Chapter 1

Curvilinear media

1
 Chapter 2
Théorie locale et globale des coques

2
 Chapter 3
Discrétisation élément finis d’une structure courbe mince

3.1 Méthode d’élément finis pour une coque

3.1.1 Triangularization mesh
The mid-surface S of the shell is meshed with triangles having straight line or curved line edges.
The triangles are characterized by the maximum edge length of all the triangles denoted by h
and the mesh is denoted by Th . Flat elements (element with very very small area M) are not
accepted.

3.1.2 Shape functions and nodal vector of degree of freedom(dof )

Let Te , e ∈ Th be a triangle whose summit-nodes have the coordinates M1 = (x1 , y1 ), M2 =
(x2 , y2 ) and M3 = (x3 , y3 ). Let P1 denote spaces of polynomials of degree 1, generated by the
bases {1, x, y} or the barycentric functions η 1 , η 2 , η 3 defined as follows:


1


 η 1 (x, y) = [(y3 − y2 )(x2 − x) − (x3 − x2 )(y2 − y)]
2M







1

η 2 (x, y) = [(y1 − y3 )(x3 − x) − (x1 − x3 )(y3 − y)] (3.1)


 2M




η 3 (x, y) =
 1
[(y2 − y1 )(x1 − x) − (x2 − x1 )(y1 − y)]

2M
and η 1 +η 2 +η 3 = 1. The space of polynomials of degree 2 is denoted by P2 and {1, x, y, x2 , xy, y 2 }
or {η 1 , η 2 , η 3 , 4η 1 η 2 , 4η 1 η 3 , 4η 2 η 3 } constitute two equivalent bases.
The values of a function u at nodes of the triangle are denoted by u1 , u2 , u3 , u4 , u5 , u6 .
If u ∈ P1 then u = a1 x + a2 y + a3 = u1 η 1 + u2 η 2 + u3 η 3 If u ∈ P2 then u = a1 x + a2 y +
a3 x + a4 xy + a5 y 2 + a6 = u1 η 1 + u2 η 2 + u3 η 3 + u4 (4η 1 η 2 ) + u5 (4η 1 η 3 ) + u6 (4η 2 η 3 ).
2

We define specific finite element to be used to interpolate vector fields restricted in a triangle
element.
1. The element TGR 12 used for the gradient recovery method is defined as follows u , u ∈ P
1 2 1
and u3 ∈ P2

3
 u1 = a1 x + a2 y + a3 , u2 = a4 x + a5 y + a6 , u3 = a7 x2 + a8 xy + a9 y 2 + a10 x + a11 y + a12 .
 
a1
 a 
 2 
 a 
 3 
 a 
 4 
  a 

x y 1 0 0 0 0 0 0 0 0 0  5 
  
u1
 u2  =  0 0 0 x y 1 0 0 0 0 0 0   a6 
(3.2)
a 
 
u3 0 0 0 0 0 0 x2 xy y 2 x y 1  7 

 a8 

 a9 

 
 a10 
 
 a11 
a12
In the barycentric base we write:
 
u11

 u21 


 u31 


 u12 

   1 2 3  u22 
u1 η η η 0 0 0 0 0 0 0 0 0
 
u32
 
 u2  =  0 0 0 η 1 η 2 η 3 0 0 0 0 0 0 
 
(3.3)
u13

u3 0 0 0 0 0 0 η η η 4η η 4η 1 η 3
1 2 3 1 2 4η 2 η 3 
 
u23

 
u33
 
 
u43
 
 
u53
 
 
u63

Let t Ue = u11 , u21 , u31 , u12 , u22 , u32 , u13 , u23 , u33 , u43 , u53 , u63 .

Ue is 12 dof nodal element vector.

