You are on page 1of 70

CIVE 751 Non-linear FE Analysis of Civil Engineering Structures

Civil Infrastructure and Environmental Engineering

Marwan Alaa Naeem – 100063027

Chapter 1: Preliminaries

and

Chapter 2: Non-linear Finite Element Analysis

1
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

2
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

3
1.2 A Review of Concepts from Linear Algebra
• Scalar: physical quantity with a value, irrespective of a direction.
– For example: speed, length, area, and mass.

– Denoted by italic symbols. For example: v.

• Vector: physical quantity with both a value and direction.


– For example: force, velocity, acceleration, and weight.

– Denoted by a roman, bold-faced, lower-case symbol. For example: v

– Vector v may have n scalar entries, so that:

v=
(1.13)

• Scalar entries can be written in a row (transpose of vector v):

v = (v1, … , vn)
4
1.2 A Review of Concepts from Linear Algebra
• Addition of vectors:
w=u+v (1.14)

– Where wi = ui + vi for i = 1, … , n.

• Multiplication of a vector by a scalar (λ) is:


w = λu (1.15)

– Where wi = λui.

• Inner/scalar product between two vectors (u and v):


uTv = (1.16)

– Commutativity: uTv = vTv

5
1.2 A Review of Concepts from Linear Algebra

• Euclidian or L2-norm:
2 = (1.17)

• The cross product (vector product) of two vectors a and b, forms vector c:
𝐜=𝐚 x 𝐛 (1.18)

– Where c is orthogonal to a and b. Hence, b x a = -a x b, and:

( )
𝑎 2𝑏3 − 𝑎3 𝑏2
c  =  𝑎 3 𝑏 1− 𝑎1 𝑏3 (1.19)
𝑎1 𝑏 2− 𝑎2 𝑏1

• Let the scalar-valued function f be a function of the components ai of the


vector a. So, the gradient b is the differentiation of f with respect to a.

𝜕𝑓 𝜕𝑓
𝒃= (1.20) 𝒃 𝒊=
𝜕𝐚 or in component form:
𝜕𝒂 𝒊 (1.21)
6
1.2 A Review of Concepts from Linear Algebra
• The multiplication of matrix A and vector b results in vector c:
c = Ab (1.22)

– Where ci = . (1.23)

• The addition of matrices is cij = aij + bij .

• The multiplication of matrix by a scalar (λ) is cij = λaij.

• The product of a (m x k) matrix A and a (k x n) matrix B results in a (m x


n) matrix C:
C = AB

(1.24)
– Where cij = .

• If (k = 1) in both matrices, the result of multiplying A by B is m x n matrix:


C = abT (1.25)
7
1.2 A Review of Concepts from Linear Algebra
• Note that:
(AB)T = BTAT (1.26)

• At certain conditions, the inverse B = A-1 can be defined as:


AB = I (1.27)

– Where I is the unit matrix (all entries are zeroes except for diagonal entries).

• The inversion of matrices is required to solve large systems of linear


equations.
– as a result of finite element discretization.

– Such systems have the form:

a11x1 + a12x2 + . . . + a1nxn = b1


a21x1 + a22x2 + . . . + a2nxn = b2
(1.28)
…+…+…=…
an1x1 + an2x2 + . . . + annxn = bn
8
1.2 A Review of Concepts from Linear Algebra

• When coefficients ann are assembled in matrix A and components bn in


vector b, Eq 1.28 can be re-written as:
Ax = b (1.29)

x =A -1
b (1.30)
• and unknowns x can be obtained through:

• Also, matrix A can be decomposed (1.31)


as:
A = LDU

[ ] [ ]
1 0 0 1 𝑢12 𝑢1𝑛
L– =Where:
𝑙 21 1 0 U  = 0 1 𝑢2𝑛 D  =diag[ d11, …,𝑑𝑛𝑛¿
𝑙 𝑛1 𝑙 𝑛2 1 0 0 1

(1.32) (1.33) (1.34)


9
1.2 A Review of Concepts from Linear Algebra
• LDU is based on Gauss elimination. Matrices L and U can also be linked
as:
x = (LDU)-1b = U-1(LD)-1b = U-1D-1L-1b

– Therefore, x can be solved as:

c = L-1b
d = D-1c (1.35)
x = U-1d

• Note that L-1b and U-1d involve multiplication, but D-1c consists of divisions.

• Since D-1 = diag[d11-1 , … , dnn-1].

• Determinant of matrix is:


detA = (1.36)

– Where Aij is an (n – 1) x (n – 1) matrix.


