Upload
klanpunk
View
232
Download
0
Embed Size (px)
DESCRIPTION
Advanced Math MSc April 07
Citation preview
ADVANCED MATHEMATICS APPLIED
TO CIVIL ENGINEERING
UNIVERSIDAD INDUSTRIAL DE
SANTANDER
Prof. LUIS E. ZAPATA ORDUZ, PhD
Facultad de Ingenierías Físico-Mecánicas
Escuela de Ingeniería Civil
Bucaramanga, I semestre 2016
II.
MATHEMATICS BEHIND THE APPLIED
MECHANICS
2
MATHEMATICAL MODELS
Introduction
There are quite a few engineering problems for which we cannot obtain
exact solutions or the exact solution may be too expensive.
Inability to obtain an exact solution:
The complex nature of governing DEs
Difficulties from dealing with the boundary and initial conditions
¡¡¡ We have to use numerical approximations !!!
3
MATHEMATICAL MODELS
In engineering problems
There are three clases of numerical methods:
(i) Finite difference method (FDM)
(ii) Finite volumen method (FVM)
(iii) Finite element method (FEM)
4
MATHEMATICAL MODELS
Finite Difference Method:
Arithmetic operations to determine derivatives
Derivatives are replaced by difference equations
Useful in simple problems (geometrics)
Very difficult to apply to complex geometries or complex BVPs
5
MATHEMATICAL MODELS
Finite Volume Method:
Based on application of conservation principles over each control volume
Especially use as a numerical method for solving PDE
This method do not requiere a structured mesh (advant. over FEM)
6
MATHEMATICAL MODELS
Finite Element Method:
Uses integral formulations rather tan difference equations
A continuous function is assumed to represent the aprox. solution
This is an aprox. solution to the exact DE
Very useful in civil engineering applications: complex problems
7
MATHEMATICAL MODELS
We will focus on the mathematics behind the Finite Element Analysis
(FEA)
FEA/FEM characteristics:
Powerful numerical technique to solve DEs with BVPs
A continuous domain is discretized to simple geometrical elements: FEs
The elements are related to each other through nodal points
The governing eqs, loading and BCs system of eqs aprox. solution
8
MATHEMATICAL MODELS
The field of mechanics can be divided into three mayor areas:
(i) Theoretical Mechanics fundamental laws and principles of mechanics
(ii) Applied Mechanics scientific/engineering applications of Theore. Mec.
(iii) Computational Mechanics simulation through numerical methods
Computational mechanics can be applied to nanomechanics,
micromechanics, continuum mechanics, and other mechanical systems.
9
MATHEMATICAL MODELS
Nanomechanics branch of nanoscience studying fundamental
mechanical properties of physical systems at the nanometer scale
Micromechanics analysis of composite or heterogeneous materials on
the level of their individual materials constituents
Continuum mechanics studies bodies at the macroscopic level, using
continuum models in which the microstructure is homogenized by
phenomenological averages
In the next we will focus our attention to those mathematical problems related
to FEM/FEA and continuum mechanics civil engineering applications
(restricted to static and linear-elastic problems)
10
11
12
MATHEMATICAL MODELS
When we are using the FEM in Civil Eng. applications we are solving
the whole elasticity field (15 eqs to determine 15 unknowns)
If stress = 0 and strain = 0 FEM = 0
Unknown I Displacement field (3 unk)
14
MATHEMATICAL MODELS
Unknown II State of Strain (6 unk)
Unknown III State of Stress (6 unk)
15
MATHEMATICAL MODELS
Equations I Equilibrium Equations 2D (3 eqs)
These eqs must be satisfied at all point inside the body
The eqs were derived from Newton´s law of motion
bx, by, and bz are body forces per unit volume b acting body´s centroid
Body forces: gravity forces, inertial forces, electric – magnetic forces
16
MATHEMATICAL MODELS
Equations II Stress-Strain Relationships (6 eqs)
Called constitutive laws or constitutive relationships
These relate the stress and strain components
For linear-elastic isotropic materials, we have:
17
MATHEMATICAL MODELS
Equations III Linear Strain-Displacement Relationships (6 eqs)
Are derived from purely geometric considerations
Define the strain components used for the characterization of deformation
The strains associated with the displacement field (Green-Lagrange strains)
These strains are expressed in terms of the displacement gradients
18
MATHEMATICAL MODELS
The strains are computed from the Green-Lagrange (G-L) tensor
Stresses corresponding to G-L strains second Piola-Kirchhoff stresses
PK2 stresses linearly related to the Cauchy stresses as follows:
19
prestresses Cauchy stresses
PK2 stresses
|Jacobian| F = Deformation gradient matrix
MATHEMATICAL MODELS
Assuming isochoric deformation (volumen-preserving) J = 1.0
Assuming the prestressed state (reference configuration) So = 0
We restrict our analysis small deformations small strains