2. The element TSD 18 used for strain deformation approach. The local shape functions should

reproduce 3 rotations and 3 translations per node. They should account for uncoupled states due
to shearing, membrane deformation and bending. So locally u can be decoupled as u = u0 + up
where u0 captures rigid body motion (i.e. (u0 ) = 0, u0 6= 0). For a cylindrical shell with radius
R and open angle 2θ0 , the following shape functions fulfill the conditions.
t
Ue = u11 , u12 , u13 , β11 , β21 , γ 1 , u21 , u22 , u23 , β12 , β22 , γ 2 , u31 , u32 , u33 , β13 , β23 , γ 3


(3.4)

where βα = u3,α + Bαρ uρ , γ = u1,2 − u2,1 − u3,12 .

Letting y = ϕ.
The solution of these equations on a cylindrical shell with an open angle 2θ0 is u0 =
(u10 , u20 , u30 ), to be completed by a particular solution up = (u1p , u2p , u3p ) as follows:

 
a1
    a2 
u10 1 R(cosθ − cosθ0 ) −Rsinθ 0 0 0  
2
 a3 
 u20  =  0 xsinθ xcosθ −Rsin θ −sinθ cosθ 
  (3.5)
 a4 
u30 0 −xcosθ xsinθ Rsinθcosθ cosθ sinθ  
 a5 
a6

4
  
a1

 a2 


 a3 


 a4 

    a5 
u1p Rx Rθ Rxθ 0 0 0 0 0 0 0 0 0
 
a6
 
 u2p  =  0 0 0 θ xθ 0 0 0 0 0 0 0 
 
(3.6)
a7

u3p 0 0 0 0 0 x2 xθ θ2 x3 x2 θ xθ2 θ3 
 
a8

 

a9

 
 
 a10 
 
 a11 
a12

such that u = u0 + up

3.1.3 Shift-Lagrange elements

9 with 9 dof for the N-T model.
• TSL
Shape functions are defined using shift parameters X and Y defined as follows:

X(x) = x + β, Y (y) = y + β, α, β ∈ R

u1 (X, Y ) = a1 + a2 X + a3 Y, (3.7)
u2 (X, Y ) = a4 + a5 X + a6 Y,
u3 (X, Y ) = a7 X 2 + a8 XY + a9 Y 2 ;

• TSL12 with 12 dof for the N model

q̄ = a10 + a11 X + a12 Y yielding 12dof for the model N

t U = (u1 , u2 , u3 , q̄ 1 , u1 , u2 , u3 , q̄ 2 , u1 , u2 , u3 , q̄ 3 ).
e 1 1 1 2 2 2 3 3 3

15
3.1.4 The classical TmN element for the N model
We have
uα ∈ P1 , u3 ∈ P2 and q̄ ∈ P1 ,
t U = (u1 , u1 , u1 , q̄ 1 , u2 , u2 , u2 , q̄ 2 , u3 , u3 , u3 , q̄ 3 , u4 , u5 , u6 )
e 1 2 3 1 2 3 1 2 3 3 3 3
or
t U e = (u1 , u2 , u3 , u1 , u2 , u3 , u1 , u2 , u3 , u4 , u5 , u6 , q̄ 1 , q̄ 2 , q̄ 3 ).
1 1 1 2 2 2 3 3 3 3 3 3

3.1.5 Some usefull relations

Let x = (x1 , x2 ) be the coordinate system on the mid-surface S. Let m ∈ S be a generic point
and T ⊂ S a subdomain. Let f be a scalar function defined on T , then
Z Z √ Z
f (m)dS = f¯ Adx1 dx2 = f¯(x)dx, (3.8)
T T̄x T̄x

where T̄x is the domain of x ∈ R2 and A = det(Aαβ ).

