In-extensional deformation Figure 41a shows the deformation of a simply supported plastic spherical cap loaded by a force.

This deformation is called extensional deformation because the shell middle surface is stretching due to the loading. The loading is carried mostly by normal forces and only a little by bending moments. The person applying the load experiences that the shell behaviour is quite stiff. Figure 41b shows the deformation of the same spherical cap but this time free from its supports. This deformation is called in-extensional deformation because it does not involve stretching of the middle surface. The loading is carried mostly by bending and only a little by normal forces. The person applying the load experiences that the shell behaviour is not stiff at all.

Figure 41a Extensional deformation

Figure 41b In-extensional deformation

Suppose that a shell roof is loaded by snow. If it deforms in-extensionally the displacements are very large and the bending stresses are very large. Clearly, thin shells need to be designed such that in-extensional deformations do not occur for any imposed force. However, in-extensional deformation gives small stresses when a displacement is imposed, for example a foundation settlement. If the deformation response to a foundation settlement would be extensional the stresses would be very large. Therefore, shells need to be designed such that in-extensional deformations occur for imposed displacements. Rijswijk shell roof 1 In the city of Rijswijk (ZH) in the Netherlands a reinforced concrete shell roof was build for a factory. The shell consisted of several half cylinders that continued over three supports (Fig. 42). Due to the heavy materials stored in the factory the foundation started settling and some columns were pulled down more than others. In the lateral direction the shell followed the deformations beautifully in an in-extensional way. However, in the axial direction the shell was very stiff. Apparently large membrane stresses occurred because large cracks were clearly visible in the shell near the settled columns. After a few years already, the building needed to be demolished due to excessive maintenance costs. The conclusion is that cylinder shell roofs should not be built continuous over more than two supports.

lateral cross-section
1

axial cross-section

Told by Mr. H. van Koten in a guest lecture in May 2007.

Figure 42. Extensional deformation of a shell roof in the city of Rijswijk Liquid storage tanks In Rotterdam port there are many liquid storage tanks. The tanks are welded out of 10 mm thick steel plates. The bottom steel plate is supported by square concrete plates that are simply placed onto the soil. Some of these tanks have a roof that floats onto the liquid. This is to prevent build up of explosive gases in half filled tanks. Unfortunately, many tank roofs get stuck against the tank walls after just a few years of operation. It appeared that some concrete plates settle more than at others. Therefore, the steel bottom plate curves and the tank wall deforms (Fig. 43). Small settlements can cause surprisingly large wall deformations. This deformation is in fact in-extensional. For the tank itself this is good because the steel stresses are small despite the large deformations. Unfortunately, as a consequence the roof gets stuck. Clearly, a floating tank roof needs be designed with a large clearance to the tank wall.

x
y

z Figure 43. In-extensional deformation of a storage tank without roof [1]

Analysis of the liquid storage tank The in-extensional deformation of a liquid storage tank can be analysed by hand. Somebody found out that the deformation is described by
2y , a 2y x u y = 2 w sin , a a 2y x u z = 4 w cos , a a u x = w cos

where a is the tank radius and w is the vertical displacement of Q .This can be checked by substituting these equations in the shell membrane equations.
xx= u x, x k x u z= 0, yy= u y , y k y u z= 0, xy= u x, y + u y , x 2k xy u z= 0,

1 where has been used that = k x k= xy 0 and k y = . Apparently, all strains of the middle a surface are zero, therefore, the described deformation is in-extensional.

The horizontal displacement at P (x = l, y = 0) is u z =

4 wl . a

Viking ship Oseberg Viking ships are known to be very flexible [2]. This has two causes. 1) The planks of a Viking ship are joint by iron rivets (Fig. 44b). This is called clinkerbuilding (Fig. 44c). The planks form a shell which can move in-extensionally. The motion is somewhat controlled by curved members called timbers and horizontal members called (Fig. 44c) 2) The Vikings had no saws to cut timbers. (In those days, manufacturing thin steel plate was difficult.) They used axes for cutting timber. To make ship members they split the timber along the grain. Timber cut along the grain (by axe) is much stronger than timber cut through the grain (by saw). Therefore, each Viking ship member was strong, light and flexible. It is not clear whether the Vikings liked their ships to be very flexible. The flexibility was just a consequence of planks joint to a shell and light timber cut along the grain.2

Figure 44a. Viking longship Oseberg, Norway, 800 AD, 21.58 m long 5.10 m wide [2]

Figure 44b. Rivets around an oar hole

Note that steel ships cannot be flexible. They would suffer from fatigue.

