Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
1 INTRODUCTION
Membrane structures are very attractive alternatives to span large distances. They are very light, elegant, and effective [1-3]. The material is optimally used since the structures are subjected only to membrane tension stresses. The art of form finding means to find the optimal deflected and finally visual shape due to a given stress distribution acting on the deformed structure. This fact should be clearly pointed out since it is different compared to conventional structural optimization where the undeformed shape is optimized with respect to displacements due to given load cases.
International Journal of Space Structures Vol. 14 No. 2 1999
The problem is very closely related to the determination of minimal surfaces. Since centuries, they are one of the oldest toys of mathematicians. There exists an enormous amount of knowledge of how to describe and to generate them [4]. From the mechanical point of view minimal surfaces are determined by an isotropic stress field that can be experimentally simulated by the soap film analogy. For the engineering application form finding of tensile membrane structures the later approach is most promising since it is directly formulated with respect to stresses which restrict and dominate the final design because of strength resistance and the
131
A General Finite Element Approach to the Form Finding of Tensile Structures by the Updated Reference Strategy
anisotropic material behavior in the directions of weft and warp [5]. Furthermore, an unlimited variation of shapes is generated by prescribing anisotropic stress distributions. Inspired e.g. by the pioneer work of Frei Otto [6] many technical procedures and algorithms had been developed which many of them are based on dynamic relaxation [7,8]. Others, like the force density method are based on special discretization and linearization techniques. Originally, it was developed for the form finding of cable structures [9-11]. Recently the method was extended to triangular membrane elements [12,13]. The existence of all the different methods is explained by the mathematical problem which arises by solving an inverse mechanical problem. It is defined by the prescribed stress distribution as the driving degree of freedom in the design process. This is inverse to standard mechanics where stresses are the structural response to the deformation of material. As a consequence the related numerical solution methods are faced to mathematical singularities which are overcome by several techniques, e.g. by a modified Newton-Raphson iteration [14-16]. The updated reference strategy which will be presented in the following is a further alternative [17-19]. It is consistently derived from continuum mechanics of elastic bodies with respect to large deflections and small strains. The numerical solution follows standard finite element discretization procedures which means that the method can be applied to any triangular or quadrilateral finite element formulation. The above mentioned singularities are regularized by a homotopy mapping [20] which is based on approximations for Cauchy and Piola-Kirchhoff stresses. It will be shown that this approach leads to the force density method if it is applied to one dimensional elastic bodies, e.g. cables. Vice versa the updated reference strategy is the consistent generalization of the force density method applicable for any structure. It can also be used as an approximation technique in structural shape optimization [19].
the intermediate state or actual configuration with actual area a is understood as to be generated by deformation of an initial state or reference configuration of area A. A point P on the actual surface with position vector x (u 1,u 2 ) had originally the position X (u 1,u 2 ) on the reference surface. The point is identified by its surface coordinates u 1,u 2 which stay invariant during deformation. The displacement u of point P is defined as u( u
1
,u
)=x (u
,u
)X(u
,u
(1)
The surface area a of the actual configuration can be written with respect to the reference configuration as a=
da =
detFdA
(2)
where detF is the determinant of the deformation gradient F. Refer to the appendix for further details on differential geometry.
The minimum of (2) is determined by the vanishing variation d a of the surface area with respect to variation d x of the shape. This states the governing equation:
2 THE GOVERNING EQUA TION OF FORM FINDING 2.1 DIRECT GEOMETRICAL APPROACH TO MINIMAL SURFACES
Minimal surfaces are defined as surfaces of minimal area which are enclosed by a given boundary. Consider the procedure to determine a minimal surface as an iterative process (Fig. 1). The surface of
132
d a=d
where
1E 2
a
d F=d gi G i= }
(d x) X
(4)
must have a normal component at any point on the surface, i.e. the condition d xg3 0 must be fulfilled everywhere. If the variation d x would map the surface back on itself the area content would not change even for a non-optimal geometry.