5
 On a sphere for example (θ, ϕ) = (x1 , x2 ) = x. Consider a portion of the sphere of radius
R, θ ∈ [θ1 , θ2 ], ϕ ∈ [ϕ1 , ϕ2 ] and f (x) = 1 = k~nk. Then dS = R2 sin θdθdϕ, and
Z Z
f dS = R sin θdθdθ = R2 (ϕ2 − ϕ1 )(cos θ1 − cos θ2 ). (3.9)
T [θ1 ,θ2 ]×[ϕ1 ,ϕ2 ]

If ϕ1 = 0, ϕ2 = 2π, θ1 = 0, θ2 = π, then the answer is 4πR2 , which is the surface of the

sphere. Therefore, through the parametrization the integral is performed on a subdomain in R2 .
The subdomain is a triangle Tx in R2 if T is curved triangle on S.

3.1.6 The barycentric coordinates system

Consider a triangle T with summit nodes denoted M1 , M2 , M3 with respective coordinate
(x1 , y1 ), (x2 , y2 ), (x3 , y3 ). Let M (x, y) ∈ T then, there exist η 1 , η 2 , η 3 ∈ R such that x =
x1 η 1 + x2 η 2 + x3 η 3 = xi η i , y = yi η i and η 1 + η 2 + η 3 = 1.
The numbers η i constitute the barycentric coordinates of a point M ∈ T and verify:
1
η 1 (x, y) = (y32 x2 x − x32 y2 y), (3.10)
J
1
η 2 (x, y) = (y13 x3 x − x13 y3 y), (3.11)
J
1
η 3 (x, y) = (y21 x1 x − x21 y1 y), (3.12)
J
where J = x12 y23 − x23 y12 and xij = xi − xj , yij = yi − yj , xi x = xi − x, yi y = yi − y. So we
can rewrite:

x = x13 η 1 + x23 η 2 + x3 = x(η 1 , η 2 ), (3.13)

1 2 1 2
y = y13 η + y23 η + y3 = y(η , η ). (3.14)
(3.15)

We can observe that the relations are linear. So

x,αβ = y,αβ = 0, α, β = η 1 or η 2 (3.16)

1 2
η,αβ = η,αβ = 0, α, β = x or y (3.17)
(3.18)

∂η 1 1
1
= y12 x2 = a11 , (3.19)
∂x J
∂η 1 1
2
= − x32 y2 = a12 ,
∂x J
∂η 2 1
1
= y13 x3 = a21 ,
∂x J
∂η 2 1
2
= − x13 y3 = a22 .
∂x J

3.1.7 Relations between derivations

Let x = (x1 , x2 ) = (x, y). A function u(x) = u(x(η 1 , η 2 )) = ū(η̄) , η̄ = (η 1 , η 2 ).

6
 ∂u ∂ ū ∂η 1 ∂ ū ∂η 2 ∂ ū ∂ ū
= + = a11 1 + a21 2 (3.20)
∂x1 1
∂η ∂x 1 2
∂η ∂x 1 ∂η ∂η
∂ ū ∂ ū ∂η 1 ∂ ū ∂η 2 ∂ ū ∂ ū
= + 2 2 = a12 1 + a22 2
∂x2 1
∂η ∂x 2 ∂η ∂x ∂η ∂η

∂2u ∂ 2 ū ∂ 2 ū ∂2u
= a211 + (a11 a21 ) 1 2 + a221 2 2 (3.21)
∂x1 ∂x1 1
∂η ∂η 1 ∂η ∂η ∂η ∂η
∂2u 2
∂ ū ∂ 2 ū ∂ 2 ū
= a21 a11 1 1 + (a11 a12 + a22 a21 ) 1 2 + a12 a22 2 2
∂x1 ∂x2 ∂η ∂η ∂η ∂η ∂η ∂η
∂2u 2
∂ ū 2
∂ ū 2
∂ ū
= a212 1 1 + (a12 a22 ) 1 2 + a222 2 2
∂x2 ∂x2 ∂η ∂η ∂η ∂η ∂η ∂η

 

u
 ū
 ∂u 

 ∂ ū 


 ∂x1



 ∂η 1 

   
   
 ∂u    ∂ ū 
  1 0 0 0 0 0 
 ∂η 2

 ∂x2
  

 
  0 a11 a21 0 0 0 



 ∂2u
  
= 0 a12 a22 0 0 0   ∂ 2 ū
 
 (3.22)

