Professional Documents
Culture Documents
Rak-54.3200 / 2015 / JN 47
Rak-54.3200
Numerical Methods in Structural Engineering
Contents
1. Modelling principles and boundary value problems in engineering sciences
2. Energy methods and basic 1D finite element methods
- bars/rods, beams, heat diffusion, seepage, electrostatics
3. Basic 2D and 3D finite element methods
- heat diffusion, seepage
4. Numerical implementation techniques of finite element methods
5. Abstract formulation and accuracy of finite element methods
6. Finite element methods for Euler−Bernoulli beams
7. Finite element methods for Timoshenko beams Research activities
are going on at our
8. Finite element methods for Kirchhoff−Love plates
department in many
9. Finite element methods for Reissner−Mindlin plates topics of the course!
10. Finite element methods for 2D and 3D elasticity
11. Extra lecture: other finite element applications in structural engineering
(time-dependent problems, nonlinearities, isogeometric methods)
Rak-54.3200 / 2016 / JN 48
2 Energy methods and
basic 1D finite element methods
Contents
1. Weak form (based on the principle of virtual work)
2. 1D finite element method (based on the weak form)
Learning outcome
A. Understanding of the main principles behind the finite element method
B. Ability to formulate and apply the finite element method for 1D model problems
References
Lecture notes: chapters 1, 2, 3.1−3, 5.1−4
Text book: chapters 1.1−11
Rak-54.3200 / 2014 / JN 49
2.0 Finite element methods intuitively
Rak-54.3200 / 2014 / JN 50
2.1 Weak form − 1D model problem
Axially loaded elastic bar: The differential equation and the boundary conditions,
i.e., the strong form of the problem is the following:
(1 - DE) EAu ' ' ( x) b( x), 0 x L
E ( x), A( x), b( x) N
(2 - eBC) u (0) u0 L
x, u ( x )
(3 - nBC) ( EAu ' )( L) N L
0 L
Rak-54.3200 / 2016 / JN 51
2.1 Weak form − 1D model problem
Axially loaded elastic bar: The differential equation and the boundary conditions,
i.e., the strong form of the problem is the following:
(1 - DE) EAu ' ' ( x) b( x), 0 x L
E ( x), A( x), b( x) N
(2 - eBC) u (0) u0 L
x, u ( x )
(3 - nBC) ( EAu ' )( L) N L
0 L
Rak-54.3200 / 2016 / JN 52
2.1 Weak form − 1D model problem
Axially loaded elastic bar: The differential equation and the boundary conditions,
i.e., the strong form of the problem is the following:
(1 - DE) EAu ' ' ( x) b( x), 0 x L
E ( x), A( x), b( x) N
(2 - eBC) u (0) u0 L
x, u ( x )
(3 - nBC) ( EAu ' )( L) N L
0 L
u axial displacement (unknown function)
E Young' s modulus (given material data)
A cross-sectional area (given geometrical data)
b axial body load (given loading data)
L length (given geometrical data)
u0 axial end point displacement (given essential/geometric boundary data)
N L axial end point force (given natural/force boundary data).
Rak-54.3200 / 2014 / JN 53
MOTIVATION …
1.1 Modeling and computation
in engineering design and analysis
Changes Physical engineering solution uP = ?
to the problem problem with
and design design criteria
step 5 Acceptance
Rak-54.3200 / 2014 / JN 55
Computer exercise 2 − Matlab (2015)
(i) Implement the finite element method with linear basis functions for the
axially loaded rod (derived above) in MATLAB with the following initial data:
E 50 GPa , L 7 m, E ( x), A( x), b( x) N
L
A0 0,25 m 2 , A1 0,5 m 2 x, u ( x )
b ? kN/m, N L ? kN 0 L
(ii) Solve the finite element problem with E , A, b NL
x, u
− two elements u u0
− 10 elements
and graphically compare the finite element approximations
of the axial displacement and stress to the corresponding
exact solutions of Home exercise 1.1 and Computer exercise 1.
Hint: Modify the given model m-file.