Figure 44c.Parts of a Viking ship Spotting in-extensional deformation In-extensional deformation and extensional deformation can occur together. For example, consider a cylinder that is loaded in the axial direction (Fig. 44). This loading will compress the middle surface and cause extensional deformation. On the other hand, a lateral loading on this cylinder will cause mainly bending and the deformation will be in-extensional. When both loads are applied together the deformation will consist of an extensional and an inextensional part.

Figure 44. Extensional and in-extensional deformation of an open cylinder Vibration mode shapes One way of spotting in-extensional deformation is to compute the natural frequencies of a shell structure. If in-extensional deformation is possible this mode will have the smallest natural frequency. In most well designed shells the modes shapes are local deformations. Inextensional deformations, on the other hand, typically are deformations that involve a large part of a shell. This approach does not work for spotting in-extensional deformation due to support settlements. Strain energy Another way of spotting in-extensional deformation is observing the strain energy in a shell. The membrane strain energy in an elementary shell part is
Esm n + 1n + 1n . = 1 2 xx xx 2 xy xy 2 yy yy

The bending strain energy is

Esb = 1 m + 1m + 1m + 1q + 1q . 2 xx xx 2 xy xy 2 yy yy 2 x xz 2 y yz

In this it is assumed that the material behaviour is elastic. The strains and curvatures are those of the middle surface. Note that strain energy does not have a direction and is always positive. A ratio can be defined as
E Esb = sm Esm + Esb

A contour plot of over the shell shows where membrane action is dominant (0 < < 1) and where bending action is dominant (1 < < 0). Dominant bending action is a clear sign of inextensional deformation. Unfortunately, most structural analysis programs cannot plot this quantity. Theorema egregium The shell compatibility equation reads
xx, yy + xy , xy yy , xx = k y xx + k xy xy k x yy

The left hand side represents membrane deformation. The right hand side represents bending deformation. The right hand side is equal to the increase of the Gaussian curvature kG during loading. This is proved in Appendix 5. Studying the compatibility equation we see that when the deformation is in-extensional the Gaussian curvature does not change. This property was discovered by the mathematician Carl Gau in 1827 [3]. It is called Theorema Egregium, which is Latin for remarkable theorem. Gau formulated it as If a curved surface is developed upon any other surface whatever, the measure of curvature in each point remains unchanged. (translated from Latin) It is true for small, large and very large deformations. Studying the compatibility equation we see that the reversed statement does not need to be true. In other words, from observing no change in the Gaussian curvature we cannot mathematically conclude that the deformation is in-extensional (Fig. 44a). change in kG no change in kG in-ext.

Figure 44a: Venn diagram of local shell behaviour Shell behaving like a plate We have seen that in-extensional deformation can be difficult to detect because often it is mixed up with extensional deformation. It appears that the Gaussian curvature itself is a better way of detecting ill-conditioned shells. When the Gaussian curvature does not change during loading perpendicular to the surface then the load is carried in bending only. Proof: The increase of the Gaussian curvature can be written as
k y xx + k xy xy k x yy = kx 2u z y 2 2k xy 2u z 2u z + ky = u z . xy x 2

The reduced shell differential equation is

E t3 12(1 )
2

2 2 2 2u z + E t u z = 2 2 p z .

When the Gaussian curvature does not change then

u z = 0.