 ∂x1 ∂x1
 
  0 0 0 a211 a11 a21 2
a21  
  ∂η 1 ∂η 1




 
 0 0 0 a21 a11 a11 a12 + a22 a21 a12 a22  


 ∂2u
  0 0 0 a212 a12 a22 a222  2

}  ∂ ū
  
  | {z 
 ∂x1 ∂x2  Bf  ∂η 1 ∂η 2 
   
   
 ∂2u
 ∂ 2 ū
  

∂x2 ∂x2 ∂η 2 ∂η 2
    
u1 1 0 0 0 0 0 ū1

 u1,1  
  0 a11 a21 0 0 0   ū
  1,1 


 u1,2  
  0 a12 a22 0 0 0 O   ū1,2 
  

 u2  
  0 0 0 1 0 0   ū 
 2 

 u2,1  
  0 0 0 0 a11 a21   ū
  2,1 

u2,2 0 0 0 0 a12 a22   ū2,2 
    
= (3.23)
  
u3   ū3 
  
  
u3,1   ū3,1 
    
  

u3,2
  
  ū3,2 

  
    
 u3,11   O Bf   ū3,11 
    
 u3,12     ū3,12 
u3,22 ū3,22
| {z } | {z }
De Dee

3.1.8 Discretization of components of stress and strain tensors

We have
 
Bm O e
De = Dee , = Bmf Dee (3.24)
O Bf

7
 e
where Bmf is a 12 × 12 matrix.
So
 
e11
Eee e
=  e22  = Ce De = Ce Bmf Dee (3.25)
2e12
 
K11
e e
EK =  K22  = CK De = CK Bmf Dee
2K12

In the above equations the vector Dee is given by:

Dee = Fe (η̄)Ue , (3.26)

where the 12 × 12 matrix Fe reads:

 
η1 η2 η3 0 0 0
 1 0 −1 0 0 0 
 
 0 1 −1 0 0 0 O 
 
 0 0 0 1
η η 2 η3 
 
 0 0
 0 1 0 −1 

0 0 0 0 1 −1
 
Fe =  (3.27)
 
η 1 η 2 η 3 4η 1 η 2 4η 1 η 3 4η 2 η 3 



 1 0 −1 4η 2 4η 3 − 4η 1 −4η 2 

0 1 −1 4η 1 −4η 1 4η 3 − 4η 2 
 

−8
 
 O 0 0 0 0 0 
 
 0 0 0 4 −4 −8 
0 0 0 0 0 −8

The nodal strain is given by:

 
11
e
F e (η) − zCK Bmf
e
F e Ue


e =  22  = Ce Bmf (3.28)
212

where Ce , CK , are 3 × 12 matrices and read:

−Γ111 1 0 −Γ211 0 0 −B11 0 0 0 0 0

 

Ce =  −Γ122 0 0 −Γ222 0 1 −B22 0 0 0 0 0  (3.29)

−2Γ112 0 1 −2Γ212 1 0 −2B12 0 0 0 0 0

K1 B11 K2 B12 B11 B11 + B11 B22 −Γ111 −Γ211 1 0 0

 
0 0
CK =  K3 0 B21 K4 B21 B22 + B12 −B21 B12 − B22 B22 −Γ122 −Γ222 0 0 1 
1 1
K5 2B2 2B1 K6 2B2 2 2B1 2 −2B1 B12 − 2B1 B22 −2Γ112 −2Γ212 0 −2 0
1 2

(3.30)
where
K1 = ∇1 B11 − B11 Γ111 − B12 Γ111 ,
K2 = ∇1 B11 − B11 Γ111 − B12 Γ111 ,
K3 = ∇2 B21 − B21 Γ112 − B21 Γ121 − B22 Γ212 , K4 = B21 Γ221 + ∇2 B22 − B21 Γ221 − B22 Γ222 ,
K5 = 2∇1 B21 − 2B21 Γ112 − 2B22 Γ112 − 2B11 Γ112 − 2B12 Γ122
K6 = −2∇1 B22 − 2B21 Γ211 − 2B22 Γ212 − 2B11 Γ212 − 2B12 Γ222