Rak-54.3200 / 2016 / JN 56
Computer exercise 2 − Matlab (2015)
(i) Implement the finite element method with linear basis functions for the
axially loaded rod (derived above) in MATLAB with the following initial data:
Rak-54.3200 / 2016 / JN 57
Computer exercise 2 − Matlab (2015)
(i) Implement the finite element method with linear basis functions for the
axially loaded rod (derived above) in MATLAB with the following initial data:
Rak-54.3200 / 2016 / JN 58
… BACK TO WORK
2.1 Weak form − 1D model problem
1. Multiply the differential equation (1) by a (smooth) test function (specified later):
EAu ' ' ( x) b( x) EAu ' ' ( x) v( x) b( x) v( x), 0 x L
Rak-54.3200 / 2016 / JN 60
2.1 Weak form − 1D model problem
1. Multiply the differential equation (1) by a (smooth) test function (specified later):
EAu ' ' ( x) b( x) EAu ' ' ( x) v( x) b( x) v( x), 0 x L
2. Integrate over the domain (interval):
L L
( EAu ' )' ( x) v( x) dx b( x) v( x) dx
0 0
Rak-54.3200 / 2016 / JN 61
2.1 Weak form − 1D model problem
1. Multiply the differential equation (1) by a (smooth) test function (specified later):
EAu ' ' ( x) b( x) EAu ' ' ( x) v( x) b( x) v( x), 0 x L
2. Integrate over the domain (interval):
L L
( EAu ' )' ( x) v( x) dx b( x) v( x) dx
0 0
3. Integrate by parts (the left hand side) for moving one derivative from u to v:
L L
( EAu ' )( L) v( L) ( EAu ' )(0) v(0) EAu' v' dx b v dx
0 0
Rak-54.3200 / 2014 / JN 62
2.1 Weak form − 1D model problem
1. Multiply the differential equation (1) by a (smooth) test function (specified later):
EAu ' ' ( x) b( x) EAu ' ' ( x) v( x) b( x) v( x), 0 x L
2. Integrate over the domain (interval):
L L
( EAu ' )' ( x) v( x) dx b( x) v( x) dx
0 0
3. Integrate by parts (the left hand side) for moving one derivative from u to v:
L L
( EAu ' )( L) v( L) ( EAu ' )(0) v(0) EAu' v' dx b v dx
0 0
Rak-54.3200 / 2014 / JN 63
2.1 Weak form − 1D model problem
1. Multiply the differential equation (1) by a (smooth) test function (specified later):
EAu ' ' ( x) b( x) EAu ' ' ( x) v( x) b( x) v( x), 0 x L
2. Integrate over the domain (interval):
L L
( EAu ' )' ( x) v( x) dx b( x) v( x) dx
0 0
3. Integrate by parts (the left hand side) for moving one derivative from u to v:
L L
( EAu ' )( L) v( L) ( EAu ' )(0) v(0) EAu' v' dx b v dx
0 0
Rak-54.3200 / 2014 / JN 64
2.1 Weak form − 1D model problem
L
u such that it satisfies u(0) u0 , dx and
2
Weak form. Find (u ' )
L L 0
EAu' v' dx N
0
L v( L) b v dx,
0L
v satisfying v(0) 0, dx .
2
for all ( v ' )
0
Rak-54.3200 / 2014 / JN 65
2.1 Weak form − 1D model problem
L
u such that it satisfies u(0) u0 , dx and
2
Weak form. Find (u ' )
L L 0
v satisfying v(0) 0, dx .
2
for all ( v ' )
0
Remark. Note that the solution and the test function, respectively, have to
satisfy the boundary conditions u(0) u0 , v(0) 0 and the regularity
conditions
L L
dx , dx .
2 2
(u ' ) ( v ' )
0 0
Then they are called kinematically admissible.
Rak-54.3200 / 2014 / JN 66
2.1 Weak form − 1D model problem
L
u such that it satisfies u(0) u0 , dx and
2
Weak form. Find (u ' )
L L 0
v satisfying v(0) 0, dx .
2
for all ( v ' )
0
Remark. Note that the solution and the test function, respectively, have to
satisfy the boundary conditions u(0) u0 , v(0) 0 and the regularity
conditions
L L
dx , dx .
2 2
(u ' ) ( v ' )
0 0
Then they are called kinematically admissible.
Remark. (Home exercise 2.1) Starting from the weak form we could
correspondingly derive the strong form (integrating by parts ”backwards”).
Rak-54.3200 / 2016 / JN 67
2.1 Weak form − 1D model problem
Rak-54.3200 / 2016 / JN 68
2.1 Weak form − 1D model problem
Rak-54.3200 / 2016 / JN 69
2.1 Weak form − 1D and 2D generalizations
EAu' v' dx b v dx
0 0
kT ' v' dx f v dx
0 0
EA, b k, f
u u0 T T0
Rak-54.3200 / 2016 / JN 70
2.1 Weak form − 1D and 2D generalizations
Rak-54.3200 / 2016 / JN 71
2.1 Weak form − 1D and 2D generalizations
1D beam: L L
Rak-54.3200 / 2016 / JN 72
2.1 Weak form − 1D and 2D generalizations
1D beam: L L
This one is of
Rak-54.3200 / 2016 / JN 73
a different form!
2.1 Weak form
Break exercise 2
L2 () : { v is defined on R |
v 2 dx }
H 1 () : { v L2 () |
(v' ) 2 dx }
but not, however, to the function space
H 01 () : { v H 1 () | v 0 on }.