10
1.2 A Review of Concepts from Linear Algebra
• Singularity of a matrix is also closely related to its determinant:

Det(AB) = detA . detB


detA = detL . detD . detU
detL = detU = 1

detA = (1.37)
• Sherman-Morrison formula:
– Let A be a non-singular n x n matrix, and let u and v be two vectors with n
entireties. Then:
(A + uvT)-1 = A-1 - (1.38)

11
1.2 A Review of Concepts from Linear Algebra
• Gauss’ divergence theorem, where volume integral can be transformed
into a surface integral:
(1.39)
∫¿ 𝐯 d 𝑉 =∫ 𝐧
T
𝐯 d 𝑆
V S

– Where n is the outward normal to the bounding surface, and div is:

divv (1.40)

• The Einstein summation convention can replace the summation in Eq 1.24


to be:
cij = aie + bej (1.41)

– Where summation is with respect to repeated index e.

– Called ‘dummy’ index.

12
1.2 A Review of Concepts from Linear Algebra
• Re-writing Gauss’ theorem in index notation:

𝜕 𝑣𝑖
∫ 𝜕 𝑥𝑖
d 𝑉 =∫ 𝑛 𝑖 𝑣 𝑖 d 𝑆
V S

– With an important quantity (Kronecker delta):

{ 𝛿 𝑖𝑗=1 if 𝑖= 𝑗
𝛿 𝑖𝑗=0 if 𝑖 ≠ 𝑗
(1.42)

• Recently, index notation has been replaced by direct tensor notation, so


Eq 1.24 is denoted as:
C=A.B (1.43)

– Where the single dot represents single contraction.

13
1.2 A Review of Concepts from Linear Algebra
• Similarly, double contraction is:
C=A:B (1.44)

– Or using index notation: c = aiebei.

• The gradient is denoted as , as following:


b=f (1.45)

– Which is equal to the gradient vector in Eq 1.20. So, Gauss’ theorem can be:

∫ ∇. 𝐯 d 𝑉 =∫ 𝐧 . 𝐯 d 𝑆
V S

• The dyadic product of vectors a and b is:


(1.46)
C=a b

14
1.2 A Review of Concepts from Linear Algebra

• With components cij = aibj. The divergence operator of a second-order


tensor A is:
a = .A (1.47)

– Such that:
aj = (1.48)

– And its trace:


c = tr(A) (1.49)

– Through c = aii.

15
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

16
1.3 Vectors and Tensors
• Since a vector has a magnitude and a direction. It is important to the
change of its component if represented in different coordinate system.
𝑦 y
𝑛2

𝑛2
𝑥
𝑛1

𝑛1 x
A vector in original x,y-coordinate system and rotated coordinate system

• Components of vector in coordinate system can be obtained in the x,y-


coordinate system by the transformation:
= Rn (1.50)

17
1.3 Vectors and Tensors
• The transformation matrix R is:
R= (1.51)

– That has a special structure:


R-1 = RT (1.52)
• A matrix that satisfies requirement 1.52 are called orthogonal.
• Which det(R) = 1.

• Tensors are physical quantities that relate vectors.


– For example: stress tensor related the force on a plane and the normal vector
of that plane.

• All tensors are matrices but NOT vice versa.

18
1.3 Vectors and Tensors
• Suppose that the second-order tensor C relates the vectors, or first-order
tensors, t and n:
t = Cn (1.53)

• In the frame, the second-order tensor related


= (1.54)

• Substituting Eq 1.50 and identical relation into Eq 1.54:


=RCRT (1.55)

• Using Eq 1.51, the identity can for two dimensions is:


= c11 cos2 + (c12 + c21) cos sin + c22 sin2
11

22 = c11 sin - (c12 + c21) cos sin + c22 cos (1.56)


2 2

12 = -c11 cos sin + c12cos - c21 sin + c22 cos sin


2 2

21 = -c11 cos sin - c12cos + c21 cos + c22 cos sin


2 2

19
1.3 Vectors and Tensors

• For symmetric second-order tensors, c21 = c12, and also 21 = 12.


2c12

Principal directions of a second-order tensor

2
c11 – c22
• The external values 11 and 22 are of interest relatively to the value of .

– Since 11 and 22 are functions of inclination angle and obtained through:


= 0 or = 0 (1.57)

– Therefore, the diagonal tensor components attain extremal values for:


tan 2 (1.58)

• Re-writing Eq 1.56:
= ½ (c11 + c22) + ½ (c11 - c22) cos 2+ c12 sin 2 (1.59)
11 20
1.3 Vectors and Tensors
• From the figure before and Eq 1.58, we infer that:
sin 2

(1.60)
cos 2

• The closed-form expression for the principles values:

{
1 1
𝑐 11  = ( c 11  +  c 22 )   − √(𝑐 11− 𝑐 22) − 4 𝑐 212
2
2 2
(1.61)
1 1
𝑐 22  = ( c 11  +  c 22 )  + √(𝑐 11 −𝑐 22)2 − 4 𝑐 2 12
2 2
• For symmetric second-order tensors, the off-diagonal components are
zero for this angle: = o, so:
+ c12 cos 2 (1.62)
21
1.3 Vectors and Tensors
• Let C be a symmetric second-order tensor, e be a vector.
– The product Ce will be parallel with e.