8
 The constitutive laws of stress tensor in the N-T model reads:
σ αβ = (ΛAαβ Aγδ + 2µAαγ Aβδ )γδ (U ) (3.31)
where the exact metric tensor Gαβ has been replaced by Aαβ in order to implement the best
first order variational equation. In the vector form, we have:
 11   
σ 11
 σ 22  = Cσ  22  (3.32)
σ 12 212
where the 3 × 3 matrix is:
(Λ + 2µ)A11 A11 ; ΛA11 A12 + 2µA12 A12 ; (Λ + 2µ)A11 A12
 

Cσ =  ΛA11 A22 + 2µA12 A12 ; (Λ + 2µ)A22 A22 ; (Λ + 2µ)A22 A12  (3.33)

12 11 12 22 12 12
(Λ + 2µ)A A ; (Λ + 2µ)A A ; ΛA A + µ(A A + A A ); 11 22 12 21

3.1.9 Discrete Variational formulation

In finite element (a triangle in this case), we have calculated the restriction of the stress and
strain tensors, as functions of shape functions and the local dof vector. The discrete variational
equation is obtained by calculating the virtual internal work and external work per element and
assembling over all the elements Te of the mesh Th as follows:
Z Z h+ X
A(u, v) = σ αβ (u)αβ (v)dSdz = L(v)|Te (3.34)
S h− Te ∈Th

Hereafter dΩ ≈ dSdz.
Let us calculate the internal virtual work. We have per element:
Z Z h+ Z Z h+
σ αβ (u)αβ (v)dSdz = σ e (u) : e (v)dSdz (3.35)
Te h− Te h−
In an element we begin by calculating on Gauss points
(i) geometric data
(Aαβ ), (Aαβ ), A = det(Aαβ ), Bαβ , Bβα , ∇ρ Bβα , ∇ρ Bαβ , Γ1αβ , Γ2αβ

(ii) The strain tensor

 
11 (U )
e (U ) =  22 (U )  = Ce Bmf
e
F e (η) − zCK Bmf
e
F e (η) Um

e
(3.36)
212 (U )
= (e − zeK )Um
e e

where
ee = Ce Bmf
e
F e (η), eK = CK Bmf
e
F e (η) (3.37)
Ume is the nodal vector, m is the number of dof which depends on the choice of the finite

element (m = 9dof for the shift Lagrange element, m = 12dof , for the classical element,
m = 18dof for the strain deformation element).

 
11 (V )
e (V ) =  22 (V )  = (Ce Bmf
e
F e (η) − zCK Bmf
e
F e (η))Vme (3.38)
212 (V )

9
((iii) Stress discretization

σ 11
   e 
11 (U )
σ e =  σ 22  = Cσ  e22 (U )  = Cσ (Ce Bmf
e
F e (η) − zCK Bmf
e
F e (η))Um
e
(3.39)
σ 12 e
212 (U )

Rigidity matrix
We have
Z Z h+ Z Z h+
e e
σ (u) :  (v)dzdS = Cσ (ee − zeK )Um
e
.(ee − zeK )Vme (3.40)
Te h− Te h−

