
elements for plates. In the semi-Loof elements - originally proposed by Irons - the transverse
displacement component and the rotation components of the normal to the undeformed plate
are interpolated independently in terms of the values at suitably located nodes. The tangent
plane of the deformed plate is defined in terms of derivatives of the interpolation of the
transverse displacement. As a result of this procedure the slopes of the tangent plane and the
rotations of the normals are represented by different sets of shape functions, and a point-
matching procedure must be used to relate them. In Kirchhoff plate elements equality is
imposed at selected points, while Mindlin elements use these points t o evaluate the shear strain.
The intention is that after matching at selected points the two sets of rotations are equal to
within the order of shear strains.
In the formulation of the geometric stiffness matrix for use in stability analysis a rotation
field is needed. In the analytical formulation of plate theory it does not matter which set of
rotations is used, because the difference between them is of the order of shear strains, i.e. by
assumption an order of magnitude smaller than the rotations associated with instability.
However, in the semi-Loof element the use of different functions to represent the two sets of
rotations may lead to differences in the performance of the resulting elements. Indeed, as
demonstrated in the Examples Section of this paper the use of the slopes of the tangent plane
in the formulation of the geometric stiffness matrix leads to far better performance of the
element.
In this work both formulations were implemented for rectangular semi-loof plate elements,
as briefly outlined in the following Section. This allowed a direct comparison between the two
alternatives by way of examples. It also enabled comparisons to be made with results from the
literature and results from test runs with the commercial finite-element code PAFEC. The
descriptions of the semi-loof element that have been available, e.g. Martins and and
, ~ not been specific about the details of the implementation
PAFEC Theoretical M a n ~ a l have
of the geometric stiffness matrix. However, tests have demonstrated that the PAFEC code
makes use of the rotations of the normals, and other results are of the same or less accuracy.
A considerable gain in accuracy is obtained by using the slopes of the tangent plane, and this
suggests that it may be of interest to develop and implement this formulation for the general
semi-loof shell element. This paper concerns the commonly used displacement formulation.
A hybrid formulation is given by Wang and Pian.' Using high-order fields for the transverse
displacement they directly relate the normal rotations to the tangent rotations.
ELEMENT INTERPOLATION
A unique feature of the semi-loof element is the interpolation scheme for displacements and
rotations. The displacements are represented via ordinary 9-point Lagrange interpolation,
while the rotation vector is interpolated by a special scheme introduced by Loo f combining
two Gauss points at each side with a node at the centre of the element. Figure l(a) shows the
nine transverse displacement and the 18 rotation components.
The 2D Lagrange interpolation follows directly from 1D interpolation by formation of
products. This gives the polynomial basis
(1; 4, t; EZ, 4r, t2;t 2 t ,4t2;t2t2) (1)
The corresponding nine Lagrange interpolation functions can be expressed by the formula
Fj(t,t)= [ftfE(Ej+ t ) + ( l - E f ) ( l - ~ ~ ) I[ + t j t ( t j + + ) + ( 1 - t f ) ( 1 -tZ)1,
j = 1, ...,9 (2)
Interpolation at Loof nodes, located at the normalized distance 1/,/3 from the centre of the
side, requires a different polynomial basis. It turns out that the term ['v2 must be replaced
by [t(t2- q2), whereby the Loof-node basis is
(1; E, t; E2, E t , tz;t2t,4 t 2 ; E t ( P - r2)) (3)
Figure 2 shows a simple direct construction of the boundary Loof-node interpolation
functions. Clearly the boundary interpolation functions must include a factor 4 or 7, and they
must therefore be constructed from the polynomial basis (3) excluding the first element, 1 . The
boundary Loof-nodes are divided into two groups. The first group containing the nodes 1 , 2,
5 and 6 have co-ordinates (E, 7) = ( ? 1/,/3, k l), while the second group containing the nodes
3, 4, 7 and 8 have co-ordinates (4, 7) = ( 51 , 5 1/,/3). For the first group of nodes (1, 2, 5 and
6 ) the principle consists in forming four functions with different symmetry about the co-
ordinate axes that vanish at all the nodes in the second group. The functions are normalized
such that the values at the nodes 1, 2, 5 and 6 are ? 1/4. The polynomial basis (3) leads to
a unique set of functions, and the interpolation functions of the nodes in the first group follow
as sums and differences of these four functions. The interpolation functions in the second
SEMI-LOOF ELEMENT 13
Figure 1. (a) The original 27 degrees of freedom; (b) the final 16 degrees of freedom
group follow by interchanging the co-ordinates E and 7. As a consequence all the boundary
Loof-node interpolation functions can be obtained from one by use of symmetry. When the
co-ordinates of Loof node j are denoted ( t j , v j ) , the general expression for the boundary node
shape functions is
This formula contains the four functions of Figure 2 in suitably normalized form. The central
Loof-node shape function is
L9(E, a ) = 1 - i ( t 2+ v 2 ) (5)
14 J. JBNSSON, S. KRENK AND L. DAMKILDE
t
t
LOCAL CONDENSATION
In the Kirchhoff plate element the original 27 degrees of freedom are reduced to 16 by local
condensation. The condensation is accomplished by imposing 1 1 conditions on the transverse
shear-strain components
Upon introduction of the interpolation representations (2), (4) and (5) the shear strain
components are
(2) The two rotation components 0'; and P', at the central node are eliminated by imposing
zero mean value of the transverse shear-strain components over the element area:
a = 1,2 (9)
(3) The displacement component w9 at the central node is eliminated by imposing a zero
mean value of the normal shear-strain component around the element contour:
With these constraint equations the original displacement vector with the 27 components
In the rotation formulation the geometric stiffness matrix before condensation is expressed in
terms of the 18 displacement components ( O ; , P i ; /?:,/3$; ...; p?, PS):
0
"1
L9
dA
16 J. JPINSSON, S. KRENK AND L. DAMKILDE
It is observed that due to the differentiation involved in forming w , the~ integrand in (15)
defining [ K B ] is of lower degree than that of (16) defining [Kg].Thus the equivalent
transverse load is smoother in the w , formulation.