– But, for each tensor there is a coordinate system where the resulting vector is
parallel to the original vector:
Ce = λe (1.63)

– With λ the scalar-valued eigenvalue. Eq 1.63 can be:


(C – λI)e = 0 (1.64)

– With I = diag[1, … , 1] the unit matrix. A non-trivial solution (e ≠ 0) then exists


if and only if the determinant of C − λI vanishes:

Det[C – λI] = 0 (1.65)


• A direct relationship can be established between the product of all
eigenvalues and the determinant of a matrix:

22
1.3 Vectors and Tensors
• A direct relationship can be established between the product of all
eigenvalues and the determinant of a matrix:
detC = (1.66)

– Know as Vieta’s rule.

– Valid for symmetric and non-symmetric matrices.

• Inverting Eq 1.55:
C=RT R (1.67)

– In the principal axes:

= (1.68)

– And = λ1 and = λ2 the principal values or eigenvalues of C.

23
1.3 Vectors and Tensors
• Applying Eq 1.51 in Eq 1.67 yields:
= λ1 + λ2

or
= λ1 () + λ2 (sin , - cos )

• Identifying = (cos , sin ) and = (sin , -cos ) as eigenvalues, so C can be


through spectral decomposition:
= (1.69)

• Where a generalization to n dimensions. The eigenprojections:


Ei = (1.70)

• The spectral decomposition of a symmetric, second-order tensor can be:


= Ei (1.71)

24
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

25
1.4 Stress and Strain Tensors
• Study the response of a body to forces exerted on it.

Force acting on an imaginary cut in a solid body (source: Non-


linear Finite Element Analysis Of Solids And Structures)

• At a small area (ΔS) with acting force (Δf) on. The stress vector t is:

∆f 𝑑f
𝑡= lim = (1.72)
∆ 𝑆→ 0 ∆ 𝑆 𝑑𝑆
– t is decomposed in a component that acts along the normal to that plane and
in two mutually orthogonal vectors denoted as σxx.

– The other component lying in the same plane σxy and σxz, in which the second
subscript refers the direction of the stress component. 26
1.4 Stress and Strain Tensors
• In 3D, nine stress component exist:

Stress tensor, σ is a 2nd order tensor


(1.73)

• Stress tensor σ is related to the stress vector t acting on a plane with


normal n. In matrix-vector notation, the relation ship is:
(1.74)

• Let the normal vector be parallel to x-axis (𝑛T=[1,0,0]), 𝑦 𝑎𝑥𝑖𝑠


(𝑛T=[0,1,0]), 𝑧−𝑎𝑥𝑖𝑠 𝑛T=[0,0,1]), the analogue of Eq 1.74 can be written
as:
– Index notation: (1.75)
– Direct tensor notation:

27
1.4 Stress and Strain Tensors
• For a nonpolar or Boltzmann continuum, shear stress is:
=
= (1.76)
=

• There are six independent stress components and the matrix


representation of the symmetric stress tensor σ can be written as:

[ ]
𝜎 𝑥𝑥 𝜎 𝑥𝑦 𝜎 𝑧𝑥
∑= 𝜎 𝑥𝑦 𝜎 𝑦𝑦 𝜎 𝑦𝑧 (1.77)
𝜎 𝑧𝑥 𝜎 𝑦𝑧 𝜎 𝑧𝑧
• There are only six independent stress components makes it also feasible to
write the stress tensor in a vector form (Voigt notation):

),,,,, (1.78)

28
1.4 Stress and Strain Tensors
• In geotechnics, for example, normal stress (, , ) is decomposed into a deviatoric
and hydrostatic stresses.
– The deviatoric changes in the shape.

– The hydrostatic changes the volume.


1
𝑝= (𝜎 𝑥𝑥 +  𝜎 𝑦𝑦 +  𝜎 𝑧𝑧 ) (1.79)
3
• The deviatoric stress tensor is:
– In matrix representation: (1.80)
– In Voigt’s notation: (1.81)
– Where:

,,,,,)
(1.82)
29
1.4 Stress and Strain Tensors
• The principle values, λ of a second-order tensor are computed based on:
det( λI) = 0 (1.83)

• Or in component form: =0

(1.84)

• With identities:

(1.85)
+2

• Eq 1.84 can be:


(1.86)

30
1.4 Stress and Strain Tensors
• , are invariants of the stress tensor (have the same value irrespective of the
(1.86)
reference frame).