S ≈ σ
In the mathematical models for the linear elastic field:
We will always have 15 unk although we may never see the 15 eqs
The 15 eqs are imbedded into the integral formulation based for the FEM
20
MATHEMATICAL MODELS
Typical Elasticity Field for a Continuous Deformable Body
Statically admisible stress field S(x,y,z) if it satisfies the equil. eqs at all
points in the domain and the surface equil. eqs at all points on the surface:
21
MATHEMATICAL MODELS
Kinematically admisible displacement field d(x,y,z) if it is continuous
and diferenciable at all points in the domain and the geometric BCs at all
points on the surface:
Strain field compatible E(x,y,z) if it is derived from a kinematically
admissible displacement field through the strain-displacement relationships
22
MATHEMATICAL MODELS
Newton´s Laws
First law A particle remains at rest, or continues to move in a straight
line with uniform velocity, if there is no unbalanced force acting on it
Second law The acceleration of a particle is proportional to the resultant
force acting on the particle and in the direction of this force (the constant
of proporcionality is the mass of the particle)
Third law The force of action and reaction between interacting bodies
are equal in magnitude, opposite in direction, and collinear
23
MATHEMATICAL MODELS
FEM/FEA calculations: Positive Sign Convention
24
MATHEMATICAL MODELS
Aproximations to DEs
There are four ways to solve any DE problem (related to FEA):
(i) Exact solution
(ii) Minimization of the quadratic functional
(iii) Strong form (alternative integral form of the DE to obtain the solution)
(iv) Weak form (alternative bilinear representation of the integral formulation
of the DE to obtain the solution)
25
MATHEMATICAL MODELS
Recall:
¿What is a DE?
An equation involving one or more variables and its derivatives with
respective to one or more independent variables.
Homogeneous DE:
Nonhomogeneous DE:
26
MATHEMATICAL MODELS
Example:
Solve the following DE using the exact form:
u´´- u = 1 0 < x < 1 u = u(x)
s.t.: u(0) = 1, u´(1) = 0
Ans.
The exact solution homogeneous solution and a particular solution:
u(0) = uh(0) + unh(0) = 1; uh(0) = 0 and unh(0) = 1
The total exact solution is: u(x) = -1+ c1ex + c2e
-x
Using the BCs: c1 = 2/(1 + e2) and c2 = c1e2
Therefore, we have:
u(x) = -1 + 2ex/(1 + e2) + 2e2-x/(1 + e2)
27
MATHEMATICAL MODELS
Approximate Solution to DE
In many cases we may not be able to solve the DEs exactly (close form
solution)
In fact, although the close form exist, in practice it may be too costly
An alternative approximate the DE using either its integral form or its
finite difference:
NOTE: The solution to the integral formulation is an approximation which is
a linear representation of the unknown parameters cj:
28
MATHEMATICAL MODELS
In this approximation:
cj are the unknown parameters
ϕj are known as trial functions (functions of position x in the domain Ω)
The problem is reduced to find cj such that the approximate solution satisfies
the equations governing the actual solution, u(x).
29
MATHEMATICAL MODELS
Integral Approximation: Strong Formulation
In the process of satisfying the governing eqs through approximations, we
obtain (not accidentally but by planning) n-algebraic relations among the n
parameters c1, c2, …, cn.
One way is to use the integration formulation: strong formulation an
approximate solution over the entire domain Ω = (0,1):
30
Nonhomogeneous solution Homogeneous solution
Known trial functions
MATHEMATICAL MODELS
The function ϕo(x) is chosen it satisfies the nonhomogeneous essential
(actual) BC
For those problems, for which essential BCs are zero ϕo(x) = 0
The trial functions ϕj are chosen satisfy the homogeneous essential
(specified) boundary conditions
The trial functions ϕj should be linearly independent and the given series
must be able to approximate the solution to any given accuracy
31
MATHEMATICAL MODELS
We know that
Therefore, the choice of the trial functions ϕj(x) must be zero for all j.
Let us choose the approximate solution of the form (i.e. using j = 2):
Let us choose the trial function of the form (for now it is magic):
32
MATHEMATICAL MODELS
¿The above trial functions satisfy the homogeneous specified BCs?:
In fact, happens to satisfy the forced BCs (in most cases it is not neccesary):
Now, the choice of ϕo(x) is such that: we know
In summary, the following BCs must be satisfied for the strong formulation:
33
MATHEMATICAL MODELS
Hence, for our example the trial functions are:
Now, we choose c1 and c2 such that they satisfy the original DE:
Since the above equation must be zero for any value of x the
coefficients of the Powers of x must be zero:
Therefore, we can conclude that there is no set of solutions to the above
equation by using the strong formulation.