d u,x= }
(d u) x
=}
1 2
(6) inserted into (5) and integration over the reference area A:
=t s :(d F.F1)detFdA=0
A
(7)
(8)
(F(detFF-1s
F-T)):xd
s =s
ab
ga gb
a b
(10) Ga Gb =Sab Ga G b
d w=t s : }
a
( u) d
d x
(5)
which states that the virtual work of a stress field in equilibrium vanishes. s is the prescribed Cauchy stress tensor which acts on the surface in equilibrium, d u,x is the derivative of the virtual displacement with respect to the geometry of the actual surface. The thickness of the membrane is denoted by t. It is comparatively thin and assumed to be constant during deformation, i.e. the Poisson effect in thickness direction is neglected. This is in accordance with the behavior of available membrane material. d u,x can be expressed in terms of the deformation gradient. From (1) it follows that d u = d x and:
The components Sa b of the 2nd Piola-Kirchhoff stress tensor are identical to the components s a b of the Cauchy stress tensor multiplied by detF. However, they are related to the base vectors of the reference configuration whereas the Cauchy stress components are related to the actual configuration. In general, if s is homogeneous, than S is not a vice versa. S and s are identical, if actual and reference configuration are identical. Note that all stress components normal to the surface area are zero, i.e.:
i3=s 3i=Si3=S3i=0
(11)
For further reading on the mechanics of tension structures see e.g. Leonard [22].
2.3 SOAP FILM ANALOGY THE EQUIVALENCE OF GEOMETRICAL AND MECHANICAL APPROACH
Up to here we developed two different equations to define (i) minimal surfaces
133
A General Finite Element Approach to the Form Finding of Tensile Structures by the Updated Reference Strategy
d a=
detFF-T:d FdA=0
(3)
displacement field are piecewise approximated by the interpolation of nodal coordinates or displacements, respectively (Fig. 3): X=^
nel
N k (u x=^
,u
(8)
)Xk ; u=^ w
1,u 2
nel
Nk (u
,u
k=1
) uk ; w (12)
Nk (u
) (Xk +uk ) w w
If an isotropic homogeneous stress field is prescribed, i.e. s = sI is a scalar multiple of the unit tensor I, (8) reduces to (3) and the shape of equilibrium is a minimal surface. This well known fact is called the soap film analogy which reflects the experiment to produce minimal surfaces of thin soap films within a rigid wire. (8) is the theoretical basis of the form finding procedure of membrane structures. The manipulation of s allows for a broad range of structural shapes other than minimal surfaces. It is the governing degree of freedom in the design procedure. On the other hand it reflects the fact that shapes of isotropic homogeneous stress distributions often cannot be build because of technical reasons of manufacturing or that minimal surfaces may be bad alternatives to carry additional load as e.g. wind and snow.
The upper bar denotes nodal values, e.g. Xk is the w reference position vector of the k-th element node. Nk are the standard Co-continuous shape functions, and nel is the number of nodes of one finite element. For the sake of simplicity no difference in notation between real and approximated fields is made. The shape functions of a three node element are
{}
X1 X3 X2 N1=1u N =u
2 1 2 1
=^ N1 /u ; N2 /u N3/u
Nk =1 ;
k=1
{}
X1 w k X3 w k Xk w 2 N1 /u N2 /u N3/u
2
; (13)
2
1 1
=1
=1
=0
N 3= u
1=0
2=1
1 1 N1= } (1+u 1)(1+u 2); N2= } (1u 1)(1+u 4 4 1 1 N3= } (1u 1)(1u 2); N4= } (1+u 1)(1u 4 4
2)
(14)
2)
134
The covariant base vectors are determined straight forward to be: nel X Ga= } a =X,a=^ N,k Xk a w u k=1 (15) nel x ga= } a =x,a=^ N,k (Xk +uk ) w a w u k=1 which gives for the three node triangle G1 =X2 X1 ; g1 =(X2 +u2 ) (X1 +u1 ) w w w w w w
F br
dA+D bs t }
A
bs (21)
F br
dA=0; r,s=1,..,ndof
(16)
Using the standard vocabulary of the finite element method (22.1) F T): } R=t det F(s F dA br A
The nodal displacement components are defined to be the free discrete parameters of form finding. They are arranged in a column matrix b of dimension ndof (number degrees of freedom). Variation of any entity, e.g. the deformation gradient F, means now variation with respect to the free parameters. By use of the chain rule of differentiation we get: (17) F gi i i d F=} d br =d gi G =} d b G ; r=1,...,ndorf br br r and ga=x,a = } g a br br=^
nen
bs
detF(s FT): }
F br
dA
(22.2)
the stiffness matrix. As a consequence of discretization the covariant base vectors gi are only linearly dependent on the parameters b r and, therefore, the second order derivative 2 F 2 gi f G1 is equal to zero. =} } br bs br bs The stiffness matrix of the membrane finite element can now be written as: Krs =t }
A
k=1
N,k } a
uk w br
br; r=l,...,ndorf
(18)