• Computing the invariants by solving the cubic equation, values of


deviatoric stress tensor are obtained:
λ 3 − 𝐽 2 λ − 𝐽 3=0 (1.87)

– Where:
(1.88)
+2
(1.89)

• It can be said that (Fung 1965):

(1.90)

31
1.4 Stress and Strain Tensors
• Consider an elementary cube (in x,y-plane), as shown:

Deformation of elementary cube’s face (source: Non- linear


Finite Element Analysis Of Solids And Structures)

• In the limiting case that Δx→0 and Δy→0, the strains in the x – and y-
directions become:

(1.91)
32
1.4 Stress and Strain Tensors
• The distortion of the elementary square in the x,y-plane is given by:
𝐶 𝐵
∆𝑢 ∆ 𝑣 𝜕𝑢 𝜕𝑣
𝛾 𝑥𝑦 = lim + = + (1.92)
∆ 𝑥→0,Δ 𝑦→0 ∆ 𝑦 ∆ 𝑥 𝜕𝑦 𝑑𝑥
• Rotation is given by:

( )( )
𝐵 𝑐
1 ∆𝑣 ∆𝑢 1 𝜕𝑣 𝜕𝑢
𝑤 𝑥𝑦 = lim − = − (1.93)
∆ 𝑥→0,Δ 𝑦→0 2 ∆ 𝑥 ∆ 𝑦 2 𝜕𝑥 𝑑 𝑦
• Using the following:
− Tensorial shear strain: =

− Normal strain:

− Shear strain: and and

33
1.4 Stress and Strain Tensors
• The strain components are:

(1.94)

– Where w is the displacement in z-direction.

[ ]
• The strain tensor is: 𝜀 𝑥𝑥 𝜀 𝑦𝑥 𝜀 𝑧𝑥
𝜀 = 𝜀 𝑥𝑦 𝜀 𝑦𝑦 𝜀 𝑧𝑦 (1.95)
𝜀 𝑥𝑧 𝜀 𝑦𝑧 𝜀 𝑧𝑧
– Note that strain tensor is symmetric.

– So, it can be in Voigt notation:

,,,,,) (1.96)
34
1.4 Stress and Strain Tensors
• The rate of internal energy in Voigt notation, direct tensor, and index
notation, respectively:
˙ ˙𝑇
𝑊 𝑖𝑛𝑡 = 𝜖 𝜎 =𝜖˙ : 𝜎= 𝜖˙𝑖𝑗 𝜎 𝑗𝑖 (1.97)

• The volumetric strain is:


+ + (1.98)

• The deviatoric strain tensor is:


I (1.99)

– In Voigt’s notation:
i (1.100)

– Where eT is:
,,,,,) (1.101)

35
1.4 Stress and Strain Tensors
• For 2D cases:
𝜎 =𝑇 𝜎 𝜎 (1.102)

– Where:

[ ]
cos 2 𝜙 sin 2 𝜙 2 sin 𝜙 cos 𝜙 (1.103)
𝑇 𝜎= sin 2 𝜙 cos 2 𝜙 − 2sin 𝜙 cos 𝜙
−sin 𝜙 cos 𝜙 sin 𝜙 cos 𝜙 cos 2 𝜙 − sin2 𝜙
– Where , , for plane stress conditions and by substituting (- ) in the latter equation,
then therefore:

(1.104)

36
1.4 Stress and Strain Tensors
• For the strain transformation:
𝝐 =𝑇 𝜖 𝝐 (1.105)

– Where:

[ ]
cos 2 𝜙 sin 2 𝜙 sin 𝜙 cos 𝜙 (1.106)
𝑇 𝜀= sin 2 𝜙 cos 2 𝜙 −sin 𝜙 cos 𝜙
−2 sin 𝜙 cos 𝜙 2 sin 𝜙 cos 𝜙 cos 2 𝜙 −sin 2 𝜙
– Where , ,

– For stress transformation:

𝑇
𝜖 =𝑇 𝜖 𝜖 (1.107)

37
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

38
1.5 Elasticity
• The constitutive equation relates the stress tensor σ to the strain tensor .

∈ = 𝐂e : σ (1.108)
• Or in the inverse form:
(1.109)

– Where De is the stiffness tensor.

• The compliance relation in Voigt’s notation:

(1.110)

– With E as Young’s modulus and v as Poisson's ratio.

