Structural models overview

In scientific literature often the following names are used for structural idealisations.

structural element name deformation included

beams Euler-Bernoulli beam bending
Timoshenko beam bending and shear
plates loaded in plane Navier equations extension
plates loaded Kirchhoff plate bending
perpendicularly to Reissner-Mindlin plate (p. 60) bending and shear
their plane Von Kármán-Föppl equations extension, bending and
large displacements
shells Shell membrane equations (p. 36) extension
Sanders-Koiter equations (p. 52) extension and bending
several theories extension, bending and shear

Shell theory
In 1888 Augustus Love1 formulated the basic equations that govern the behaviour of thin
elastic shells [21, 22]. He used Jacob Bernoulli’s2 hypothesis which was also used by Gustav
Kirchhoff 3 in formulating the plate theory. In the years that followed there was much
discussion on this shell theory. Some inconsistencies were found. Many scientists proposed
other equations, such as Wilhelm Flügge4 (1934) [23], Ralph Byrne5 (1944) [24], Eric
Reissner6 (1952) [25] and Paul Naghdi7 (1957) [26]. Also Love himself proposed improved
equations [27]. Lyell Sanders8 was the first to remove all inconsistencies from Love’s first
equations [28]. Independently, Warner Koiter9 proved that Love’s initial assumptions were
correct after all and he also derived the correct shell equations [29, 30]. In 1959 there was a
conference in the aula of Delft University where Sanders presented the correct shell equations
and Koiter presented the correct shell equations. One of Koiter’s papers on the subject has the
clear title “All you need is Love” [31].

Love’s first equations are called the first approximation theory. Including improvements they
are referred to as the Sanders-Koiter equations (p. 52). Other theories account for out-of-plane
shear deformation and are called higher-order approximation theory. They are intended for
thick shells (p. 14).

The equations are difficult to solve analytically, therefore, simplifications were proposed for
specific shell shapes by many scientists. For example, equations for cylindrical shells were
proposed by Lloyd Donnell10 (1934) [32] and Leslie Morley11 (1959) [33].

1
Augustus Love (1863–1940) was a mathematician and professor in Oxford [Wikipedia].
2
Jacob Bernoulli (1654–1705) was a professor of mathematics in Bazel [Wikipedia].
3
Gustav Kirchhoff (1824–1887) was a German physicist and professor in Berlin, Breslau and
Heidelberg. He is also well-known in physics for many discoveries such as Kirchhoff’s laws on
electrical current [Wikipedia].
4
Wilhelm Flügge (1904–1990) was professor of civil engineering in Göttingen. After the second world
war he and his wife moved to the USA and became professors in Stanford [Wikipedia].
5
Ralph Byrne (191..–194..) was associate professor of applied mechanics in Caltech. He died in the
second world war […].
6
Eric Reissner (1913–1996) was professor of applied mechanics in MIT and San Diego. His father,
Hans Reißner (1874–1967) was an aircraft engineer and professor in Aachen and Berlin. The family
moved from Berlin to the Illinois just before the second world war [Wikipedia].
7
Paul Naghdi (1924–1994) was born in Tehran. He studied in the USA and became professor of
mechanical engineering in Berkeley [Wikipedia].
8
Lyell Sanders (1924–1998) was professor of structural mechanics in Harvard […].
9
Warner Koiter (1914–1997) was professor of applied mechanics in Delft [Wikipedia].
10
Lloyd Donnell (1895–1997) was professor of mechanical engineering in Illinois [Wikipedia].

51
Sanders-Koiter equations
The following 21 equations describe membrane action and bending action in thin shell
structures. Equation 6 is derived below. The other equations are not derived in these notes but
they can be obtained in the same way. The derivation of Sanders and that of Koiter can be
found in literature [28] and [29, 30] respectively. The derivation of Koiter is based on tensor
analysis and is most rigorous. The equations are valid for elastic material behaviour and small
displacements. They correctly predict no stresses for rigid translations and rigid rotations. The
equations correctly produce symmetrical stiffness matrices (Betti’s reciprocal theorem). The
Sanders-Koiter equations include the equations for plates. In other words, with appropriate
values for k xx , k yy , k xy ,  x ,  y the Sanders-Koiter equations simplify to the equations for
plates loaded in plane, plates loaded perpendicular to their plane (Kirchhoff theory), circular
plates and the shell membrane equations (p. 36). This clearly is a remarkable achievement of
the 20th century scientists. The Sanders-Koiter equations are a scientific masterpiece. 12