where br is the r-th component of b, i.e. the r-th degree of freedom of the discretized problem. Discretization of (8) yields: F w=br t detF(s FT): } dA=0 (19) A br
bs or in components:
(detF(s xFT)): }
F br
dA
(23)
Krs =t
which has to be fulfilled for any choice of d b and, finally, we arrive at the non-linear system of ndof equations:
E 3 detFs
A
Rr =t detF s
A
ab
(ga.r gb ) dA
ab
ab
(ga.rgb br
.s
)+(detFs
),s (ga.rgb ) dA
(24)
( ),r = }
and ( ),s = }
bs
w br
=t detF(s FT): }
A
F br
dA=0
(20)
Note that even for constant components of stress tensor s the derivative s a b ,s does not vanish since the related covariant base vectors ga are functions of the discretization parameters. For the case of isotropic stress fields (s = s ^I ) or minimal surfaces the expressions reduce to Rr = } Krs= } 2a b bs r a br =ts ^ detFga.r gadA
A
(25)
.s g
135
A General Finite Element Approach to the Form Finding of Tensile Structures by the Updated Reference Strategy
and Krs is identified as the Hessian matrix of the surface area, i.e. the matrix of second order derivatives with respect to the discretization parameters.
be performed explicitly:
d w=s Ac }
X
l1 L l2
s X xAc
l
(gi.rgl)d br
where the last expression of (27) reflects the alternative formulation using the 2nd Piola-Kirchhoff stress X: S L X S=S I=detFF1s X IFT=s X } I ` (28)
Rr = } Krs= }
s X Ac ` `
(29) (gl.rg1) 1
2
s XAc `
(gl.rgl.s) }
(gl.rg1)(gl.sg1) g1gl.s
||g1 || ||G1 ||
where integration is performed along the arc length S of the reference configuration and the cable cross section area Ac is assumed to be constant during deformation, i.e. the Poisson effect is neglected.
We further discretize the cable by simple two-node elements (Fig. 4). By this assumption we can set i g1i = ` , i G1i = L, g11 = 1/` 2, the element length before, L, and after, ` , deformation. Now, the integration can
136
membrane structures. In general, the edges are moving tangentially inwards during the form finding procedure forcing a tangential adjustment of the surface finite element mesh. It is practically impossible to consider this effect by prescribed move directions. We have to find other methods to regularize the problem. General mathematical methods to approach a solution of a singular problem are the methods of numerical continuation [20] also called homotopy methods or path following methods. The idea is to modify the original problem by a related one which fades out as we approach the solution. The methods are used e.g. to determine the non-linear load deflection relations in structural limit state analysis [23] or structural optimization [24]. E.g., consider the problem f(x) min. The objective f(x) is assumed to be singular at the solution; the standard methods of optimization fail. However, a related problem f(x) min min is known which is not singular. The function f(x) can be used to regularize the original problem by mapping with the continuation factor l : (30) min fl (x)=min[l f (x)+(1l )f (x)] The solution of (30) approaches the solution of the original problem as we trace l from 0 to 1. The factor l is identified as the step length parameter of a curve in the extended parameter space (x, l ). The method is the more successful as the related function f(x) is closer to the original function f(x). Applied to the problem of form finding of membranes and minimal surfaces we can use the alternative formulation of the virtual work (9) by use of the 2nd Piola-Kirchhoff stress tensor S as a related problem. The difference to the original problem is that we now assume the 2nd Piola-Kirchhoff stresses to be given instead of the Cauchy stresses s . The modified stationary condition states as:
Thus the Hessian of the modified problem has the form: K(l
)rs=l
t (detF FT),s:F,rdA s
A
+(ll )t (F,sS):F,rdA
A
(32)
.sdA
(33)
is positive definite if S is positive definite, which is the case for tension stresses. As a consequence, the total stiffness matrix is also positive definite, if l is small enough. The quality of the approximation improves as the reference configuration gets closer to the optimal shape. Note, that for given S (33) is a constant expression since the covariant base vectors ga are only linear dependent on the discretization parameters br. In non-linear structural analysis (33) is also known as geometrical stiffness or initial stress matrix. X For an isotropic stress distribution (S = S I) or minimal surfaces (33) reduces to:
.sdA
(34)
+(1l )t (FS):F,rdA
A
(35)
5.2 APPROACHING THE OPTIMUM THE UPDA TED REFERENCE STRA E G Y T (URS)
Several strategies are suggested to trace the continuation factor l towards the optimal solution: (i) Adaptation of l by trial and error.