39
1.5 Elasticity
• Eq 1.110 can be compacted as:
∈ = 𝐂e σ (1.111)

– Where Ce is the elastic compliance matrix.

• Eq 1.110 presents the strain tensor, as a function of the stress tensor, σ.


• To obtain inverse relation we rewrite the first three equations as:

(1.112)

• Adding the above equations and using Eq 1.79 and 1.98, we obtain:

(1.113)
40
1.5 Elasticity
• Where K is the bulk modulus that relates the volumetric strain and the
hydrostatic pressure:
𝐸
𝐾= (1.114)
3 (1 − 2 𝑣 )

• Substituting Eq 1.113 into 1.112 and inversion yields ‘the elastic stiffness
relation’:

(1.115)

– Where:
(1.116)
41
1.5 Elasticity
• The above stiffness relation is:
σ  = D e ∈ (1.117)

• By plugging Eq 1.116 into 1.115:

[ ]
0 0 0
1− 𝑣 𝑣 𝑣
0 0 0
𝑣 1− 𝑣 𝑣
0 0 0
e 𝐸 𝑣 𝑣 1 −𝑣
D = 1 −2 𝑣 0 0 (1.118)
(1+ 𝑣)(1 − 2 𝑣 ) 0 0 0
2 1− 2 𝑣 0
0 0 0
0 2 1 −2 𝑣
0 0 0
0 0 2

42
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

43
2.1 Equilibrium and Virtual Work
• Let a body V with boundary S, is in the balance of momentum. Where the
stress vector t and gravity acceleration as vector g, the linear momentum
balance is:

∫ 𝐭 𝑑𝑆+∫ 𝜌 𝐠 𝑑𝑉 =∫ 𝜌 𝐮¨ 𝑑𝑉 (2.1)
𝑆 𝑉 𝑉

– Where is the density, and the dots represents differentiation with respect to
time.
– Using Eq 1.75, the above equation yields so:

∫ 𝐧 ⋅𝜎 𝑑𝑆+∫ 𝜌 𝐠 𝑑𝑉 =∫ 𝜌 𝐮¨ 𝑑𝑉 (2.2)
𝑆 𝑉 𝑉

– Where n is the outward unit vector at the boundary of the body.

44
2.1 Equilibrium and Virtual Work
• The divergence theorem is used to get:

∫ (∇⋅𝜎 ¿+𝜌𝐠−𝜌 𝐮)𝑑𝑉=0¿


¨ (2.3)
𝑉

• Recalling:
∇⋅ 𝜎 +𝜌 𝐠=𝜌 𝐮¨ (2.4)

• Introducing the operator matrix L:

[ ]
𝜕 𝜕 𝜕
0 0 0
𝜕𝑥 𝜕𝑦 𝜕𝑧
(2.5)
𝑇
𝐿 = 0 𝜕 𝜕 𝜕
0 0
𝜕𝑦 𝜕𝑥 𝜕𝑧
𝜕 𝜕 𝜕
0 0 0
𝜕𝑧 𝜕𝑦 𝜕𝑥
45
2.1 Equilibrium and Virtual Work
• With Eq 2.5, Eq 2.4 can be compacted to:

𝑳𝑇 𝜎 +𝜌 𝐠=𝜌 𝒖¨ (2.6)
• Then, Eq 2.6 is to be transformed into a weak form. By multiplying the
equation by the virtual displacement field and integrate over V:
∫ 𝛿𝐮 (𝐋 𝜎+𝜌𝐠−𝜌 ¿𝒖)𝑑𝑉=0¿
𝑇 𝑇
¨ (2.7)
𝑉
• Applying the divergence theorem:

∫ ¿ ¿ (2.8)
𝑉

– With boundary conditions

– and initial conditions: 0 .

46
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

47
2.2 Spatial Discretisation by Finite Elements
• In this section, a pure displacement-based formulation is adopted. In
which the displacements at nodes of elements are considered as the
fundamental unknowns.
• The continuous displacement field u elementwise as:

𝑢=∑𝑛𝑘=1 h𝑘 (𝜉 , 𝜂 , 𝜁 )𝑎 𝑘 (2.9)

– Where are “shape functions” or “interpolation functions” of an element that is


supported by n nodes.
– The interpolation functions are polynomials expressed in terms of .