If this condition is fulfilled over some shell area then the differential equation in this area reduces to
E t3 12(1 )
2

2 2u z =p z ,

which is the differential equation of plates loaded in bending. Though the shell is curved, the load p z will be carried by bending moments and not by normal forces. Q.E.D. This proof can be easily extended to the Sanders-Koiter equations. Shell design 1 Suppose we are designing a thin shell and part of its area has very large stresses (1). We increase the thickness and this reduces the stresses. However, we find that for acceptable stresses the thickness needs to be very large (2). The large stresses seem to be caused by large bending moments (3) and not by normal forces. Now we have observed three symptoms of inextensional deformation and we think that this might be the problem. We plot the increase of the Gaussian curvature for each load case. It appears that for one of the load cases the increase of the Gaussian curvature is almost zero in the problem area. Now we know for sure that our shell suffers from in-extensional deformation. A solution can be to add a stiff beam to the shell edge. Plotting Gaussian curvature Unfortunately, most finite element programs cannot make a contour plot of the Gaussian curvature or the increase of the Gaussian curvature. Programmers need to implement this, which is not an easy task. Only for high accuracy elements the Gaussian curvature can be computed from information available within an element (see shell finite elements). Kresge Auditorium Kresge Auditorium is a building at MIT campus in Cambridge (close to Boston, USA) (Fig. 44b). It was completed in 1955. Its shape is spherical with three edge beams and three point supports. The reinforced concrete edge-beams prevent in-extensional deformation of the reinforced concrete shell. The edge beams cause edge disturbances in the shell. The height is 15 m. The span between two supports is 48 m. The shell thickness is 90 mm. The architect is Eero Saarinen. The engineering firm is Ammann & Whitney. The contractor is the George A. Fuller Company. The money was donated by Sebastian Kresge ($1.5 million). In the original design the curtain walls were horizontally supported by the edge beams. In the vertical direction the curtain walls were self-supporting with an expansion joint to the edge beam. However, after removal of the timber formwork much creep occurred in the concrete (more than 130 mm deflection). Therefore, the curtain walls were redesigned to also vertically support the edge beams [3a].

The current roof cover dates from 1980. It consists of copper sheets. Earlier roof covers were made of plastic applied as a liquid (lasted 8 years) and soldered lead sheets (lasted 15 years) [3a]. They cracked due to temperature deformation of the roof. (In the Boston climate half a roof can be covered in snow while the other half is heated by the sun.) The cracks and lack of ventilation made the concrete wet. Corrosion and freezing severely damaged the concrete. Extensive and costly repairs have taken place, including replacing large parts of the edge beams. If the building would not be architecturally important it would have been replaced a long time ago. Fortunately, the problems seem to be solved now [3a].

Figure 44b. Kresge Auditorium (MIT, Cambridge USA) Kresge auditorium has whispering galleries (see acoustics). Nevertheless, the acoustic properties are quite good and it is often used as a concert hall. Deitingen petrol station In Switzerland next to highway A1 is Deitingen petrol station (Fig. 44c). It has two reinforced concrete canopies that have been designed by Heinz Isler. They have been built by Willi Bsiger AG in 1968. Note that this shell does not have edge beams. Its deformation is inextensional but apparently this does not give problems. The span is m. The smallest thickness is mm. The radius of curvature is m. The ratio a/t = . The formwork of this shell consisted of steel scaffolding, curved glulam beams (approximately 180 x 50 mm spaced 800 mm) and wood floor boards. The formwork parts were reused on other projects. The concrete is watertight and a roof cover has not been applied. The surface is just painted [4]. Heinz Isler (19262009) designed more than 1200 reinforced concrete shell structures. Most were built between 1955 and 1979. He did not use computers for structural analysis. Instead, he used plastic models to determine deflections, stresses and buckling loads (Fig. 44d) [4a].

Figure 44c. Deitingen petrol station

Figure 44d. Model of a shell structure made by Heinz Isler [4a] Gau-Bonnet theorem
1 1 1 A sphere has a in every point a Gaussian curvature of kG = k1k2 = ( )( ) = . a a a2

It has a surface area of A= 4a 2 . The total Gaussian curvature of a sphere is

kG dA=
A

1 a2

A= 4 .

When this calculation is repeated for an ellipsoid, a pseudo sphere or a brick the results are also 4. The total Gaussian curvature of any object without holes is 4. If the object has one hole the total Gaussian curvature is 0. If the object has two holes the total Gaussian curvature is -4. This is the Gau-Bonnet theorem which was published by the mathematician Pierre Bonnet in 1848 [4].