(31)
nd
For the modified problem Cauchy as well as 2 Piola-Kirchhoff stresses are prescribed. As the 2nd Piola-Kirchhoff stresses are defined in terms of the fixed reference configuration, S remains constant if the shape is modified; the derivatives with respect to the discretization parameters vanish: S,r = 0.
The continuation factor l is set constant and small enough to stabilize the problem. Trace l experimentally from 0 to 1 in series of analyses until the solution fails. Take the last solution of the modified problem (31) as optimal solution. This procedure is simple, however, not very satisfactory in particular for large problems.
137
A General Finite Element Approach to the Form Finding of Tensile Structures by the Updated Reference Strategy
(ii)
Optimization of l methods.
by extended system
A solution of the singular problem can directly be determined by so called extended system methods. These methods have recently been used to determine buckling or limit loads in the context of non-linear structural analysis [24-26]. A detailed list of related literature is e.g. given by Reitinger [24]. Usually the methods consider critical points with one unique zero eigenmode. The basic idea is to extend the singular problem by additional equations which state the singularity condition and a normalization of the related eigenvector. After discretization the extended system of equations displays as: R(l )r(b,l ) K(l )rs(b,l )fs l(f ) =0 ; stationary condition =0 ; singularity condition (36) =0 ; normalization of eigenvectorf
linear and is solved within one step for the cost of additional approximation steps. In practice one will start with a sufficiently small value, e.g. l 0.5, and will enlarge l for the next steps to improve convergence. The procedure is absolutely robust and gives reliable results very fast. The amount of training is negligible.
The singularity condition is enhanced by an additional penalty term to be able to solve (36). Several exact and approximate penalty formulations are known in literature. Exact formulations fail because the original problem is multiple singular. An approximate formulation given by Reitinger might instead be applied which underestimates the correct solution and gives save approximations: R(l
)r(b,l
=0 =0 =0 (37)
(38)
And for the stiffness with respect to the discretization parameter u: =q (39) Ac L ` Finally, we arrive at the expression q which is well known as force density and relates the prescribed cable force n to the deformed length ` of the member. The force density q may be interpreted as a 2nd PiolaKirchhoff force. It is usually set constant during calculation which yields linear systems of equations. Finally, the force density method appears to be a consistent part of the more general theory of the updated reference strategy.
L LL
K=Ac (F,uS):Fuds=Ac }
1` 2
}
Ln
} ds= }
Although theoretical fascinating the extended system methods may appear in this context as not stable enough for routine usage. (iii) The Updated Reference Strategy (URS) The solution of the modified problem (31) with any suitable choice of l is used as a new improved reference state for the next approximation. The reference configuration is iteratively adapted towards the optimal solution. If the differences of subsequent solutions are small enough to be accepted the series of approximations is terminated. The strategy converges to the optimum for any value of the continuation factor l which is small enough to ensure the regularity of the Hessian matrix. For the special case l = 0 (31) is even
138
changing surface during all stages of deformation. Load of that type is therefore often called follower load. Because of that, follower loads contribute to the stiffness matrix. It depends on the boundary conditions whether this load is conservative or not [27], i.e. the stiffness matrix is symmetrical or not, respectively. The equation of virtual work (5) is enhanced by the pressure term:
u=x= }
x br
x u
a
=x,a,
pd uda=d br
E E
u
X (x, 3 x, )x, du p 1 2 r
du
(43)
d w=t s :d u,xda
a
Linearization gives:
pd u da=0
(40)
lin( pd uda)=d br
a
E E
u
1u
X (x, 3 x, )x, du p 1 2 r
1d u
The geometrical equivalence are surfaces of constant positive mean curvature as a generalization of minimal surfaces which are defined by zero mean curvature.