48
2.2 Spatial Discretisation by Finite Elements
• Consider vector in which all displacement degrees of freedom of the nodes
connected to this element are assembled as:

()
𝑎1
𝑎2
𝑎 𝑒 = ... (2.10)
...
𝑎𝑛

• In a 3 x 3n matrix H:

[ ]
h1 0 0 h2 0 0 ... ... h𝑛 0 0
𝐻= 0 h1 0 0 h2 0 ... ... 0 h𝑛 0 (2.11)
0 0 h1 0 0 h2 ... ... 0 0 h𝑛
• The interpolation of the continuous displacement field for all points within
an element can be written in a compact manner is:

𝐮=𝐇 𝑎𝑒 (2.12)
49
2.2 Spatial Discretisation by Finite Elements
• Consider vector in which all displacement degrees of freedom of the nodes
connected to this element are assembled as:

()
𝑎1
𝑎2
𝑎 𝑒 = ... (2.10)
...
𝑎𝑛

• In a 3 x 3n matrix H:

[ ]
h1 0 0 h2 0 0 ... ... h𝑛 0 0
𝐻= 0 h1 0 0 h2 0 ... ... 0 h𝑛 0 (2.11)
0 0 h1 0 0 h2 ... ... 0 0 h𝑛
• The interpolation of the continuous displacement field for all points within
an element can be written in a compact manner is:

𝐮=𝐇𝐚 𝑒 (2.12)
50
2.2 Spatial Discretisation by Finite Elements
• The displacements in the element related vector can be related to the
global displacements contained in vector a through an incidence or
location matrix :
𝑎𝑒=𝑍 𝑒 𝑎 (2.13)

• From the previous formulas, the weak form of the balance of momentum
can be reformulated as:

(2.14)

– The global nodal virtual displacements as found in which are independent of


the spatial coordinates and can be brought out of the integral.

51
2.2 Spatial Discretisation by Finite Elements
• For any virtual displacement at the semi-discrete balance of momentum:
..
𝑀 𝑎= 𝑓 𝑒𝑥𝑡 − 𝑓 𝑖𝑛𝑡
(2.15)
– With the mass matrix :
𝑇 𝑇
𝑴=¿𝑒=1¿𝑛 𝒁 ∫ 𝜌𝑯 𝑯 𝑑𝑉 𝒁 𝑒
𝑒 𝑒 𝑉𝑒
(2.16)

– And external force vector:


𝑇 𝑇 𝑇 𝑇
𝐟 𝑒𝑥𝑡 =¿𝑒=1¿𝑛 𝐙 ∫ 𝜌𝐇 𝑔𝑑𝑉+¿𝑒=1¿𝑛 𝐙 ∫ 𝐇 𝑡𝑑𝑆
𝑒 𝑒 𝑉𝑒 𝑒 𝑒 𝑆𝑒
(2.17)

– And internal force vector:

𝐟 𝑖𝑛𝑡=¿𝑒=1¿𝑛𝑒 𝒁 𝑇𝑒 ∫ 𝑉 𝑩𝑇 𝜎𝑑𝑉
𝑒
(2.18)
– And:
𝐁=𝐋𝐇 (2.19)

52
2.2 Spatial Discretisation by Finite Elements
• To facilitate that , a mapping from a cuboidal parent element in a
coordinate system onto the actual, arbitrary geometry on the element in
the x, y, z- coordinate system.
• If ] and if we have:
𝑥=𝑥( 𝜉 ) (2.20)

• Using standard transformation, the integration domain in x, y, z


coordinates can be converted into the simple cuboidal domain of the
parent element as:
𝑛𝑒 +1 +1 +1
𝑓 𝑖𝑛𝑡 =∑ 𝑍 𝑒 ∫ ∫ ∫ (𝑑𝑒𝑡𝐽 )𝐵 𝜎 𝑑𝜉 𝑑𝜂 𝑑𝜁
𝑇 𝑇
(2.21)
𝑒=1 −1 − 1 −1

53
2.2 Spatial Discretisation by Finite Elements
• J is the Jacobian Matrix of the mapping 𝒙=𝒙(𝝃) as defined through:
𝜕 𝑥
𝐽 =
𝜕 𝜉 (2.22)
– In component form:

[ ]
𝜕𝑥 𝜕𝑥 𝜕𝑥
𝜕𝜉 𝜕𝜂 𝜕𝜁
(2.23)
𝜕𝑦 𝜕𝑦 𝜕𝑦
𝐉=
𝜕𝜉 𝜕𝜂 𝜕𝜁
– By interpolation:
𝜕𝑧 𝜕𝑧 𝜕𝑧
𝜕𝜉 𝜕𝜂 𝜕𝜁
(2.24)

– With the set being spatial coordinates of node k.

54
2.2 Spatial Discretisation by Finite Elements
• We shall adhere the isoperimetric concept, then the Jacobian Matrix takes
the following form:
𝑛
𝜕 h𝑘
𝐽 =∑ 𝐱 𝑘𝑇
𝑘= 1 𝜕 𝜉 (2.25)

• Numerical Integration is applied to compute the integral of the equation:

(2.26)
– With the weight factor of integration point i, and ni the number of integration
points in element e.