We have:
Z Z h+ Z hZ h+ h
e e t e
σ (u) :  (v)dzdS = e Cσ ee − z t eK Cσ ee − z t ee Cσ eK (3.41)
Te h− Te h−
i i
+ z 2t eK Cσ eK dz Um
e
.Vme dS

which reduces in the case of constant thickness (h− = −h/2, h+ = h/2) to Me Um

e .V e :
m
Z Z −h/2
σ e (u)e (v)dzdS = Me Um
e
.Vme (3.42)
Te h/2

where ii √
h3 t e
Z h
Me = ht ee Cσ ee + K Cσ eK J Adη (3.43)
Tref 12
The element stiffness matrix Me is a m × m matrix according to the choice of the finite
elements, m = 12 in this case. Remind that Gauss formula are used for all integrations. The
elementary stiffness matrix Me is the contribution of the element Te to the global stiffness
matrix M. Let X denote the vector of all the dof. The global vector X can be rearranged
in three parts: X1 for unknown dof, X2 for dof that interact with other structures such as
soil-structure interaction on heavy duty (industrial foundations) and X3 for imposed dof or dof
that satisfy some constraints. After renumerotation of each part, the global matrix M is filled
with contributions from each element, i.e. the elementary stiffness matrix is immersed into M
by expressing Me in the global matrix M. Let Pe denote the position vector matrix such that
Ume = P e X, V e = P e X. Then the contribution of M denoted Mg is t P e M P and
m e e e e
X
M= Mge . (3.44)
Te ∈Th

Algorithmic for Rigidity matrix

Let Nd be the number of dof and Ne the number of elements, then P e is a m × Nd matrix and
Ne position matrices are needed to fill M. In practice the matrix P e is not stored. Instead, each
e or V e is referenced in the total dof vector by a connector N um(n , i) and M is filled
dof i in Um m e

10
as follows:

ne = 1, ....., Ne (3.45)
i, j = 1, ....m
I = N um(ne , i)
J = N um(ne , j)
M(I, J) = M(I, J) + Me (i, j)

The external virtual work per element reads:

Z Z Z
i i
(P vi dS + q vi dγ1 + (mt θν − mν θt )dγ (3.46)
Te γ1 ∩∂Te γ1 ∩∂Te

where θν = −(∇α v3 + Bαρ vρ )ν α and θt = θ.t.

We have

v1 = v11 η 1 + v12 η 2 + v13 (1 − η 1 − η 2 ) (3.47)

1 2 1 2
v2 = v21 η + v2 η + v23 (1 − η − η )
v3 = v31 η 1 + v32 η 2 + v33 (1 − η 1 − η 2 ) + v34 (4η 1 η 2 ) + v35 (4η 1 η 3 ) + v36 (4η 2 η 3 )

   1 2 3 
v1 η η η 0 0 0 0 0 0 0
 v2  =  0 0 0 η 1 η 2 η 3 0 0 0 0 0 0  Vme = Lem (η)Vme
v3 0 0 0 0 0 0 η η η 4η η 4η η 4η 2 η 3
1 2 3 1 2 1 3

(3.48)
!
Z Z √
P i vi dS = t e
L (η)[P i ] AJdη Vme = FSe .Vme , (3.49)
Te Tref

where J = 2|Te |, t [P i ] = [P 1 , P 2 , P 3 ].
Let us denote x1 = x and x2 = y. The coordinates of the three summit nodes of the
triangle Te are (x1 , y1 ), (x2 , y2 ) and (x3 , y3 ). We still denote xi − xj = xij and yi − yj = yij
and J = 2|Te |, where |Te | is the area of Te . The local coordinates or the barycentric coordinate
systems is still denoted η 1 , η 2 , η 3 and η 1 +η 2 +η 3 = 1, η̄ = (η 1 , η 2 ), x = x13 η 1 +x23 η 2 +x3 = x(η̄),
y = y13 η 1 + y23 η 2 + y3 = y(η̄). We also remind that if v(x, y) = v(x(η̄), y(η̄)) = v(η̄), then

∂v y23 ∂v̄ y13 ∂v̄ ∂v x23 ∂v̄ x13 ∂v̄

= 1
− 2
; =− 1
+ (3.50)
∂x J ∂η J ∂η ∂y J ∂η J ∂η 2
From the definition of θα = v3,α + Bα1 v1 + Bα2 v2 , we obtain:

θ1 = v3,1 + B11 v1 + B12 v2 , (3.51)

θ2 = v3,2 + B21 v1 + B22 v2 ,

We deduce that
 
θ1
= Leγ .Vme (3.52)
θ2

11
where the 2 × 12 matrix Leγ is defined as follows:

y23 − y13 y13 − y23

 
B11 η 1 B11 η 2 B11 η 3 B12 η 1 B12 η 2 B12 η 3 l1 B21 η 1 ....