Table 6. Sanders-Koiter equations

equilibrium v v y 1
equations k xx nxx  k xy (nxy  n yx )  k yy n yy  x   k y vx  k x v y  pz  0
x y
nxx n yx 2
  k y (nxx  n yy )  k x (nxy  n yx )  k xx vx  k xy v y  p x  0
x y
n yy nxy 3
  k x (n yy  nxx )  k y (nxy  n yx )  k yy v y  k xy vx  p y  0
y x
mxx mxy 4
  k y (mxx  m yy )  2k x mxy  vx  0
x y
m yy mxy 5
  k x (m yy  mxx )  2k y mxy  v y  0
y x
k xy (mxx  m yy )  (k xx  k yy )mxy  nxy  n yx  0 6
constitutive Et 7
equations nxx  ( xx   yy )
1  2
Et 8
n yy  ( yy   xx )
1  2
nxy  n yx Et 9
  xy
2 2(1  )
E t3 10
mxx  (  xx   yy )
12(1   2 )
E t3 11
m yy  (  yy   xx )
12(1   2 )

11
Leslie Morley (1924–2011) was a scientist in the Royal Aircraft Establishment and a professor in
Brunel University, London [Wikipedia].
12
The following dates provide a time frame. In 1822 Claude-Louis Navier formulated the Navier-
Stokes equations which describe the behaviour of fluids [Wikipedia]. In 1850 Gustav Kirchhoff
completed the differential equation that describes the behaviour of plates [Wikipedia]. In 1865 James
Clerk-Maxwell unified many laws into Maxwell’s equations that describe electric and magnetic fields
[wikipedia]. In 1916 Albert Einstein found the Einstein field equations describing the structure of the
universe [wikipedia]. In 1925 and 1926 Werner Heisenberg, Max Born and Pascual Jordan found the
Heisenberg equation of quantum mechanics describing materials on a very small scale [Wikipedia].

52
 E t3 12
mxy   xy
24(1  )
kinematic u x 13
equations  xx   k xxu z  k xu y
x
u y 14
 yy   k yy u z  k y u x
y
u x u y 15
 xy    2k xy u z  k x u x  k y u y
y x
u z 16
x    k xxu x  k xy u y
x
u 17
 y   z  k yy u y  k xy u x
y
u x u y 18
 z  1 (   k xux  k y u y )
2 y x
 x 19
 xx   k xy  z  k x  y
x
 y 20
 yy   k xy  z  k y  x
y
 x  y 21
 xy    (k xx  k yy ) z  k x  x  k y  y
y x

Ping pong ball

Consider a sphere that is deformed into an ellipsoid (fig. 72). Think of a ping pong ball that is
squeezed by your hand. The code below shows the evaluation of the Sanders-Koiter equations
2u 2u
(p. 52) by Maple. The deformation u z  b cos , u x  0.49 b sin has been obtained by trial
a a
and error to minimize the load p x . The code produces figure 73. Displacement u y and
distributed force p y are zero and p x is almost zero. Only pz is needed to obtain this
deformation.

uz
u
ux z
a

Figure 72. Deformation of a spherical ping pong ball into a prolate ellipsoid shape

> a:=20: t:=0.4: E:=1400: nu:=0.3: b:=1:

> kxx:=-1/a: kyy:=-1/a: kxy:=0: alphax:=1: alphay:=sin(u/a):
> ux:=0.49*b*sin(2*u/a): uy:=0: uz:=b*cos(2*u/a):
>
> ky:=diff(alphay,u)/alphay/alphax: kx:=diff(alphax,v)/alphax/alphay:
> epsilonxx:=diff(ux,u)/alphax-kxx*uz+kx*uy:

53
 > epsilonyy:=diff(uy,v)/alphay-kyy*uz+ky*ux:
> gammaxy:=diff(ux,v)/alphay+diff(uy,u)/alphax-2*kxy*uz-kx*ux-ky*uy:
> phix:=-diff(uz,u)/alphax-kxx*ux-kxy*uy:
> phiy:=-diff(uz,v)/alphay-kyy*uy-kxy*ux:
> phiz:=1/2*(-diff(ux,v)/alphay+diff(uy,u)/alphax-kx*ux+ky*uy):
> kappaxx:=diff(phix,u)/alphax-kxy*phiz+kx*phiy:
> kappayy:=diff(phiy,v)/alphay+kxy*phiz+ky*phix:
> rhoxy:=diff(phix,v)/alphay+diff(phiy,u)/alphax+(kxx-kyy)*phiz-kx*phix-ky*phiy:
> nxx:=E*t/(1-nu^2)*(epsilonxx+nu*epsilonyy):
> nyy:=E*t/(1-nu^2)*(epsilonyy+nu*epsilonxx):
> nxym:=E*t/(2*(1+nu))*gammaxy:
> mxx:=E*t^3/(12*(1-nu^2))*(kappaxx+nu*kappayy):
> myy:=E*t^3/(12*(1-nu^2))*(kappayy+nu*kappaxx):
> mxy:=E*t^3/(24*(1+nu))*rhoxy:
> vx:=diff(mxx,u)/alphax+diff(mxy,v)/alphay+ky*(mxx-myy)+2*kx*mxy:
> vy:=diff(myy,v)/alphay+diff(mxy,u)/alphax+kx*(myy-mxx)+2*ky*mxy:
> tmp:=kxy*(mxx-myy)-(kxx-kyy)*mxy:
> nxy:=nxym-tmp/2:
> nyx:=nxym+tmp/2:
> px:=-(diff(nxx,u)/alphax+diff(nyx,v)/alphay+ky*(nxx-nyy)+kx*(nxy+nyx)-kxx*vx-kxy*vy):
> py:=-(diff(nyy,v)/alphay+diff(nxy,u)/alphax+kx*(nyy-nxx)+ky*(nxy+nyx)-kyy*vy-kxy*vx):
> pz:=-(kxx*nxx+kxy*(nxy+nyx)+kyy*nyy+diff(vx,u)/alphax+diff(vy,v)/alphay+ky*vx+kx*vy):
>
> plot({ux,uy,uz,px/1.5,py/1.5,pz/1.5},u=0..Pi*a-1);

ux

uz
pz

Figure 73. Loading pz and deformation u x , u z of a ping pong ball computed by Maple

Ingredients for deriving the Sanders-Koiter equilibrium equations

1. Equilibrium of forces, which follows directly from Newton’s second law of motion F = m a,
where F is the vector sum of forces on an object, m is the mass of the object and a is the
acceleration of the object written as a vector. This law has been confirmed on earth by many
experiments and in space by many observations. In extreme situations Newton’s laws are not
accurate and replaced by Einstein’s field equations.
2. Equilibrium of moments, which follows from Newton’s second law applied to a group of
point masses.
3. Assumption that the shape is not changed much by the deformation (equilibrium in the
undeformed state). Unfortunately, whether this assumption is fulfilled depends strongly on the
situation. Usually, it is fulfilled because well designed shells are very stiff.

Ingredients for deriving the Sanders-Koiter constitutive equations

4. Hook’s law including Poisson’s ratio expressed in stresses and strains of an isotropic material.
Experiments show that this law represents many materials accurately, until the stress becomes
too large.
5. Definitions of section membrane forces and section moments
6. Hypothesis that out of plane shear deformation is negligible. This hypothesis is correct for
thin shells, which has been shown by comparison to higher order approximation shell
theories.

54
7. Bernoulli's hypothesis; plane sections remain plane (p. 55). This has been checked by finite
element models of beams consisting of small solid elements.
8. Stress in the material in the z direction is negligible compared to the stresses in the x and y
directions. This can be verified with simple hand calculations.

Ingredients for deriving the Sanders-Koiter kinematic equations

9. Assumption that the deformations are small. Unfortunately, whether this assumption is
fulfilled depends strongly on the situation. Usually, it is fulfilled because well designed shells
are very stiff.
10. Small rotations
11. Definitions of strains of the middle surface and deformation curvature of the middle surface.
These definitions are chosen such that the formulas for the strain energy are intuitive. For
example, the strain energy of torsion is 1 mxy  xy .
2

Exercise: Is it correct to summarise that the Sanders-Koiter equations follow from Newton’s
second law, Hooke’s law, Bernoulli’s hypothesis and a few geometrical assumptions?

Plane sections remain plane

Bernoulli's hypothesis13 is: plane sections remain plane. It was used by Euler14 in deriving the
beam and column equations. It was generalised to straight needle remains straight by
Germain15 and by Kirchhoff and used in deriving the plate equations. Love, Sanders and
Koiter used this hypothesis too for deriving the Sanders-Koiter equations (p. 52). As a
consequence the constitutive equations of thin shells are exactly the same as those of thin
plates.

