Professional Documents
Culture Documents
00
Printed in Great Britain. 0 1988 Civil-Comp Ltd and Pergamon Press plc
Abstract-A solution strategy for computing surfaces based on minimal area calculation is described which
shows a number of advantages compared with other strategies. It is derived from a discrete network type
formulation of the shape and its properties. Any air inflated structure might be formulated in this approach
and the definition of the boundary of the surface is not limited to fixed geometrical values but also allows for
free boundaries defined by cable forces.
679
680 L. GR~~NDIG
determined by minimizing the area, because the topo- triangles allow for a sufficiently close description of the
logical type of the area is strictly defined by the surface:
relations of proximity. From these properties it is
possible to arrive at a powerful instrument for cal- S=xA,=minVI. (2)
culating minimal surfaces and therefore for finding
the shape of structural membranes and for air inflated Here A, represents the area of the triangle p. The area
structures. of the triangle is a function of the coordinates of the
edge points of the triangles.
MATHEMATICAL APPROACHES FOR CALCULATING
Equation (2) can be formulated in a quadratic
MINIMAL SURFACES form using & = JAp and arranging f, in the vector
fapplying the scalar product
Mathematical approaches for minimal surfaces
normally start with the differential eqns (1). Observing S =frf = min. (3)
the boundary conditions a solution of the differential
equations is searched for, with the exception of very Equation (3) is a least squares approach according to
special solutions a closed formula for the solution Gauss. The following conditions, being necessary for
surface cannot be obtained. Therefore the differential the minimizing of eqn (3) are valid with f=f(x)
equations have to be discretized and solved for using and xr=[x,y,z ,..., x,y, z], the coordinates of the
proper points of support. edge points of all triangles arranged in a vector x
In the approach of [2] a discrete triangular facet of coordinates
area is used. The surface area is described as the sum of
the individual areas of the triangles. The areas of the
triangles are functions of the positions of the edge
points of the triangles with the coordinates x, y, z.
0g;=o. (4)
Keeping x and y as constant z is allowed to vary in such As the area is subdivided into m triangles and the
a way that the gradient of the sum of the surface areas number of triangles is smaller than the number of edge
around every point of the surface will disappear and points, eqn (4) cannot be solved in the usual way by
will become zero. According to [2] this strategy is keeping aflax fixed and using Taylor expansion only
restricted to surfaces with a geometrically defined forf. The resulting system of equations of the following
boundary which has to be planar in case of volume kind would become singular:
restrictions.
Several other strategies [3,4] start from a finite ($r($dx = -(;rj& (5)
element description of the surface and describe the
minimal surface in their physical meaning as an area
with constant stress. The system of eqns (5) does not allow for a solution of
the adjustment problem, because an essential property
The global strategy described in the following starts,
of minimal surfaces has been disregarded, namely the
like the strategy of [2], from a discrete description of the
area with points and triangles connecting the points. minimal area being positive and significantly different
The restrictions of the approach of [2], however, from zero. All triangular surfaces will become positive
are not valid any more and any boundary curves are if the surface is obtained by stretching it into a given
boundary frame and these surfaces will be significantly
allowed (Fig. 1).
larger than zero. Newton’s method is applied in a strict
way starting from the nonlinear system of eqns (4) and
THE APPROACH OF MINIMIZATION OF
applying the chain rule of differentiation the following
TRIANGULAR AREAS
system of equations results:
As a minimal surface structure reflects the minimum
area of all the triangles between neighbouring points
on the surface, the following condition is valid if the +‘(t%-)d
=_(“b
ax I/“=cons, ax
NII N22
Fig. 2. Triangle p with corner points i,j, k and edge-lengths u = cx. 00)
a, b, c.
The coordinate difference u, = xi - x, in the triangle
The formation of the derivatives proves to be a p leads to the general definition of the elements of
tedious exercise in general and Schwenkel[2] decided in the pth row of C, which is a branch-node matrix of
his approach to determine the derivatives by numerical the network
differentiation. This is not true if the following ap-
proach is to be used. C(m, n) = 1, if node i = n in the triangle p
Because the triangles can be arranged in any
position in space a representation of the area is useful C(m,n)= -1, ifnodek=ninthetrianglep (11)
which does not favour any of the coordinates of the
corner position. Heron’s formula allows for this. For C(m, n) = 0 elsewhere.