55
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

56
2.4 Incremental-iterative Analysis
• For quasi-static process equation:
𝐟 ext − 𝐟 ∫ ¿ =𝟎 ¿ (2.27)

• Decomposing the vector of unknown stress components at time t + ∆t:


𝜎 𝑡 + Δ 𝑡 =𝜎 𝑡 + Δ 𝜎 (2.28)

• Substituting in Eq 2.27 using Eq 2.18:


𝑡+Δ 𝑡 𝑛𝑒 T T 𝑡 𝑛𝑒 T T
𝐟 ext − ∑𝑒=1 𝐙𝑒 ∫ 𝑉 𝐁 𝜎   d 𝑉 − ∑𝑒=1 𝐙𝑒 ∫ 𝑉 𝐁 Δ 𝜎 d 𝑉 =𝟎
𝑒 𝑒
(2.29)
• Rearranging:
𝑛𝑒

∑ ❑ 𝐙𝑒 ∫❑𝐁 Δ 𝜎 d 𝑉 =𝐟 ext − 𝐟 ∫ ¿ ¿
T T 𝑡+ Δ 𝑡
𝑡 (2.30)
𝑒=1 𝑉

– Where the superscript t is attached to to underscore that the internal force is


evaluated at time t.
– Similarly, t + ∆t to .
57
2.4 Incremental-iterative Analysis
• Newton Raphson Method aids in linearising the dependence of the stress
increment on the displacement and strain:
Δ 𝜎 =Δ 𝜎 ( Δ 𝜖 ( Δ 𝐮)) (2.31)

• The stress increment can be linearized as:


(2.32)

• Defining:

( )
𝑡
𝜕𝜎
D= (2.33)
𝜕𝜖
• The material tangential stiffness matrix:

𝛿𝜎 =𝐃 𝛿 𝝐 (2.34)
• The internal work:
𝛿 𝑊 int   =∫ ❑ 𝛿 𝜖 T 𝜎 d 𝑉 (2.35)
𝑉
58
2.4 Incremental-iterative Analysis
• The kinematic relation between the variation of the strain tensor and the
continuous displacement field is:
𝛿 𝝐 =𝐋 𝛿 𝐮 (2.36)

• For quasi-static loading conditions:


∫ ❑𝛿𝜖 T
𝜎 d𝑉 =∫ ❑𝜌𝛿𝐮 T
𝐠d 𝑉 +∫ 𝐭d𝑆
❑𝛿𝐮 T
(2.37)
𝑉 𝑉 𝑆
• Using Eq 2.36, the stress variation is:
δσ = DLδu (2.38)

• After inserting Eq 2.12:


(2.39)

– Or using Eq 2.13
(2.40)

59
2.4 Incremental-iterative Analysis
• Inserting this result into Eq 2.30 and using Eq 2.19 to introduce B yields
the linearised equation for a finite load increment:
𝑛𝑒

∑ ❑𝐙𝑒 ∫ ❑𝐁 𝐃𝐁𝐙𝑒 Δ𝐚 𝑑𝑉 =𝐟 𝑒𝑥𝑡 − 𝐟 𝑖𝑛𝑡


T T 𝑡 +Δ 𝑡 𝑡
(2.41)
𝑒=1 𝑉𝑒

• Since incremental nodal displacements do not depend upon the spatial


coordinates, they can be brought outside the integral and one obtains the
following linearized set of N equations:
𝐊 Δ 𝐚 =𝐟 𝑡𝑒𝑥𝑡
+Δ 𝑡
− 𝐟 𝑡int   (2.42)

• The tangential stiffness matrix, K upon small increment loading is:


(2.43)

60
2.4 Incremental-iterative Analysis
• Eq 2.42 can be written as:
´ +𝐟 𝑡 − 𝐟 𝑡
𝐊 Δ𝐚=Δ 𝐟 ext 𝑒𝑥𝑡 𝑖𝑛𝑡 (2.44)
– Where is the normalised external load vector.

• We can rewrite the load increment as with a scalar-valued incremental


load parameter, so:

𝐊 Δ 𝐚 = Δ 𝜆 𝐟´ ext +𝐟 𝑡𝑒𝑥𝑡 − 𝐟 𝑡𝑖𝑛𝑡 (2.45)

61
2.4 Incremental-iterative Analysis
• This figure represents linearization which yields to drifting from true
solution (equilibrium state).
– Especially with large increments.