Compatibility equation
Sanders-Koiter equations 13 to 21 (p. 52) can be combined resulting in the following
equation.

 2 xx  2  xy  2 yy
    k yy  xx  k xy  xy  k xx  yy
y 2 xy x 2

(See derivation in appendix 4.) It shows that the strains of the middle surface are connected to
the bending deformation. Only deformations that fulfil this equation are possible. Therefore, it
is called the compatibility equation.

Rigid translation
The Sanders-Koiter equations (p. 52) are accurate for small displacements. However, for large
rigid translations they are accurate too. For example, consider a reinforced concrete industrial
chimney with a height of 70 m, a radius a = 2.6 m and a wall thickness t = 0.1 m. During a
storm the chimney top moves b = 1.0 m which is not exceptional for a chimney of this height.

A rigid translation of the whole chimney (fig. 74) can be described exactly by the
displacements

v v
u x  0, u y  b cos , u z  b sin .
a a

13
Jacob Bernoulli (1654–1705) was mathematics professor in Basel, Switzerland [Wikipedia].
14
Leonhard Euler (1707–1783) was mathematics professor in Saint Pietersburg and Berlin [Wikipedia].
15
Sophie Germain (1776–1831) was a French mathematician. She was one of the first women who
contributed to science. For many years she worked on the thin plate theory [Wikipedia].

55
Obviously, this translation should not produce strains.

b
uy

uz
a

v
uy
uz
z
Figure 74. Rigid translation of a cylinder cross-section

1
From the chimney geometry it follows that k xx  0, k yy   , k xy  0,  x  1,  y  1 .
a
Substitution of these in the kinematic equations 13 to 21 gives

 xx  0,  yy  0,  xy  0,  xx  0,  yy  0,  xy  0 ,

which is the correct result. Consequently, the large deflection of the chimney top can be
described by the Sanders-Koiter equations.

Exercise: Large rigid rotations do produce unrealistic strains and stresses. Check the Sanders-
Koiter equations for this.

Shell differential equations

When the Sanders-Koiter equations (p. 53) are substituted into each other the following two
coupled partial differential equations are obtained (assuming px  p y  0 and v x , v y ,
nxy  n yx are small ).

E t3
   2 2 u z  p z
12(1   ) 2

and

22  E t u z  0 ,

where,
 2 (.)  2 (.)  2 (.)
(.)  k xx  2k xy  k yy ,
y 2 xy x 2
 2 (.)  2 (.)
 2 (.)  
.
x 2 y 2
 is the Airy stress function16 which is related to the membrane forces

 2  2  2
nxx  , n yy  , nxy  n yx   .
y 2 x 2 xy

16
George Airy (1801–1892) was an astronomy professor in Cambridge, England [Wikipedia]

56
Differential equation type
Partial differential equations of the second order are subdivided in three types; elliptic,
parabolic and hyperbolic [53]. Physicists use this to predict the nature of the solution and
select a solution method. The membrane part of the shell differential equations (p. 56) is

  pz

In a well-designed thin shell this part dominates the behaviour. It can be shown that the type
of this differential equation depends on the Gaussian curvature kG (p. 23).

kG  0  elliptic, the solution is local

kG  0  parabolic, the solution extends along one straight line
kG  0  hyperbolic, the solution extends along two straight lines,
which are called characteristics

Shallow shell differential equation

For cylinders and spheres k xx , k yy , k xy are uniform. This reduces the shell differential
equations (p. 56) to

E t3
 2 2 2 2u z  E t u z   2 2 p z .
12(1   2 )

This is a linear eight order partial differential equation in curvilinear coordinates u and v (p.
29).

A shallow shell is a shell with a sagitta (p. 1) that is small compared to its span. For these
shells the curvatures do not change much over the surface and the above differential equation
can be a good approximation.