any given triangle with the edge-lengths a, b, c and the
sum value 2s = a + b + c, the area can be expressed by C is defined completely allowing p to represent
(Fig. 2) any triangle of the structure. Because there are three
types of coordinate differences in any triangle accord-
A, = ,/(s(s - n)(s - b)(s - c)). (7) ing to (9), three matrices C,, C,, C, are required to
describe all the existing coordinate differences in the
The sides of the triangles are functions of the
whole structure
coordinates of the comer points. Therefore the areas
of the triangles can be expressed in any geometrical
111= c,x
position using the same formula. Applying formulas
(4) and (6) fp = ,/A has to be formed and the Jaco-
bian matrices aflax and N,,, Nr2 will be needed. The
lA2= c,x (12)
Taylor expansion forfleads to so-called ‘error equa-
tions’ for all triangles, and v represents the square uj = c,x.
root of the residual of the triangular area.
Now arranging the ratios s,Ju,sb~b,sc/c in three dia-
gonal matrices L,,Lr,Lr of rank m (the number of
f= g dx+f,. (8) triangles), and arranging the coordinate differences
0
ui, ur, uj and the square roots of the triangular areas
The pth column of the Jacobian matrix aflax only A in diagonal matrices as well, results in
contains elements in the rows defined by the coordi-
nates of the points i,j, k. These elements are obtained u = A(L,U,C, + L.,U,C, + L,U,C,)dk +f: (13)
by ordinary differentiation and result in the following
expression: The Jacobian matrix aflax therefore is
A(L,U,C, + L21J2C2+ L,U,C,) and (4) becomes
up = 2 (xi - xk) (dxi - dx&
E ;= (CIUIL, + C*U,L, + C,U,L,)Af= 0. (14)
0
sb = l/s + l/(s - a) - l/(s -b) + I/@ -c) with er=[l,l,..., 1, 1], A any diagonal matrix and
a its corresponding vector s. If a, b are vectors and
S, = l/S + l/(S - a) + l/(S - b) - l/(S -c). A, B their corresponding diagonal matrices according
682 L. GR~~NDIG
Fig. 5. Minimal surface with fixed outer boundary and fixed Fig. 6. Minimal surface with variation of the position of the
inner ring. inner ring.
Fig. 7. Minimal surface with fixed central position. Fig. 8. Minimal surface with elevated central position.
Fig. 9. Minimal surface subjected to inner pressure. Fig. 10. Minimal surface with boundary partly relaxed.
complex roof structure which was built in West 5. H.-J. Schek, The force density method for form finding
Germany as a wooden roof for a Health Spa. The and computation of general networks. Comput. Meth.
appl. Mech. Engng3,115-134 (1974).
minimal surface was a starting situation for the actual 6. L. Grtlndig, Die Berechnung von vorgespannten Seil-
geometry of the surface. The surface was modified netzen und Hiingenetzen unter Beriicksichtigung ihrer
adapting architectural wishes and constructural topologischen und physikalischen Eigenschaften und der
needs [9]. The structure lines were materialized with Ausgleichungsrechnung. DGK, Reihe C, Nr. 216, 1976
und Mitteilungen des SFB 64 ‘Weitgespannte Fliichen-
wooden beams being doubly curved (limber).
tragwerke’, Nr 34 (1976).
7. L. Grilndig, The ‘FORCE-DENSITY’ approach and
numerical methods for the calculation of networks. In
REFERENCES Proc. of 3. Intern. Symposium ‘Weitgespannte Flachen-
tragwerke’, Stuttgart (1985).
J. Nitsche, Vorlesungen iiber MinimalJliichen. Springer, 8. K. Linkwitz, L. Griindig, U. Hangleiter and J. Bahndorf,
Berlin (1975). Mathmatisch-numerische Methoden der Netzberech-
D. Schwenkel, Form studies relating to air-supported nung. Mitteilungen des SFB 64 ‘Weitgespannte Fliichen-
structures by minimization of element surface areas. tragwerke’, Nr 72.
CAD 15, No. I (1983). 9. L. Griindig and J. Bahndorf, Formfinding of a roof
E. Haug and J. Gelbermann, Numerical design and structure for a Health Spa. First Int. Conference
analysis on air supported structures. IASS International on Lightweight Structures in Architecture, Sydney,
Symposium on Air Supported Structures, Vendig (1977). Australia (1986).
D. S. Wakefield, Tensyl: An integrated CAD approach 10. L. Grtlndig, J. Bahndorf and M. Neureither, Mini-
to stressed membrane structures. Proc. of the Second malfl&hen zur Formfindung von Membranen. ARCUS,
Int. Conference. Civil-Camp 85. Munchen, Heft 2 (1987).