Shell design 2 Consider a point load perpendicular to the surface of a shell. Under the point load the Gaussian curvature has decreased (or increased). According to the Gau-Bonnet theorem the total Gaussian curvature does not change. So somewhere else the Gaussian curvature must have increased (or decreased). Apparently, shells carry load by moving around Gaussian curvature. Point load on a sphere In 1946 Eric Reissner solved the Sanders-Koiter equations for a spherical cap loaded by a point load perpendicular to the surface [5]. The deflection u z under the point load P is
= uz 3 Pa 1 2 . 2 4 Et

P uz

The membrane forces under the point load are

n1 = n2 = 3P 1 2 . 8 t

4 at

The moments under the point load are

m = = 1 m 2 1 4 4 12 P at 1 + , d 4 1 2

where d is the diameter of the small circular area over which the point load is distributed. Point load on a shell of positive Gaussian curvature

In 2012, Amir Semiari made finite element models of surfaces of varying curvature [6]. He found that Reissners solution of a point load on a sphere can be adapted to shells of any positive Gaussian curvature by replacing the radius a by the Gaussian curvature kG .
= uz 3 P 1 2 . 2 4 Et kG

This formula is accurate when the deformation is in-extensional and the distance from the t point load to any shell edge is more than 1.65 . The membrane forces under the point k2 load are the same as for a sphere. The moments under a point load are
m = = 1 m 2 1 4 4 12 P d t 1+

kG 4 1 2

Point load on a cylinder In 1977, Chris Calladine studied circular cylindrical shells loaded by point loads (Fig. 44e) []. He found that there is a difference between long cylinders and short cylinders.
a long cylinder l > at t
P a 2 1 2 , = u z 0.80 Et t
3

short cylinder

a at < l < at t

P a 4 l 2 1 2 , u z 0.35 with fixed ends= Et t a P a 4 l 2 with diaphragm= ends 1 2 , u z 0.50 Et t a

5 1

with free ends = u z 0.50

P a a 1 2 . Et t l

where l is the cylinder length. 3 The membrane forces under a point load are 4
n1 = 0, n2 = 3P 1 2 . 8 t

The principal normal force directions are the principal curvature directions. The moments under a point load are the same as for a point load on a shell of positive Gaussian curvature.

Figure 44e. Point loads on a circular cylindrical shell Point load on a shell of negative Gaussian curvature In , The membrane forces under the point load are 4

The coefficient 0.80 has been determined by Geoffrey van Bolderen using finite element analysis [7a]. The coefficients 0.35, 0.50 and 0.50 are due to []. 4 The formulas are also valid for a negative load P. In that case, exchange n1 and n2 .

10

n1 =

3P 3P 1 2 , n2 = 1 2 . 8 t 8 t

(Apparently, for a negative Gaussian curvature, there is no umbilic in the membrane force trajectories under a point load.) The principal normal force directions are the principal curvature directions. The moments under a point load are the same as for a point load on a shell of positive Gaussian curvature. Plate twisting A flat plate can be twisted. Doing so a Gaussian curvature is introduced. Since the Gaussian curvature changes, the deformation is extensional and membrane forces develop (Fig. 44a). The phenomenon can be easily observed in a piece of fabric (Fig. 44b). Ask somebody to hold two corners of the fabric and hold the other two corners yourself. Stretch the fabric firmly. Move slowly one of the corners out of plane. You will observe that the middle of the fabric becomes floppy. If the fabric were a plate the middle would be compressed and the edges would be tensioned.

Figure 44b. Twisted piece of fabric Figure 45 shows a square plate that is initially flat and supported in four corners. One of the corners has been displaced perpendicular to the surface. At a displacement of t the Gaussian curvature is small and hardly any membrane forces develop. The stresses due to bending moments are much larger than the stresses due to membrane forces. At a displacement of 10t substantial membrane forces have developed and a significant Gaussian curvature can be observed. At a displacement of 20t the plate deformation has changed. The membrane forces and Gaussian curvature are almost zero.

u=t u = 10t u = 20t Figure 45. Geometrically nonlinear finite element analysis of a square plate The transition from significant Gaussian curvature to almost zero Gaussian curvature occurs suddenly at a displacement u = 16.8t . This has been discovered by Dries Staaks in 2003 in his graduation project [7]. The explanation is that the compression forces in the plate middle become so large that the plate buckles. In the plate middle the membrane forces are
2 1 n = = 1 n 2 64 E t b kG ,

11

where b is the plate length and width. In the plate edge the membrane forces are
1 E t b2k n1 = 32 n2 = 0 .5 G

Before buckling the Gaussian curvature is kG = u 2 b 4 . The latter equation can be derived it the same way as hypar curvature. The previous formulas are for square panels. Unfortunately, for rectangular panels no formula is available. Important applications are glass faades and glass roofs (Fig. 46a).