+d brD bs
E E
u
(44)
1u
du
and we can identify the additional contributions R(p)r and K(p)rs to unbalanced force vector and stiffness matrix R(p)r= K(p)rs
E E =E E
u u
1u 1u
X (x, 3 x, )x, du p 1 2 r
du
2 1
E E
u
du
X (x, 3 x, ), x, du p 1 2 s r
1d u
(45)
which have to be added to (24) to yield the total expressions. In the case of standard displacement elements the second term of K(p)rs vanishes. This is because the geometry x is only linearly dependent on the discretization parameters b r and the second derivatives of x with respect to br are, therefore, zero. Making use of (a 3 b) c = (b 3 a) c we integrate (45) by parts and get after some rearrangement: K(p)rs=X p =X p
Given the pressure magnitude X, the pressure vector p acts in the direction of the surface normal (Fig. 5):
X p=p g3;g3= }
=}
g13 g2 j
E E
u u
1
1 u
(x,13 x,2),sx,r du
(41)
X p
E E
u
X 1p
1d u
(x,s3 x,2)x,r du
1d u 2
1u
(x,13 x,2),rx,s du
(46)
1
The virtual work of the pressure load can now be written as:
Adding K(p)rs = X e u 1 e u 2 (x,13 x,2),sx,rdu p both sides and dividing by 2 gives finally:
du
on
pd u da=
E E
E E
u 1 u
jpd udu
1du 1d u 2
(47) (42)
u 1 u
X (g 3 g )d udu p 1 2 2
Considering the discretization of the displacement u in terms of the parameters br and the definition (49) of the base vectors ga, i.e.
Obviously, the domain integral of (47) is symmetrical. The boundary integrals vanish at finite element interfaces if adjacent elements share common
139
A General Finite Element Approach to the Form Finding of Tensile Structures by the Updated Reference Strategy
tangent base vectors and common displacement derivatives u,r. That is the case for conforming elements as e.g. for the chosen displacement elements. Thus, the boundary conditions decide on the symmetry of the overall stiffness matrix and the conservative or non-conservative character of the pressure loading. The boundary integrals vanish if the displacement is prescribed at the structures boundary (u,r ; 0) or if (g1 3 u,s)u,r ; 0 or (u,s3 g2)u,r ; 0 at the supported edge. The latter conditions reflect the situation that at every point on the boundary the displacement derivatives u,r and the tangent base vector share a common plane, e.g. support in normal direction. The typical example of an inflated bubble which is fixed around a hole where the filling media comes through is a conservative problem with symmetrical stiffness matrix.
The shape of a catenoid is determined starting from a cylindrical initial shape of diameter 2 and height 1.2 (Fig. 6). All three solution strategies have been applied and the numerical solutions are compared with the analytical result. One eighth of the shape was discretized by 144 4-noded isoparametric elements. The best approximation of shape was achieved by the updated reference strategy. The comparison of surface area is misleading because of the discretization error. Even for strategy trial and error the principal stresses are almost homogeneous. Other extended system methods than Reitingers failed. strategy analytical Trial and error URS ext. system Surface area 7.0 6.998 6.993 6.993 no. of time steps
l
0.85 0.8 0.8 0.979
1 5 1+ ext. system
140
shape of a discretized sphere. The problem is modelled by 80 4-node iso-parametric elements. The continuation factor is 0.95 throughout the whole calculation.
7.4 TENTS
Several student projects [28-30] of membranes with cable reinforced free edges are shown to give an impression of the variety of applications, Fig. 9. Some of them are related to existing structures as e.g. the Tanzbrunnen by Frei Otto.
141
A General Finite Element Approach to the Form Finding of Tensile Structures by the Updated Reference Strategy
142
9 REFERENCES
1. 2. 3. 4.
Otto, F. and Rasch, B., Finding Form, Deutscher, Werkbund Bayern, Edition A. Menges, 1995. Schock, H.-J., Segel, Folien und Membrane, Birkhuser, Basel, 1997. Berger, H., Light structures structures of light, Birkhuser, Basel, 1996. Dierkes, U., Hildebrandt, S., Kster, A., Wohlrab, O., Minimal surfaces I and II. Grundlehren der mathematischen Wissenschaften 295-296, Springer, Berlin, Heidelberg, 1992. Suzuki, T., Hangai, Y., Shape analysis of differently stressed surfaces by the finite element method, in: Srivastava, N. K., Sherbourne, A. N. and Rooda, J., eds., Innovative Large Span Structures (vol. 2): Proc. IASS-CSCS International Congress, Toronto, Canada, 1991, 400-411. Otto, F. et. al., Zugbeanspruchte Konstruktionen, Vols. I and II, Ullstein Verlag, Franfurt, Berlin, 1962, 1966. Barnes M., Form and stress engineering of tension structures, Structural Engineering Review, 6, 1994, 175202. Lewis, W. J., Lewis, T. S., Application of Formian and dynamic relaxation to the form finding of minimal surfaces, Journal of the IASS, 37(3), 1996, 165-186. Linkwitz, K. and Schek, H.-J., Einige Bemerkungen zur Berechnung von vorgespannten Seilnetzkonstruktionen, Ingenieur-Archiv, 40, 1971, 145-158 Schek, H.-J., The force density method for form finding and computations of general networks, Computer Methods in Applied Mechanics and Engineering, 3, 1974, 115-134.