• In an incremental-iterative solution method a first estimate for the


displacement increment, is made through:
Δ 𝐚1 =𝐊 −
0
1
𝐫0 (2.46)

Purely incremental solution procedure (source: Non- linear


Finite Element Analysis Of Solids And Structures)
62
2.4 Incremental-iterative Analysis
• With: 𝐫 0=𝐟 𝑡ext+ Δ𝑡  − 𝐟 int   ,0 (2.47)

• The residual factor, or out-of-balance vector at the beginning of the load


increment.
• The subscript 1 of ∆a signifies that we deal with the estimate in the first
iteration for the incremental displacement vector.
• The subscript 0 of the internal force vector relates to the fact that this
vector is calculated using the stresses at the beginning of the loading step:
𝑛𝑒 T 𝑛𝑖 T
𝐟 ∫ , 0=∑𝑒=1 𝐙𝑒 ∑𝑖 =1 𝑤𝑖 ( de 𝑡 𝐉𝑖 ) 𝐁 𝑖 𝜎 𝑖 ,0 (2.48)

• From the incremental displacement vector a first estimate for the strain
increment 1 can be calculated, using the stress–strain law, the stress
increment can be computed:
𝟏 =  𝟎 +  𝟏 (2.49)
63
2.4 Incremental-iterative Analysis
• The internal force vector that is computed based on the stresses is not in
equilibrium with the external loads .
– For this reason, a correction to the displacement increment is necessary.

• Donating this correction by da2:

with (2.50)
• the updated tangential stiffness matrix
• The displacement increment after the second iteration in the loading step
follows from:
Δ 𝐚2 = Δ 𝐚1 +d 𝐚 2 (2.51)

64
2.4 Incremental-iterative Analysis
• In a similar fashion to the calculation of the strain and stress increment in the first
iteration the quantities and are now computed.
• From the latter quantity an improved approximation for the stress at the end of
the loading step, can be made. This process can be summarized as:

(2.52)

65
2.4 Incremental-iterative Analysis
• Delta-incremental method is less robust, particularly when we have materially
non-linear models.
– In which different behavior in loading than in unloading is noticed”, e.g.
plasticity.

• The implicit assumption that the


tangential stiffness matrix is updated
after each iteration.
• As long as the stresses are determined in
a proper manner and the resulting
internal force vector is computed on the Modified Newton-Raphson iteration scheme (source: Non-
linear Finite Element Analysis Of Solids And Structures)

basis of these stresses and inserted on the


right-hand side of the equation of
66
Agenda
• Chapter 1: Preliminaries
– Section 1.2: A Review of Concepts from Linear Algebra

– Section 1.3: Vectors and Tensors

– Section 1.4: Stress and Strain Tensors

– Section 1.5: Elasticity

• Chapter 2: Non-linear Finite Element Analysis


– Section 2.1: Equilibrium and Virtual Work

– Section 2.2: Spatial Discretisation by Finite Elements

– Section 2.4: Incremental-iterative Analysis

– Section 2.5: Load versus Displacement Control

67
2.5 Load versus Displacement Control
• As an alternative approach, displacement increments can be prescribed
instead of load.
• Stress development within the system which in turn results in nodal forces
at the nodes where the displacement is prescribed.
• This approach is preferred because:
– The tangential stiffness matrix is better conditioned for displacement control
than for load control resulting in faster convergence of the iterative process.
– Under load control, the tangential stiffness matrix becomes singular at a limit
point in the load-deflection diagram, not only when a global failure occurs, but
also, we have a local maximum along this curve.

68
2.5 Load versus Displacement Control
• Decomposing the incremental displacement vector, into a vector that
contains only free degrees of freedom (i.e.) which have to be calculated,
Similarly, the stiffness matrix can be partitioned as:

Δ 𝑎=
[ ]
Δ𝑎 𝑓
Δ𝑎𝑝
(2.53)

• In a similar manner, the stiffness matrix can be partitioned as:

𝐾=
[
𝐾 𝑓𝑓
𝐾 𝑝𝑓
𝐾 𝑓𝑝
𝐾 𝑝𝑝 ] (2.54)

• Using Eq 2.53 and 2.54, Eq 2.42 can be:

[ 𝐾 𝑓𝑓
𝐾 𝑝𝑓
𝐾 𝑓𝑝
𝐾 𝑝𝑝 ][ ]
Δ𝑎𝑓
Δ 𝑎𝑝
=− ¿ (2.55)

69
2.5 Load versus Displacement Control
• For the first iteration this elimination process yields:
(2.53)

• In the subsequent iterations the formula for computing the unknown


degrees of freedom changes into:
(2.57)

Singularity of tangential stiffness matrix at limit point and divergence of iterative


procedure (source: Non- linear Finite Element Analysis Of Solids And Structures)
70

You might also like