Figure 46a. Canopy of twisted glass panels at a bus stop in Delft, the Netherlands Exercise: A reinforced concrete hypar (a = 140 m) will be cast on a timber formwork. The formwork will consist of straight beams in parallel to the hypar edges and multiplex plates. The plates will be twisted. Clearly, we do not want them to buckle. The factory dimensions of the plates are 2440 x 1220 x 18 mm. Do the plates need to be cut to a smaller size? Gaussian curvature of boats Steel boats are made of plates that have zero Gaussian curvature both before assembling and after assembling (Fig. 46). An edge where the plates are connected is called chime.

chime

keel

These formulas are also valid for positive Gaussian curvatures. In that case, exchange n1 and n2.

12

Figure 46. Curved plates at the bow of a steel boat (zero Gaussian curvature) Prestressing tents Tents are made of fabric parts that are sewed together. The Gaussian curvature of the fabric is zero (It leaves the factory on a role). Therefore, in a traditional circus tent every fabric part has zero Gaussian curvature (Fig. 47a). However, architects like smooth shapes which do have Gaussian curvature (Fig. 47). If we impose a Gaussian curvature to a fabric it wrinkles, unless it is prestressed. In the direction of the seams the required prestress is
1 E t b2k , nxx = 24 G

where b is the fabric width. Perpendicular to the seams the required prestress is
1 n yy = 6144

E t b 4 kG k x z

where z is accepted maximum distance from the theoretical smooth surface to the tent fabric [8].

Figure 47a. Traditional circus tent (zero Gaussian curvature)

Figure 47. Canopy at the European patent office in Rijswijk, Netherlands (negative Gaussian curvature) architect Lewis X Associates, consultant Tentech, contractor Poly-Ned Trajectories The software can plot principal directions on the surface of a shell (Fig. 48). By hand we can draw lines that follow the principal directions (Fig. 49). We call these lines trajectories. In a vector field, for example the shear force, the trajectories have one direction. In a tensor field, for example the moments, the trajectories have two perpendicular directions. In other courses other words are used for trajectories, for example hydraulic engineers call them flow lines, electro engineers call them field lines and mathematicians call them integral curves.

13

Figure 48. Four hypar shells

star edge beam
support

lemon

interior beam

star

symmetrical lemon edge beam

monstar
star

star interior beam Figure 49. Moment trajectories in one of the hypar shells of Fig. 48 due to self-weight

Umbilics An umbilic is a point in a tensor field where both principal values are the same. For example nxx = 40 kN/m, n yy = 40, nxy = 0. Consequently, n1 = 40 and n2 = 40 and Mohrs circle is just a point. Principal directions cannot be determined. The reason is that the principal directions are defined as the directions in which there is no shear stress. In an umbilic there is no shear stress in any direction. Umbilics are also called umbilical points or isotropic points. Umbilical patterns The trajectories around an umbilic have a particular pattern. If the tensor field is linear in x and y around the umbilic then either a monstar or a star occurs (Fig 50). Both patterns have three trajectories that go trough the umbilic. These trajectories are called ridges. The ridges of a monstar are always within a 90 angle. The ridges of a star are always not within a 90

14

angle. When any two ridges have an angle of exactly 90 then the third ridge does not occur and the usual orthogonal pattern occurs. When the three ridges of a monstar coincide a lemon occurs. 6 When two ridges of a monstar coincide a pattern occurs that does not have a name. Let us call it a flame. 7 Figure 51 shows the trajectory pattern as a function of the ridge angles. More patterns are possible if the tensor field around an umbilic is nonlinear in x and y. Then the number of ridges is unlimited. For example the moment trajectories around a point load perpendicular to a shell. These are not studied in these notes as yet.

monstar

star

lemon

flame

Figure 50. Trajectory patterns around an umbilic in case of a linear tensor field
1 2