8 CONCLUSIONS
Form finding, i.e. the problem to determine the shape of equilibrium of a stress field with given intensity but unknown orientation, is a challenging esthetical as well as analytical and numerical task. Classical methods of consistent linearization fail because the problem is singular at the solution. The updated reference method overcomes the difficulty by a rigorous algorithm which is based on differential geometry, continuum mechanics and numerical continuation. Internal pressure is implemented considering the interaction of shape and load direction and the consequence for the non-linear characteristics of form finding. The method can be applied to the form finding of membranes and cable structures using any finite element discretization scheme, e.g. two-, three- or four-node elements or even elements of higher order. It is shown that the force density method for cable nets is a special case of the presented method. Several typical examples illustrate accuracy, behavior, and applicability of the method.
5.
6. 7.
8.
9.
10.
143
A General Finite Element Approach to the Form Finding of Tensile Structures by the Updated Reference Strategy
11.
Linkwitz, K., Least squares methods in non-linear form finding and analysis of pre-stressed and hanging nets, in: Non-Linear Analysis and Design for Shell and Spatial Structures: Proc. SEIKEN-IASS Symposium, Tokio, 1993, 601-610. Singer, P., Die Berechnung von Minimalflchen, Seifenblasen, Membrane und Pneus aus geodtischer Sicht, PhD Thesis, University of Stuttgart, 1995. Maurin, B. and Motro, R., The surface stress density method as a form-finding tool for tensile membranes, Engineering Structures, 20(8), 1998, 712-719. Haug, E. and Powell, G. H., Finite element analysis of nonlinear membrane structures, Report UCSESM 72-7, University of California at Berkeley, 1972. Haug, E. and Powell, G. H., Finite element analysis of nonlinear membrane structures, in: Tension and space structures (vol.2): Proc. 1971 IASS Pacific Symposium, Tokio and Kyoto, 1972, 165-175. Haber, R. and Abel, J., Initial equilibrium solution methods for cable reinforced membranes, Computer Methods in Applied Mechanics and Engineering, 30, 1982, 263-284. Bletzinger, K.-U., Form finding of membrane structures and minimal surfaces, in: Rozvany, G. and Ohlhoff, N., eds., Proc. 1st World Congress of Structural and Multidisciplinary Optimization, Goslar, Germany, 1995, 563-568. Bletzinger, K.-U., Form finding of tensile structures by the updated reference strategy, in: Chilton, J., Lewis, W. et. al., eds., Structural Morphology Towards the New Millennium, Proc. IASS International Colloquium, August 15-17, 1997, University of Nottingham, UK, 1997. Bletzinger, K.-U., Shape optimization by homotopy methods with special application to membrane structures, in: Proc. 6th AIAA/NASA/ISSMO Symposium on Multidisciplinary Analysis and Optimization, Bellevue, Washington, USA, 1996, 122-129. Allgower, E. L. and George, K., Numerical Continuation Methods, Springer Series in Computationa l Mathematics, Springer, Berlin, Heidelberg, New York, 1990. Kreyszig., E., Differentialgeometrie, Akademische Verlagsgesellschaft, Leipzig, 1968. Leonard, J. W., Tension Structures, McGraw-Hill, New York, 1988 Ramm, E., The Riks/Wempner approach An extension of the displacement control method in non-linear analysis, in: Hinton, E., Owen, D. R. J., and Taylor, C., eds. recent advances in non-linear computational mechanis, Pineridge Press, UK, 1982, 63-86.
24.