 .... B 1 η 2 B 1 η 3 J J 

2 2
Leγ =
 
 B 2 η 1 B 2 η 2 B 2 η 3 − x23 x13 x23 − x13 4x23 η 2 x13 (4 − 8η 2 ) 
 2 2 2 l2 l3 .... 
J J J J J

.... l4 l5
(3.53)
4y23 η 2 − 4y13 η 1 4x23 η 1 − 4x13 η 2 x23 x 13 η 1
where l1 = , l2 = , l3 = − (4 − 8η 2 ) − , l4 =
J J J J
y23 4y13 1 4y23 2 y13
(4 − 8η 1 ) + η , l5 = − η − (4 − 8η 2 ).
J J J J
Remind that the term on the border do not concern all the dof in Vme . On a border of a
triangle only two coordinates are considered. r So we either have η 1 + η 2 = 1, η 1 + η 3 = 1 or
dxα dxβ
η 2 + η 3 = 1. The line element is dγ = Aαβ , where x(λ) is the curve on the border.
dλ dλ
For example consider a border between two nodes having (x1 , y1 ) and (x2 , y2 ) as coordinates.
Then from η 3 = 0 and η 1 + η 2 = 1, we have:

x = x1 η 1 + x2 η 2 = x12 η 1 + x2 (3.54)
1 2 1
y = y1 η + y2 η = y12 η + y2

dx dy
Therefore λ = η 1 and = x12 , = y12 , dγ 2 = A11 x212 + 2A12 x12 y12 + A22 y12
2 .
dη 1 dη 1

Algorithmic for source term

Finally, the external force per element is:

F e = FSe + Fγm
e
. (3.55)

The global vector force is filled as below:

ne = 1, ....., Ne (3.56)
i = 1, ...., m
I = N um(ne , i)
F (I) = F (I) + F e (i)

The final global force vector Fg is obtained by adding all the concentrated forces Fc :

Fg = F + Fc (3.57)

Linear problem
The final discrete problem is now
MX = Fg (3.58)
This equation is rearranged before the resolution. The total dof vector is subdivided in (X1 , X2 , X3 )
where X3 denotes imposed dof, X2 consists of dof that interacts with external structures such

12
as soil-structure interaction. X1 and X2 are unknown dof. The equation now reads.

M11 X1 + M12 X2 + M13 X3 = F1 (3.59)

M21 X1 + M22 X2 + M23 X3 = F2
M31 X1 + M32 X2 + M33 X3 = F3

F3 is the reaction on imposed dof. Suppose F2 is an elastic reaction due to soil for example and
F2 = −Ks X2 . The equation now reads

M11 X1 + M12 X2 = F1 − M13 X3 (3.60)

M21 X1 + (M22 + Ks )X2 = −M23 X3

X2 = −(M22 + Ks )−1 (M21 X1 − M23 X3 ) = −(M22 + Ks )−1 M21 X1 − (M22 + Ks )M23 X3 (3.61)

which leads to

M11 X1 − M12 (M22 + Ks )−1 X1 = F1 − M13 X3 + (M22 + Ks )−1 M23 X3

(M11 − M12 (M22 + Ks )−1 )X1 = Fd (3.62)

or
M11 X1 = Fd . (3.63)
This final equation can be solved by various linear methods. The unknown X2 is next
deduced. From X = (X1 , X2 , X3 ) all the others (eαβ , Kαβ , αβ , σ αβ ) are calculated locally.

Exercise 5.1
Calculate for an element the local strain and stress tensors of the N model by using the 15 dof
finite element. Calculate the elementary stiffness matrix and nodal force.

13