O S
O ridges
1
F
1 2

y
2 1

O
F

F M

O
2
M

O
1 2

L F
O O
S

M
F

M
M
F

O O
1 2

M .. monstar S .... star L ... lemon F ... flame O ... orthogonal

Figure 51. Umbilical patterns as a function of the ridge angles 1 and 2 Identifying umbilical patterns It is often difficult to recognise the umbilical pattern from the finite element principal directions, especially when several umbilics occur close to each other. Fortunately, they can be recognised computationally too.
6

The name lemon is related to the fruits shape that can be recognised in the trajectory pattern. In fact, the name monstar is derived from lemon-star. 7 The theory of umbilics has been developed by mathematicians studying differential geometry [9]. They probably thought that the flame was not interesting and did not need to be mentioned. This has created a lot of confusion amongst engineers who observed trajectories in which two ridges crossed at angles different than 90 which they thought would not be possible. By giving it a name future confusion might be avoided.

15

Six factors a can be computed from the tensor finite element results around an umbilic.
a1 = nxx nxx a2 = x y n yy n yy a3 = a4 = x y nxy nxy a5 = a6 = x y

The directions of the ridges are the roots of

f = a6 tan 3 + (a2 a4 + a5 ) tan 2 + (a1 a3 a6 ) tan a5 (a2 a4 ) tan 3 + (a1 a3 4a6 ) tan 2 ( a2 a4 + 4a5 ) tan ( a1 a3)

For example, Figure 52 shows f for a1 = 1, a2 = 2, a3 = 3, a4 = 4, a5 = 5, a6 = 6 N/mm. The roots can be computed using the Newton-Raphson algorithm. When the directions of the ridges are known the umbilical pattern can be identified using Figure 51. Excersise: What umbilical pattern follows from Figure 52 ?

Figure 52. Function f of , the three roots are angles of ridges with the x axis Table 1. Values of a1 to a6 a1 Monstar 2 Star 0 Lemon 2 Flame 1 Orthogonal 0 for the patterns of Figure 50
a2 0 2 0 -1 1 a3 0 0 0 0 0 a4 0 0 0 0 0 a5 0 1 0 0 0 a6 3 0 1 1 0

Invariants The factors a defined in the previous section depend on the direction of the local x axis and y axis. For making contour plots it is useful to have quantities that do not depend on the reference frame. The following quantities have this property. They are called invariants. They are valid for all points of a shell, not only for the umbilics.

16

1= (a1 a3 )a6 (a2 a4 )a5

2 1 2 2 2 1 = 2 4 ( a1 a3 ) + 4 ( a2 a4 ) + a5 + a6 3 = 2 2 = 4 4(3a6t1 + t2 )(3a5t2 + t1 ) (t2t1 9a5 a6 )2 t1 = a3 a1 + a6

t2 = a2 a4 + a5
2 2 = 5 4(a1a3 a5 )(a2 a4 a6 ) (a1a4 + a2 a3 2a5 a6 ) 2

The first invariant 1 gives information on the umbilical pattern. Where 1 > 0 monstars occur, where 1 < 0 a stars occur, where 1 = 0 orthogonal patterns or nonlinear patterns occurs. The second invariant 2 gives information on the gradient of the tensor field. This can be used for determining the finite element size. The third invariant 3 has no particular meaning. The forth invariant 4 gives information on the number of ridges in umbilics. Where 4 > 0 one ridge occurs (lemon), where 4 < 0 three ridges occur (stars or monstars), where 4 = 0 two ridges occur of which one is double (flame). The fifth invariant 5 gives information on the shape of the contour lines of both principal values. Where 5 > 0 the contour lines are ellipses, where 5 < 0 the contour lines are hyperbolas (Fig 50). It is conjectured that six independent invariants exist. After all, there are also 6 variables. However, not all have been found as yet.