Reitinger, R., Bletzinger, K.-U., and Ramm, E., Shape optimization of buckling sensitive structures, Computing Systems in Engineering, 5, 65-75, 1993. Wagner, W., Zur Behandlung von Stabilittsproblemen der Elastostatik mit der Methode der finiten Elemente, Habilitationsschrift, Bericht-Nr. F 91/1, Institut fr Baumechanik und Numerische Mechanik, Universitt Hannover, 1991. Wriggers, P. and Simo, J. C., A general procedure for the direct computation of turning and bifurcation points, International Journal Numerical Methods in Engineering, 30, 1990, 155-176. Schweizerhof, K., Nichtlineare Berechnung von Tragwerken unter verformungsabhngiger Belastung mit finiten Elementen, PhD. Thesis, University of Stuttgart, 1982. Wahl, C., Numerische Formfindung von vorgespannten Seil- und Flchentragwerken, Diploma Thesis, Institut fr Baustatik, University of Stuttgart, 1995. Krapf, A., Formfindung und Statik vorgespannter Membrantragwerke, Diploma Thesis, Institut fr Baustatik, University of Stuttgart, 1996. Seebacher, F., Formfindung, Zuschnitt und Statik vorgespannter, textiler Membrankonstruktionen, Diploma Thesis, Institut fr Baustatik, University of Karlsruhe, 1998.
25.
12.
13.
26.
14. 15.
27.
28.
16.
29.
17.
30.
18.
APPENDIX
The basis of differential geometry will be briefly displayed [21]. The position vectors x and X to a point P on a spatially curved surface in the actual and reference configuration, respectively, are vector functions of the surface parameters u 1 and u 2: x=x(u
1,u 2);
19.
20.
X=X(u
1,u
2)
(48)
as:
g1= }
; g2= } 1
and G1}
X u
1
=G2}
X u
2
(49)
The covariant base vectors are tangential to the corresponding coordinate lines, e.g. g1 is tangential to the coordinate line u 1 = const. in the actual configuration. Covariant base vectors are in general not orthogonal and are not of unit length. Their scalar
144
products gij the components of the covariant metric tensor reflect the metric of the surface, i.e. the length of the covariant base vectors and the angle between them: g11=g1g1 1; g22=g2g2 1; g12=g21=g1g2 0 (50) The surface normal is determined by g3 which is defined by the normalized cross product of g1 and g2: g3= } g13 g2 ||g13 g2|| ;||g3||=1 and G3= } G13 G2 ||G13 G 2||
vector parallelogram which is given by the covariant base vectors g1 and g2. The area content is determined as da=||g13 g2||=(g13 g2)g3=det(gij)=j and dA=||G13 G2||=(G13 G2)G 3=det(Gij)=J and The total area a of a surface in the actual configuration can be expressed in terms of the surface coordinates as a=
(56)
;||G3||=1 (51)
E E
u 1u
jdu
du
E E
u 1u 2
||g13 g2||du
du
(57)
Th contravariant base vectors g i are the dual of the covariant base vectors g j. They are defined by the rule g igj=d
i j
and GiGj=d
i j
(52)
where d ji is the Kronecker delta. Any vector or tensor can be written with respect to coor covariant base vectors gi, Gi or gi, Gi: a=aigi= iGi=aig i= iGi; (53) ~ ~ i T=T ijg gj=T ijGi Gj=Tij g i g j=T ij G Gj i The unit tensor I is defined in terms of the co- or covariant base vectors as (54) I=GijGi G j=G ijGi G j=gijg i g j=g ijg gj i means the tensor product and we make use of Einsteins summation convention: a=a igi=^
3
The reference configuration is transformed into the actual configuration by the deformation gradient F: x F= } =g G i; FT=G i gi; X i (58) X F1= } =Gi g i; FT=gi Gi x The deformation gradient F transforms base vectors by gi=FGi=gj GjGi=d Gi=FTgi=Gj g jgi=d g i=FTGi=g j G jGi=d G i=F Tg i=G j gjg i=d and the differential area by da=j=||g13 g2||=detFdA=detFJ=detF||G13 G2|| j ||g13 g2|| } =detF= } = } dA J ||G13 G2||
j i gj=gi j i Gj=G i j i j g =g j G j=Gi i
(59)
i=1
a=1
ba g a
(55)
da
(60)
Note, that Latin dummy indices run from 1 to 3 and Greek from 1 to 2. Note also that covariant components, e.g. ai, go together with contravariant base vectors gi and vice versa. The differential piece of area da is defined as the
The surface area of the actual configuration can be written with respect to the reference configuration
145