Plate boundary conditions

In general, the solution to an eight order partial differential equation has 8 constants in the u
direction and 8 constants in the v direction. The constants can be solved by 4 boundary
conditions on each edge. Figure 75 shows the boundary conditions of a canopy that is fixed
on one edge. Note that there are too many boundary conditions. So, some boundary
conditions cannot be fulfilled.

z
n yy  0 n yx  0
nxx  0
m yy  0 mxy  0
nxy  0
vy  0 u
mxx  0 v
ux  0
mxy  0
uy  0
vx  0
uz  0
x  0
n yy  0 n yx  0
m yy  0 mxy  0
edge part vy  0
Figure 75: Boundary conditions of a canopy

57
This problem also occurs in plates. It was solved by Gustav Kirchhoff 3 in 1850 [54]. He
derived the correct boundary conditions of plates from virtual work. Others interpreted his
solution as that the stresses due to the torsion moment mxy go round in the edge (fig. 76-1).
Therefore, mxy on the edge needs be replaced by a concentrated shear force V in the edge (fig.
76-2).
mxy dx  Vdx  V  mxy

From equilibrium of a somewhat larger edge part (fig. 76-3) it follows that

pz dxdy  v y dx  (vx  dvx )dy  (V  dV )  V  vx dy  0 .

This can be simplified to

dvx dV
pz dy  v y  dy   0.
dx dx
dV
When dy  0 then v y   0 which can be written as
dx
V
vy   .
x

Now we have 4 boundary conditions per edge and the differential equation can be solved (fig.
77).
vy

vx
pz
mxy vx dvx

dy V
V V
mxy V  dV

1 2 dx 3 dx
Figure 76: Forces on an edge part

nxx  0 n yy  0 n yx  0 z
nxy  0 V
m yy  0 v y  
mxx  0 x
u
V
vx   v ux  0
y
uy  0
uz  0
x  0

n yy  0 n yx  0
V
m yy  0 v y  
x
Figure 77: Boundary conditions according to Kirchhoff (plates)

58
Thus, according to Kirchhoff, mxy need not be zero on a plate or shell edge in the x or y
direction. Also vx need not be zero on an edge in the y direction, and v y need not be zero on an
edge in the x direction. Clearly, in reality they are zero.

We need to interpret mxy on an edge as a concentrated shear force V in the edge.

We need to interpret v on an edge as a change in the concentrated shear force V.

However, the plate boundary conditions are not entirely correct for shells (see shell boundary
conditions p. 62).

Exercise: In plates mxy = 0 in a fixed edge along the x or y direction. In shells this can be
observed too, however, there are exceptions. Can we show this with the Sanders-Koiter
equations?

Exercise: Proof that mxy = 0 in a free corner.

Reissner-Mindlin plate theory17

It is possible to come up with a new shell theory that does not have interpretation problems of
the boundary conditions. In fact, the Reissner-Mindlin theory [55] for thick plates predicts
mxy , vx and v y on edges realistically without interpretations (fig. 75). However, to compute
these values accurately we need to use very small finite elements on the edges. For example
when a plate is 180 mm thick we need to use finite elements that are less than 10 mm wide.
This is impractical due to large computation time and therefore almost never applied. A
practical element for a 180 mm plate is more than 250 mm wide. For that mesh mxy will not
be zero on the edges also not when the Reissner-Mindlin theory is used. Therefore, also in the
Reissner-Mindlin theory we need to interpret the torsion moment on an edge as a concentrated
shear force in the edge.

Edge shear stresses

The shear stress in a plate edge or shell edge is 18

3 vx 3 V
 xz   10 .
2 t 2 t2

17
The name of this theory refers to Eric Reissner and Raymond Mindlin. Eric Reissner (1913–1996)
was a professor of applied mechanics at MIT and the University of California San Diego [56].
Raymond Mindlin (1906–1987) was a professor of applied science at Columbia University, USA [57].
From our point of view they were very skilled in mathematics. They had to be because they did not
have computers.
18
In 2010, Johan Blaauwendraad (professor of structural mechanics at Delft University) used
Reissner’s plate theory (p. 59) to derive the stresses in plate edges. He showed that the shear stress
distribution is exponential and the factor of the peak stress is 3 10 [57]. In 2013, Rutger Zwennis (at
2
that time a student at Delft University) modelled a plate loaded in torsion using volume finite elements
[58]. He showed that the peak stress due to V includes the factor 4.48 instead of 3 10 = 4.74. Who is
2
right? The Reissner plate theory is not exact because Reissner made several assumptions in the
derivation. The finite element analyses is not exact either because the number of elements is restricted.
In these notes the safe choice of 3 10 has been made. Future computers will be able to determine the
2
factor very accurately.

59
The formula is valid when the local x axis points in the direction of the edge and the local y
axis points outwards (fig. 78). Unfortunately, finite element programs using shell elements do
not compute this stress. If important, we need to calculate and check this stress by hand.