Figure 55. Sydney Opera House (Australia)

17

Architect: Jrn Utzon Engineering: Ove Arup and partners Contractor: Hornibrook Group Pty Ltd. The building was designed in 1955 and completed in 1973. The shell roofs are made of precast concrete panels supported by precast concrete ribs. Cladding: white tiles. Costs: $102 million Literature 1. J. Blaauwendraad , J.H. Hoefakker, , 2. 3a. T.E. Boothby, Historic Preservation of Thin-Shell Concrete Structures, online, 26 May 2012, http://www.engr.psu.edu/ae/thinshells/module%20III/case_study_3.htm 3. C.F. Gauss,Disquisitiones generales circa superficies curvas Oct. 8, 1827, Commentationes societatis regiae scientarium Gottingensis recentiores, Gttingen 1873, Vol. 4, pp. 217 - 492 (in Latin), 4. H. Bsiger, The building of Isler shells, IASS Journal, Vol. 53 (2011) No. 3, pp. 161-169 4a. E. Ramm, Heinz Isler shells the priority of form, IASS Journal, Vol. 53 (2011) No. 3, pp. 143-154 4b. T. Kotnik, J. Schwartz, The architecture of Heinz Isler, IASS Journal, Vol. 53 (2011) No. 3, pp. 185-190 5. E. Reissner, Stresses and small displacements of shallow spherical shells II, Journal of Mathematics and Physics, Vol. 25 (1946), pp. 279-300 6. A. Semiari, Doorbuiging van schalen door puntlasten, Bachelor report, Delft University of Technology, 2012 7a. G.S. van Bolderen, Doorbuiging van cilindrische schaalconstructies door puntlasten, Bachelor report, Delft University of Technology, 2012 7. D. Staaks, Koud torderen van glaspanelen in blobs, MSc report, Eindhoven University of Technology, the Netherlands, 2003 8. P.C.J. Hoogenboom, P.A. de Vries, R. Houtman, "Requirements for cutting patterns of smooth membrane structures", IASS Journal, Vol 50 (2009), No. 1. pp. 23-32 9. G. Darboux, Leons sur la thorie gnerale des surfaces Vol. 4, Gauthier-Villars, 1896, p. 455.

18

Appendix 5 The quantity k y xx + k xy xy k x yy is approximately equal to the increase of the shell Gaussian curvature kG during loading. Proof In the local coordinate system the shell surface can be approximated by
z= 1 k x 2 + k xy xy + 1 k y2 . 2 x 2 y

A displacement can be approximated as

2 1 2 1 1 u = z u zo + x x + y y 2 xx x 2 xy xy 2 yy y .

The deformed shape is

2 2 1 1 1 z+u = z u zo + x x + y y + 2 ( k x xx ) x + ( k xy 2 xy ) xy + 2 ( k y yy ) y .

The curvatures after deformation are

2 ( z + uz ) = k x xx , x2 2 ( z + uz ) = k xy 1 , 2 xy xy

2 ( z + uz ) = k y yy . y2

Before deformation the Gaussian curvature of the middle surface is

2 . = kG k x k y k xy

After deformation the Gaussian curvature is

= (k x xx )(k y yy ) (k xy 1 )2 . kGd 2 xy

The increase in Gaussian curvature is

kGd kG = k y xx + k xy xy k x yy + xx yy 1 2 . 4 xy

The last two terms are very small compared to the other terms and can be neglected for shells with significant curvatures. They cannot be neglected for flat plates. Q.E.D.

The increase of the Gaussian curvature can also be written as

kGd kG = ( k y 1 ) + (k xy 1 ) (k x 1 ) . 2 yy xx 4 xy xy 2 xx yy

19

Appendix 6 Trajectories around umbilics Ridge angles as function of constants a1 to a6 nxx = p + a1 x + a2 y

n yy = p + a3 x + a4 y nxy = a5 x + a6 y

(1)

= x r cos = y r sin
= nrr 1 (n + n yy ) + 1 (n n yy ) cos 2 + nxy sin 2 2 xx 2 xx = nss 1 (n + n yy ) 1 (n n yy ) cos 2 nxy sin 2 2 xx 2 xx = nrs 1 (n n yy )sin 2 + nxy cos 2 2 xx

(2)

(3)

The principal directions in the r-s coordinate system is

tan 2 = 2nrs nrr nss

(4)

For a ridge = 0 or = 1 . 2 Substitution of Eqs (1), (2), (3) and (5) in (4) gives
0= a6 tan 3 + (a2 a4 + a5 ) tan 2 + (a1 a3 a6 ) tan a5 (a2 a4 ) tan 3 + (a1 a3 4a6 ) tan 2 (a2 a4 + 4a5 ) tan (a1 a3)