~ After evaluation of these matrices they are
transformed via the condensation implied by the formulas (8)-(12). The differences in the
resulting condensed geometric stiffness matrices are difficult to trace analytically. Their
performance is illustrated by some examples in the following Section.
EXAMPLES
In this Section the difference in performance using either of the rotation representations in the
geometric stiffness matrix is illustrated through classic plate buckling examples for a
rectangular plate with length I and width b. The thickness t , is uniform, and the material is
isotropic with Young’s modulus E and Poisson’s ratio v. The theoretical buckling stress of
0.85 0.85
2x2 4x4 8x8 16x16 2x2 4x4 8x8 16x16
Element mesh Element mesh
0.90
0.85
i 2x2 4x4 8x8 16x16
0.85
2x2 4x4
ID5 I
8x8 16x16
Element mesh Element mesh
- Free
-
-7 Simply supported
Clamped
Figure 3. Performance of the semi-loof element in compression
SEMI-LOOF ELEMENT 17
0.85
2x2 4x4
Element mesh
8x8 16x16
0.90
O“”
i 2x2 4x4 8x0
Element mesh
18x16
0.95/ ! : T6Li p= 1
0.90
0.85 0.85
T
DL 3
1
2x2 4x4 8x0 16x16 2x2 4x4 8x5 16x16
Element mesh Element mesh
-
~ r r r m
Simply supported
Clamped
Figure 4. Performance of the semi-Loof element for linear normal stress and shear
The buckling coefficient k depends on the aspect ratio p = / / b , the boundary conditions and
the Poisson ratio Y. Theoretical values for the buckling coefficient for the cases investigated
can be found in Timoshenko and Gere.’
Eight different cases have been treated, and a mixture of different loading and boundary
conditions have been considered in order to investigate the accuracy of the two different
stability formulations. In Figures 3 and 4 the results for different uniform element meshes have
been plotted. The results are scaled relative t o the theoretical values from Timoshenko and
Gere.’
18 J . JQNSSON, S. KRENK AND L. DAMKILDE
Buckling coefficient k
Table 11. Comparison with results of Martins and Owen2 and PAFEC
curved), but it is possible to use four or five points of integration. Table I1 indicates that by
using PAFEC with 4-point integration we obtain the same values as Martins and Owen.2
CONCLUSION
The semi-Loof element relies on an elaborate interpolation scheme in which the translation and
the rotation of the normals are represented in terms of two different sets of interpolation
functions. A compact form of these interpolation functions is given, and two different
formulations of the geometric stiffness matrix are investigated. In one of these the destabilizing
effect is represented via the rotation of the tangent plane, while the other makes use of the
rotation of the normals. It is demonstrated by examples that the use of the tangent plane
rotation leads to consistently better results. In spite of this, available codes and results from
the literature indicate that the second, and less accurate, methods is the common choice.
REFERENCES
1. B. M. Irons, ‘The semi-Loof shell element’, Finite Elements for Thin Shells and Curved Members,
D. G. Ashwell and R. G. Gallagher (Eds.), Wiley, 1976.
2. R. A. F. Martins and D. R. J . Owen, ‘Structural instability and natural vibration analysis of thin
arbitrary shells by use of the semi-Loof elements’, Int. j . numer. methods eng., 11, 481-498 (1977).
3. R. A. F. Martins and D. R. J . Owen, ‘Elastoplastic and geometrically nonlinear thin shell analysis
by the semi-Loof element’, Comput. Struct., 13, 505-513 (1981).
4. PAFEC Theoretical Manual, PAFEC Limited, Strelley Hall, Strelley, Nottingham, England, 1984.
5. C. Wang and T. H . H. Pian, ‘Hybrid semiloof element for buckling of thin-walled structures’,
Comput. Struct., 30, 811-816 (1988).
6. H. W. Loof, ‘The economical computation of stiffness of large structural elements’, Assoc. of Int.
Symp. on Use of Comp. in Struct. Eng., University of Newcastle-upon-Tyne, 1966.
7. S. P. Timoshenko and J. M. Gere, Theory of Elastic Stability, 2 edn., McGraw-Hill, New York,
1961.
8. J . Jsnsson, ‘Recursive finite elements for buckling of thin-walled beams’, Ph.D. Thesis, R 263,
Department of Structural Engineering, Technical University of Denmark, 1990.
9. R. Guyan, ‘Recution of stiffness and mass matrices’, AZAA J . , 3, 380 (1965).
10. R. D. Henshell and J. H. Aug, ‘Automatic masters for eigenvalue economisation’, Earthquake eng.
struct. dyn., 3, 375-383 (1975).