The concentrated shear force produces a local stress peak. In many structures a local stress
peak is not important because the stress will redistribute (steel yields, reinforced concrete
cracks). However, a stress peak is important for materials that do not yield such as glass. A
stress peak is also important for fatigue.

3 vx outer surface
mxy
2 t 6
t2
z

 xz
x edge
3 V edge shear stress
10
2 t2 y
Figure 78. Shear stresses in a free shell edge

Reinforced concrete plate edges

In reinforced concrete plates it is common practice to put hairpins in the edges (fig. 79). A
hairpin is a reinforcing bar that is bend in the shape of a U. The hairpins have the same
diameter and spacing as the bars perpendicular to the edge. There is a good reason for these
hairpins. They carry the concentrated shear force (fig. 80).
tie (hairpin)
strut (concrete)
hairpin plate edge
V
plate edge

Figure 79. Reinforcement in a Figure 80. Strut-and-tie model of a

cross-section of a concrete plate edge reinforced concrete plate edge

Edges that are not in the x or y direction

If an edge is not in the x direction or y direction the shear force vx and the torsion moment
mxy need to be transformed to the edge direction. For this we need to rotate the local
coordinate systems of the edge finite elements such that one of the axes is in the direction of
the edge. The obtained concentrated shear force on a free or simply supported edge can be
2
easily checked because it is equal to V   mxy  mxx m yy ,

Proof: Plate moments are a tensor (p. 93). m1 and m2 are the principal values (p. 94). m1m2 is
2
an invariant (p. 169) of this tensor. Therefore, m1m2  mxx m yy  mxy 2
 mss mtt  mst . Suppose

60
that the s axis is perpendicular to the shell edge. Since the edge is free or simply
2
supported mss  0 . Therefore, mxx m yy  mxy 2
 mst  V 2 . Q.E.D.

Palazzetto dello sport […]

The palazzetto dello sport was built for the 1960 summer Olympics in Rome (fig. 1). It hosted
basketball. Nowadays, it is a sports and community centre.
The buttresses are made of prefab concrete. The shell and ribs are made of reinforced concrete
that was cast in situ. The formwork of the shell consisted of 1620 cassettes supported by steel
tube scaffolding. The cassettes were made of 25 mm thick ferrocement (fig. 81). Ferrocement
is a thin layer of mortar with a steel wire mesh inside.

Construction sequence of the dome Completed

- Placing the buttresses
- Building the scaffolding for the cassettes. The scaffolding included
circular rings made of curved rails of an old railway track. These rings
were horizontally elevated onto temporary columns of steel tubes.
- Building a timber template of a large part of the shell internal surface August 1956
- Drawing the grid onto the template
- Fabrication of moulds for the cassettes. First, onto the timber template December 1956
the inside shape of one cassette was made of bricks and plaster (fig. 82).
Second, a cassette was made onto this inside shape. Third, this cassette
was moved down and several moulds were made of this cassette. Etc.
- Prefabrication of 30 cassettes a day
- Placing the cassettes onto the scaffolding (fig. 83, 84 )
- Placing reinforcing bars in and on the cassettes
- Pouring concrete (fig. 85) February 1957

architect: Annibale Vitellozzi (1903-1990)

engineer: Pier Luigi Nervi (1891-1979)
contractor: Bartoli

Computer analyses were not used. Structural calculations were done by hand and checked by
scale model tests.

Figure 81. Cross-section of the shell and ribs Figure 82. Mould fabrication

61
Figure 83. A cassette [..]

Figure 84. Scaffolding and cassettes [..]

Figure 85. Construction site during concrete pouring [..]