(5)

(6)

The denominator is important because when any two ridges have an angle of /2 the third ridge is cancelled out of this fraction. Therefore, there can be one, two or three roots. The roots can be computed using, for example, the Newton-Raphson algorithm. The derivative of Eq. (4) is
(t1 tan 2 + t2 tan + t3 )(1 + tan 2 ) 2 d tan 2 = (7) d ((a2 a4 ) tan 3 + (a1 a3 4a6 ) tan 2 ( a2 a4 + 4a5 ) tan ( a1 a3)) 2
t= 1 a6 ( a1 a3 4a6 ) ( a2 a4 + a5 )( a2 a4 ) t2 = 2(a2 a4 )(a1 a3 ) 8a5 a6 t3 = a5 (a2 a4 + 4a5 ) (a1 a3 a6 )(a1 a3 )

Constants a1 to a6 as function of the ridge angles To obtain the patters of Fig. 49 we choose a Cartesian reference frame in an umbilic and assume a linear variation in the second order tensor, for example the normal forces.

20

nxx = p + a1 x + a2 y n yy = p + a3 x + a4 y nxy = a5 x + a6 y

(1)

Cylinder coordinates are introduced.

= x r cos = y r sin

(2)

The principal direction is defined by

tan 2 = 2nxy nxx n yy

(3)

For the ridges holds

=

(4)

Substitution of Eqs (1), (2) and (4) in (3) gives a third degree polynomial in tan .
a6 tan 3 + (a2 a4 + a5 ) tan 2 + ( a1 a3 a6 ) tan a5 = 0

(5)

Suppose the roots of this polynomial are b1 , b2 and b3 . Then it can be written as
(b1 tan )(b2 tan )(b3 tan ) = 0.

(6)

This can be evaluated as

tan 3 (b1 + b2 + b3 ) tan 2 + (b1b2 + b2b3 + b3b1 ) tan b1b2b3 = 0 .

(7)

Comparing Eq. (5) to (7) we observe,

a2 a4 + a5 = (b1 + b2 + b3 ) a6 a1 a3 a6 = b1b2 + b2b3 + b3b1 a6 a5 = b1b2b3 a6

(8)

The roots are the angles of the ridges (Fig. 50)

= b1 tan 1 = b 2 tan 2 = b 3 tan 3

(9)

Substitution of Eqs (9) in (8) and evaluation gives

21

a1 a3 = (ccc + ssc + css + scs ) C a2 a4 = ( sss + ccs + csc + scc) C a5 = sss C a6 = ccc C

(10)

where C is an unknown factor and

sss =sin 1 sin 2 sin 3 css =cos 1 sin 2 sin 3 sin 1 cos 2 sin 3 scs = ssc =sin 1 sin 2 cos 3 ccs =cos 1 cos 2 sin 3 csc =cos 1 sin 2 cos 3 sin 1 cos 2 cos 3 scc = ccc =cos 1 cos 2 cos 3

(11)

Invariant proof
> restart: > x:=r*cos(f)-s*sin(f): > y:=r*sin(f)+s*cos(f): > nxx:=p1+a1*x+a2*y: > nyy:=p2+a3*x+a4*y: > nxy:=p3+a5*x+a6*y: > nrr:=1/2*(nxx+nyy)+1/2*(nxx-nyy)*cos(2*f)+nxy*sin(2*f): > nss:=1/2*(nxx+nyy)-1/2*(nxx-nyy)*cos(2*f)-nxy*sin(2*f): > nrs:= -1/2*(nxx-nyy)*sin(2*f)+nxy*cos(2*f): > b1:=diff(nrr,r): > b2:=diff(nrr,s): > b3:=diff(nss,r): > b4:=diff(nss,s): > b5:=diff(nrs,r): > b6:=diff(nrs,s): > simplify((b1-b3)*b6-(b2-b4)*b5); a1 a6 a2 a5 a3 a6 + a4 a5

Invariants as a function of the ridge angles

1 =

+ cos(1 2 ) cos(1 3 ) cos(2 3 ) ) C 2 (1 4

= 2 1 cos(1 2 ) cos(1 3 ) cos(2 3 )C 2 2 3 = 4 = 5 = 6 =

(12)

22