Rak-54.3200 / 2014 / JN 74
2.2 1D finite element method −
model problem
1. Divide the solution interval (domain) into n subintervals ei (elements)
with nodes xi and the element size hi xi xi 1 :
e1 ei en
x0 0 xi 1 xi xn L
Rak-54.3200 / 2014 / JN 75
2.2 1D finite element method −
model problem
1. Divide the solution interval (domain) into n subintervals ei (elements)
with nodes xi and the element size hi xi xi 1 :
e1 ei en
x0 0 xi 1 xi xn L
2. Choose a trial function for the finite element approximation as a sum
uh ( x) 0 ( x)d 0 1 ( x)d1 n ( x)d n j ( x)d j
n
j 0
Rak-54.3200 / 2014 / JN 76
2.2 1D finite element method −
model problem
1. Divide the solution interval (domain) into n subintervals ei (elements)
with nodes xi and the element size hi xi xi 1 :
e1 ei en
x0 0 xi 1 xi xn L
2. Choose a trial function for the finite element approximation as a sum
uh ( x) 0 ( x)d 0 1 ( x)d1 n ( x)d n j ( x)d j
n
j 0
with suitable local basis functions i of some polynomial order (now linear)
0 i i 1 n
1
x0 0 xi xi 1 xn L
Rak-54.3200 / 2014 / JN 77
2.2 1D finite element method −
model problem
1. Divide the solution interval (domain) into n subintervals ei (elements)
with nodes xi and the element size hi xi xi 1 :
e1 ei en
x0 0 xi 1 xi xn L
2. Choose a trial function for the finite element approximation as a sum
uh ( x) 0 ( x)d 0 1 ( x)d1 n ( x)d n j ( x)d j
n
j 0
with suitable local basis functions i of some polynomial order (now linear)
0 i i 1 n
1
i ( x j ) ij
x0 0 xi xi 1 xn L
The unknown scalar values d i uh ( xi ) are called the degrees of freedom.
Rak-54.3200 / 2014 / JN 78
2.2 1D finite element method −
model problem
Ensure that the trial function satisfies the essential boundary conditions:
Rak-54.3200 / 2014 / JN 79
2.2 1D finite element method −
model problem
Ensure that the trial function satisfies the essential boundary conditions:
0 v(0) c0 0
Rak-54.3200 / 2014 / JN 80
2.2 1D finite element method −
model problem
Ensure that the trial function satisfies the essential boundary conditions:
0 v(0) c0 0
4. Insert the functions − trial and test − into the weak form:
L L
Rak-54.3200 / 2014 / JN 81
2.2 1D finite element method −
model problem
Ensure that the trial function satisfies the essential boundary conditions:
0 v(0) c0 0
4. Insert the functions − trial and test − into the weak form:
L L
Rak-54.3200 / 2014 / JN 82
2.2 1D finite element method −
model problem
This results in a simple equation system
Kd f
with the stiffness matrix (computable for i,j = 1, …, n), force vector (computable
for i = 1, …, n) and the displacement vector (unknown for i = 1, …, n):
L
K K ij , K ij EAi ' j ' dx,
0
di d0
L L
f f i , f i b i dx N L i ( L) u0 AE dx, d d j .
0 0
dx dx
Rak-54.3200 / 2014 / JN 83
2.2 1D finite element method −
model problem
This results in a simple equation system
Kd f
with the stiffness matrix (computable for i,j = 1, …, n), force vector (computable
for i = 1, …, n) and the displacement vector (unknown for i = 1, …, n):
L
K K ij , K ij EAi ' j ' dx,
0
di d0
L L
f f i , f i b i dx N L i ( L) u0 AE dx, d d j .
0 0
dx dx
details on blackboard or exercises 0 1 2
1
the general case
an example with two elements
x0 0
e1 e2
x2 L
Rak-54.3200 / 2014 / JN 84
2.2 1D finite element method −
model problem
This results in a simple equation system
Kd f
with the stiffness matrix (computable for i,j = 1, …, n), force vector (computable
for i = 1, …, n) and the displacement vector (unknown for i = 1, …, n):
L
K K ij , K ij EAi ' j ' dx,
0
di d0
L L
f f i , f i b i dx N L i ( L) u0 AE dx, d d j .
0 0
dx dx
details on blackboard or exercises 0 1 2 3 4
1
the general case
an example with two elements
x0 0
e1 e2
x2 L
implications of quadratic basis functions
Rak-54.3200 / 2014 / JN 85
2.2 1D finite element method −
model problem
This results in a simple equation system
Kd f
with the stiffness matrix (computable for i,j = 1, …, n), force vector (computable
for i = 1, …, n) and the displacement vector (unknown for i = 1, …, n):
L
K K ij , K ij EAi ' j ' dx,
0
di d0
L L
f f i , f i b i dx N L i ( L) u0 AE dx, d d j .