34
MATHEMATICAL MODELS
DE´s Residual (R)
R is obtained by setting the DE to zero:
Now, the R is multiply by a weighted function v(x) and integrated over Ω:
We may obtain as many linearly independent eqs as there are linearly
independent functions v(x).
In our example, v(x) must be restricted to n = 2 because we have two
coefficients, cj (c1 and c2)
35
MATHEMATICAL MODELS
To overcome this problem approximate the solution to satisfy the DE in
the weighted-integral sense (called weak formulation).
Let us take (¿arbitrarily?)
Gauss-Jordan:
c1 = 324/347
c2 = - 40/347
Hence, the solution to the weak formulation becomes:
36
MATHEMATICAL MODELS
Strong Form: Methods of Weigthted Residual (MWR)
In MWR asume that a solution may be approximated analytically or
piecewise analytically.
In general, a solution to a DE is expressed as a superposition of a base set
of functions that satisfy initial and BC of the problem:
where,
ϕj(x) called trial functions chosen to satisfy the specified BC
ϕo(x) function that is chosen to satisfy the essential nonhomogeneous BC
cj generalized coordinates (are constants determined by a chosen method)
38
MATHEMATICAL MODELS
Key points about the MWR
The assumed solution is not exact the goal is to choose the coefficients
cj such that the residual R becomes small (in fact zero) over Ω
The procedure consists in premultiplying the DE by a weighted function Φi
and integrating over Ω
From theoretical studies the exact solution always satisfies R = 0 if the
weight functions are analytic (even in subdomains)
The number of weight functions Φi should be equal to the number of unk
constants cj the result is a set of n-algebraic eqs for the unk cj
39
MATHEMATICAL MODELS
Four categories
There are four main categories of weight functions applied in MWR in order to
determine the unknown missing coefficients:
1. Collocation Method
In this method R is forced to zero n-times (n = unk coefficients cj)
The selection of collocation points is arbitrary
The weighting functions family of Dirac functions in the Ω
Dirac function in this method: Φi = δ(x – xi)
40
MATHEMATICAL MODELS
2. Subdodmain Method
The integral of the error function is force to zero over selected subintervals
The Ω is divided into n-subdomains Ωn (size free)
The weighted function is chosen such that in each subdomain:
41
MATHEMATICAL MODELS
3. Least-Squares Method
Error is minimized respect to the unk. coefficients in the assumed solution
Hence, the weight function is:
42
MATHEMATICAL MODELS
4. Galerkin Method
We choose the weight functions identical to the trial functions
The method requires the error to be orthogonal to the weight functions
NOTE: In general, for methods 1 – 4 we are solving a matrix equation in order
to obtain the unknown coefficients cj
43
MATHEMATICAL MODELS
Exercise:
Solve the following DE:
u´´- u = 1 0 < x < 1 u = u(x) s.t.: u(0) = 1, u´(1) = 0
1. Determine the weighted functions (use j = 2)
u(x) = -1+ c1ex + c2e
-x
unh(0) = 1 and uh(0) = 0
u´nh(1) = 0 and u´h(1) = 0
Hence, we have:
ϕo(0) = 1 and ϕ´o(1) = 0 nonhomogeneous (actual) BCs
ϕj(0) = 0 and ϕ´j(1) = 0 homogeneous form of the specified BCs
44
MATHEMATICAL MODELS
Solving for the nonhomogeneous term:
BCs(hom) since BCs =2 let us the polynomial: ϕo(x) = a + bx
Solving for the BCs: ϕo(0) = 1 a = 1 and ϕ´o(1) = 0 b = 0
The polynomial is ϕo(x) = 1
Solving for the first homogeneous term (j = 1):
BCs(nonh) = 2 let us the polynomial: ϕ1(x) = a + bx + cx2
Solving for the BCs: ϕ1(0) = 0 a = 0 and ϕ´1(1) = 0 b + 2c = 0
The polynomial is ϕ1(x) = c(x2 – 2x) (but c = 1 absorbed into c1)
The trial function is ϕ1(x) = (x2 – 2x)
45
MATHEMATICAL MODELS
Solving for the second homogeneous term (j = 2):
BCs(nonh) = 2 let us the polynomial: ϕ2(x) = a + bx + dx2
Solving for the BCs: ϕ2(0) = 0 a = 0 and ϕ´2(1) = 0 b + 3d = 0
The polynomial is ϕ2(x) = d(x3 – 3x) (but d = 1 absorbed into c2)
The trial function is ϕ2(x) = (x3 – 3x)
Hence, the trial functions are:
ϕo(x) = 1 ϕ1(x) = (x2 – 2x) ϕ2(x) = (x3 – 3x)
The complete approximate solution:
46
MATHEMATICAL MODELS
2. Determine the DE´s residual (for j = 2)
3. Approximate the DE by using collocation method (use j = 2)
Rta.