62
 Literature
21. A.E.H. Love, “The Small Free Vibrations and Deformation of a Thin Elastic Shell”,
Philosophical Transactions of the Royal Society of London, Vol. 179, 1888, pp. 491-546
22. A.E.H. Love, “A treatise on the mathematical theory of elasticity”, Dover Publications, New
York, 1944
23. W. Flügge, “Statik und Dynamik der Schalen”, Springer-Verlag, Berlin, 1934 (in German).
24. R. Byrne, “Theory of small deformations of a thin elastic shell”, University of California,
Publications in Mathematics (new series), 1944, Vol. 2, pp. 103-152.
25. E. Reissner, “Stress strain relations in the theory of thin elastic shells”, Journal of
Mathematical Physics, 1952, Vol. 31, pp. 109-119.
26. P.M. Naghdi, “On the theory of thin elastic shells”, Quarterly Applied Mathematics, 1957,
Vol. 14, pp. 369-380.
27. A.E.H. Love, ….
28. J.L. Sanders, “An improved first approximation theory for thin shells”, National Aeronautics
and Space Administration (NASA) Report No. 24, Washington, DC, 1959.
29. W.T. Koiter, “A consistent first approximation in the general theory of thin elastic shells”,
Report, Delft University of Technology, 1959.
30. W.T. Koiter, “A consistent first approximation in the general theory of thin elastic shells”,
Proceedings of IUTAM Symposium “On the Theory of Thin Elastic Shells”, Delft 24-28
August 1959, North-Holland publishing company, Amsterdam 1960, pp 12-33.
31. W.T. Koiter, “All you need is Love”, ….
32. L.H. Donnell, “A new theory for the buckling of thin cylinders under axial compression and
bending”, Transactions of the American Society of Mechanical Engineers, 1934, Vol. 56, pp.
759-806.
33. L.S.D. Morley, “An improvement on Donnell’s approximation for thin-walled circular
cylinders”, Quarterly Journal of Mech. Applied Mathematics, 1959, Vol. 12 (Part 1), pp. 89-
99.

63
Appendix 4: Compatibility equation
In this appendix the shell compatibility equation (p. 55) is derived. Consider the local
coordinate system x, y, z (p. 29). The curvatures k xx , k yy , k xy are assumed constant in x and
y. Sanders-Koiter equations 13 to 21 (p. 53) are used. In the local coordinate system there are
no curved coordinate lines therefore terms with radii of curved coordinate lines (p. 35) are
removed from the equations.
> restart:
> ux:=Ux(x,y):
> uy:=Uy(x,y):
> uz:=Uz(x,y):
> kG:=kxx*kyy-kxy^2:
> km:=(kxx+kyy)/2:
> epsilonxx:=diff(ux,x)-kxx*uz:
> epsilonyy:=diff(uy,y)-kyy*uz:
> gammaxy:= diff(ux,y)+diff(uy,x)-2*kxy*uz:
> phix:=-diff(uz,x)-kxx*ux-kxy*uy:
> phiy:=-diff(uz,y)-kyy*uy-kxy*ux:
> phiz:=1/2*(-diff(ux,y)+diff(uy,x)):
> kappaxx:=diff(phix,x)-kxy*phiz:
> kappayy:=diff(phiy,y)+kxy*phiz:
> rhoxy:=diff(phix,y)+diff(phiy,x)+(kxx-kyy)*phiz:
> simplify(diff(epsilonxx,y,y)+diff(gammaxy,y,x)-diff(epsilonyy,x,x)):
 2   2   2 
kxx  2 Uz( x, y ) 2 kxy  Uz( x, y ) kyy  2 Uz( x, y ) 
 y   y x   x 
> simplify(-kyy*kappaxx+kxy*rhoxy-kxx*kappayy-kG*(epsilonxx+epsilonyy)-2*kG*km*uz):
 2   2   2 
kxx  2 Uz( x, y ) 2 kxy  Uz( x, y ) kyy  2 Uz( x, y ) 
 y    y  x   x 
Apparently,

 2 xx  2  xy  2 yy u z2 u z2 u 2
    k xx  2k xy  k yy z ,
y 2 xy x 2 y 2 yx x 2

and
u z2 u z2 u 2
k yy  xx  k xy  xy  k xx  yy  kG ( xx   yy )  kG kmu z  k xx  2k xy  k yy z .
y 2 yx x 2

The terms kG ( xx   yy ) and kG kmu z can be neglected. For example, a steel sphere has a
radius of a = 100 mm and a thickness of t = 1 mm (a / t = 100). The local deformation is u z =
1 mm (u = t). The yield strain of the material is 0.002. The yield curvature of the material is
0.002 / (t / 2) = 0.004 mm-1.

kG ( xx   yy ) = 1/100² (0.002 + 0.002) = 4×10-7mm-2

kG kmu z = 1/100² (1/100 +1/100)/2 ×1 = 10×10-7 mm-2
k yy  xx  k xy  xy  k xx  yy = –1/100 ×0.004 + 0 –1/100 ×0.004 = –800×10-7 mm-2

Consequently,

 2 xx  2  xy  2 yy
    k yy  xx  k xy  xy  k xx  yy .
y 2 xy x 2

Q.E.D.

64