0 0
dx dx
Remark. The stiffness matrix is (very often) symmetric (due to derivative orders) and
its entries are concentrated in a narrow diagonal band forming a band matrix (due to
local trial and test functions). These features can can be utilized in computer
impelementation − implying small amounts of memory needs and quick processing.
Rak-54.3200 / 2014 / JN 86
2.2 1D finite element method −
model problem
This results in a simple equation system
Kd f
with the stiffness matrix (computable for i,j = 1, …, n), force vector (computable
for i = 1, …, n) and the displacement vector (unknown for i = 1, …, n):
L
K K ij , K ij EAi ' j ' dx,
0
di d0
L L
f f i , f i b i dx N L ni u0 AE dx, d d j .
0 0
dx dx
Remark. The stiffness matrix is (very often) symmetric (due to derivative orders) and
its entries are concentrated in a narrow diagonal band forming a band matrix (due to
local trial and test functions). These features can can be utilized in computer
impelementation − implying small amounts of memory needs and quick processing.
Remark. Test and trial functions have to be (only) once locally differentiable (and will
be then integrated over the domain) and (only) locally evaluable on the boundary.
Rak-54.3200 / 2014 / JN 87
2.2 1D finite element method −
model problem
5. Use an appropriate solver for the equation system:
1
j ( x) d j
n
dK f uh ( x) j 1
Rak-54.3200 / 2014 / JN 88
2.2 1D finite element method −
model problem
5. Use an appropriate solver for the equation system:
1
j ( x) d j
n
dK f uh ( x) j 1
A( x)
Rak-54.3200 / 2014 / JN 89
2.2 1D finite element method −
model problem
5. Use an appropriate solver for the equation system:
1
j ( x) d j
n
dK f uh ( x) j 1
A( x)
7. Evalute possible error indicators, change the discretization (steps 1−4) ... rerun ...
Rak-54.3200 / 2014 / JN 90
2.2 1D finite element method −
model problem
5. Use an appropriate solver for the equation system:
1
j ( x) d j
n
dK f uh ( x) j 1
A( x)
7. Evalute possible error indicators, change the discretization (steps 1−4) ... rerun ...
Rak-54.3200 / 2014 / JN 91
2.2 1D finite element method −
further steps
Rak-54.3200 / 2014 / JN 92
2.X Related methods
L L
( EAu ' )' ( x) v( x) dx b( x) v( x) dx
0 0
L
{( EAu ' )' ( x) b( x)} v( x) dx 0
0
Rak-54.3200 / 2014 / JN 93
2.X Related methods
0 v(0) c0 0
4. Insert the functions − trial and weight − into the residual form:
L
i 0 [ j 0 d j ( EA j ' )'b] i dx [ j 0 d j ( EA j )' ( L) N L ] i ( L) ci 0
n n n
0
Rak-54.3200 / 2014 / JN 94
2.X Related methods
L
K K ij , K ij i ( EA j ' )' dx i ( L)( EA j )' ( L), d d j
0
L
f f i , f i i [b u0 ( EA 0 ' )' ]dx i ( L)[ N L d 0 ( EA 0 )' ( L)] .
0
Remark. Weight functions can be chosen fairly freely since they are only
integrated over the domain (interval) and evaluated on the boundary (points),
while the trial functions have to twice differentiable (and will be then integrated
over the domain) and once differentiable (and will be then evaluated on the
boundary).
Rak-54.3200 / 2014 / JN 95
2.X Related methods
Collocation methods. Choosing (in step 3 above) the weight functions as Dirac
delta distributions corresponding to a set of grid points xi implies the following:
3. Choose the Dirac delta distributions for the weight functions as
i ( x) ( x xi ), having their basic feature for any function f f (x) as
L
( x x ) f ( x)dx f ( x ),
0
i i
n
i 0
[ n
j 0
n
d j ( EA j ' )'b]( xi ) [ j 0 d j ( EA j )' ( L) N L ] ci 0
K K ij , K ij ( EA j ' )' ( xi ) ( EA j )' ( L), d d j
f f i , f i b( xi ) u0 ( EA 0 ' )' ( xi ) [ N L u0 ( EA 0 ' ) ( L)].
Rak-54.3200 / 2014 / JN 96
2.X Related methods
Coffee exercise 2
Starting from the total (elastic) potential energy expression for an axially loaded rod,
L L
1
(v) EA(v' ) 2 dx N L v( L) b v dx,
20 0
derive the corresponding expression of the principle of virtual work by applying the
principle of minimum potential energy. Hint: Assuming that u minimizes the energy
functional above find the minimum at 0 for function f defined as
f ( ) : (u u),
where u is chosen properly.
Rak-54.3200 / 2014 / JN 97
QUESTIONS?
ANSWERS”
LECTURE BREAK!