c1 = 666/731
c2 = -81/731
47
MATHEMATICAL MODELS
5. Approximate the DE by using subdomain method (use j = 2)
Rta.
c1 = 195/209 and c2 = -24/209
3. Approximate the DE by using least-squares method (use j = 2)
Rta.
c1 = 5730/6143 and c2 = -72800/632729
6. Approximate the DE by using Galerkin method (use j = 2)
Rta.
c1 = 5010/5537 and c2 = -80/791
48
MATHEMATICAL MODELS
Weak Formulation
The formulation previously described is known as strong formulation
We needed to evaluate the following integral (highest order derivative)
This implies that v´´ Є (-∞, 0) or (0, ∞) meaningful approx. solution
49
MATHEMATICAL MODELS
The Strategy:
Reducing the order of differentiability through mathematical techniques
By integrating by parts: strong form weak form
The key is “weakened” the order of continuity of primary variable
The major advantage is when choosing the trial functions
For an ODE: strong form (2m) weak form (up to m-order)
50
MATHEMATICAL MODELS
Test Function (v)
Any cont & diff function vanishes at points of the BC (essential)
Weak formulation test functions obtained directly from trial functions
Test Function: (approximate solution) – (exact solution) very small
51
MATHEMATICAL MODELS
Weak Form
The goal is to have the same order of derivatives for the primary variables
and the test function v(x).
52
MATHEMATICAL MODELS
Therefore, the following eq is called the weak form of the DE
The test function must be such that:
Or symbolically:
53
MATHEMATICAL MODELS
Quadratic Functionals
If B(u,v) = B(v,u) (symmetric bilinear term) the functional for the given
DE can be written as the following quadratic functional:
vs.
The Euler-Lagrange eq. for this quadratic functional is same as the DE. In
our case the quadratic functional is:
54
MATHEMATICAL MODELS
Galerkin Approximation
The Galerkin approach (weak form) is one of the most often used
approximate methods to determine numerical solutions of ODEs and PDEs
ϕo(x) = 0 satisfies the nonhomogeneous essential BCs
ϕj(x) = 0 trial functions: satisfy the homogeneous essential BCs
cj constants called generalized coordinates
NOTE: Galerkin method the trial functions satisfy both BCs essential/natural
55
MATHEMATICAL MODELS
When using Galerkin approx.:
Regardless of using the weak or strong formulation accuracy of the
aprox. solution depends upon the choice of trial functions (TFs)
To assume a proper TF for the unknown exact solution is not an easy task
Specially true when the exact solution to have a large variation over Ω
To overcome TFs are expressed using piecewise continuous functions
56
MATHEMATICAL MODELS
Assuming the following approximate function:
Hence, TFs must satisfy:
(i) Be linearly independent
(ii) Satisfy both essential and natural BCs
(iii) Form a complete set (no missing terms)
(iv) Test Functions = Trial Functions v(x) = Φi(x) for 1 ≤ i ≤ n
NOTE: in Galerkin you can use both the strong or weak form to get answers.
NOTE: in the following, the approximation solution losses the upper bar.
57
MATHEMATICAL MODELS
Galerkin: Strong Form
The strong form is:
Recall: Test Functions = Trial Functions v(x) = Φi(x) for 1 ≤ i ≤ n
In symbolic form, the i-th eq (1 ≤ i ≤ n) is written as:
58
MATHEMATICAL MODELS
In the above equation:
Kij the stiffness coefficient
pi The generalized forces corresponding to the i-th generalized coordinate
In matrix notation, the set of equations can be written as:
K c = P
K known as the stiffness matrix
c vector of generalized coordinates
P vector of generalized forces
59
MATHEMATICAL MODELS
Exercise:
Solve the following DE by using the strong form of the Galerkin method:
u´´- u = 1 0 < x < 1 u = u(x) s.t.: u(0) = 1, u´(1) = 0
Rta.
Same trial functions: ϕ0(x) = 1, ϕ1(x) = (x2 – 2x) and ϕ2(x) = (x3 – 3x)
The approximate sol.: usg(x) = ϕ0(x) + c1ϕ1(x) + c2ϕ2(x)
usg(x) = 1 + c1 (x2 – 2x) + c2 (x3 – 3x)
60
MATHEMATICAL MODELS
The strong form (for i = 1, 2):
In matrix form: K c = P
61
MATHEMATICAL MODELS
The generalized stiffness coefficients (for i = 1, 2):
62
MATHEMATICAL MODELS
The generalized force coefficients (for i = 1, 2):
Solving the system:
63