Вы находитесь на странице: 1из 44

Path following methods

Prof. Dr. Eleni Chatzi, Dr. Konstantinos Agathos


Lecture 4 - 10 October, 2019

Institute of Structural Engineering, ETH Zürich

October 11, 2019

Institute of Structural Engineering Method of Finite Elements II 1


Outline

1 Introduction

2 The Newton-Raphson method

3 The Newton-Raphson method in structural mechanics

4 Path-following Methods

Institute of Structural Engineering Method of Finite Elements II 2


Learnig goals

Understanding the general solution process for nonlinear


problems

Understanding limitations of the process

Gaining a basic understanding of path following solution


methods

Institute of Structural Engineering Method of Finite Elements II 3


Significance of the lecture

Applications include any problem described by nonlinear PDEs, such


as:

Structural mechanics problems involving:

Geometrical nonlinearities

Material nonlinearities

Material failure

Fluid mechanics

Institute of Structural Engineering Method of Finite Elements II 4


The Newton-Raphson method in 1-D

Problem
Given the function:

f (x ) : R → R

find:

x̂ : f (x̂ ) = 0

where:
f (x ) is a general nonlinear function

The derivative of f (x ) is available

Institute of Structural Engineering Method of Finite Elements II 5


The Newton-Raphson method in 1-D

Solution:

If an initial estimate x0 is available, the function can be locally


approximated by a truncated Taylor series:

df
f (x ) ≈ f (x0 ) + |x (x − x0 ) = f (x0 ) + f 0 (x0 ) ∆x0
dx 0

Then this equation can be solved to obtain a better approximation


to the solution:
−1
∆x0 = − f 0 (x0 ) f (x0 ) → x1 = x0 + ∆x0

Institute of Structural Engineering Method of Finite Elements II 6


The Newton-Raphson method in 1-D

The process can be repeated iteratively until a sufficient accuracy is


obtained:
−1
∆xi = − f 0 (xi ) f (xi ) → xi+1 = xi + ∆xi

The absolute value of f (xi+1 ) can be used to determine the


achieved level of accuracy

Alternatively the difference between two consecutive solutions


(|xi+1 − xi |) could be used

Institute of Structural Engineering Method of Finite Elements II 7


The Newton-Raphson method in 1-D

The whole solution can be written as an algorithm:

Data: f (x ),f 0 (x ), x0 , tol, maxit


Result: x̂

for i ≤ maxit do
 −1
df
∆xi = − dx |xi f (xi )
xi+1 = xi + ∆xi
if |f | ≤ tol then
return
end
end
return
Algorithm 1: Newton-Raphson algorithm

Institute of Structural Engineering Method of Finite Elements II 8


The Newton-Raphson method in 1-D

But how do we know that the method will eventually converge to


the correct value?

It can be shown that: (x̂ − xi+1 ) ≤ C (x̂ − xi )2

Provided that:

The initial estimate x0 is close enough to the solution x̄

f is sufficiently smooth

Institute of Structural Engineering Method of Finite Elements II 9


The Newton-Raphson method in n-D
The method can be extended to vector valued functions:

f (x) : Rn → Rn
and systems of nonlinear equations:

f (x) = 0
By considering that:
  ∂f
f (x) ≈ f xi + | i ∆xi
∂x x
−1  
∂f

∆xi = − |i f xi
∂x x

xi+1 = xi + ∆xi
For the n-D version of the method, we will use superscripts to indicate the
iteration i and we will drop the index for the increment to simplify notation.
Institute of Structural Engineering Method of Finite Elements II 10
The Newton-Raphson method in n-D
Or component wise:

 ∂f xi ∂f1 (xi ) ∂f1 (xi ) 


1( )

···
    
f1 (x) f1  xi
 ∂x1 i ∂x2 ∂xn  ∆x1
  ∂f2 (x ) ∂f2 (xi ) ∂f2 (xi ) 
f2 (x) f2 xi   ∂x ··· ∆x2 
   
 . ≈ . + 1 ∂x2 ∂xn 
 .   .   . .. ..   .. 

 .   .   .. ..  . 
. . . 
i
  
fn (x) fn x ∂fn (xi ) ∂fn (xi ) ∂fn (xi ) ∆xn
n×1 n×1 ∂x1 ∂x2 ··· ∂xn n×1
n×n

 ∂f xi ∂f1 (xi ) ∂f1 (xi )
−1
1( )
···
 i+1     
x1 i
x1 f1 xi 
 ∂x1 i ∂x2 ∂xn 
 i+1   i   ∂f2 (x ) ∂f2 (xi ) ∂f2 (xi )  
x2  x2   ∂x ··· f2 xi 
∂x2 ∂xn  
  
 . = . − 1
.. 
 .  .  . .. .. .. 
 .   .   .. . 
 
. . .  
i+1 i fn x i
  
xn xn ∂fn (xi ) ∂fn (xi ) ∂fn (xi )
n×1 n×1 ∂x1 ∂x2 ··· ∂xn n×1
n×n

Institute of Structural Engineering Method of Finite Elements II 11


The Newton-Raphson method

We recall the linearized equilibrium equation:

KT ∆un = −Ri = fext − fint i

Typically external loads are scaled by a factor λ:

KT ∆un = −Ri = λfext − fint i

Institute of Structural Engineering Method of Finite Elements II 12


The Newton-Raphson method

For a given value of λ, the above corresponds to a Newton-Raphson


iteration where:

∂Ri
R = Ri + ∆un = fint i − λfext + KT ∆un = 0
∂un

The iteration is repeated until

|R| <= tol


where tol is the tolerance.

Institute of Structural Engineering Method of Finite Elements II 13


The Newton-Raphson method

Typically:

Different (increasing) values are given to the load factor

Displacements are obtained using the Newton-Raphson method

The response of the structure is obtained for progressively


increasing loading

The load-displacement curve of the structure can be obtained

Institute of Structural Engineering Method of Finite Elements II 14


The Newton-Raphson method

The above procedure is called load control and can be illustrated as


follows:

Increment load

Apply load increment

Solve with
Newton-Raphson

Repeat for next


increment

Institute of Structural Engineering Method of Finite Elements II 15


The Newton-Raphson method

The above procedure is called load control and can be illustrated as


follows:

Increment load

Apply load increment

Solve with
Newton-Raphson

Repeat for next


increment

Institute of Structural Engineering Method of Finite Elements II 15


The Newton-Raphson method

The above procedure is called load control and can be illustrated as


follows:

Increment load

Apply load increment

Solve with
Newton-Raphson

Repeat for next


increment

Institute of Structural Engineering Method of Finite Elements II 15


The Newton-Raphson method

The above procedure is called load control and can be illustrated as


follows:

Increment load

Apply load increment

Solve with
Newton-Raphson

Repeat for next


increment

Institute of Structural Engineering Method of Finite Elements II 15


The Newton-Raphson method

The above procedure is called load control and can be illustrated as


follows:

Increment load

Apply load increment

Solve with
Newton-Raphson

Repeat for next


increment

Institute of Structural Engineering Method of Finite Elements II 15


The Newton-Raphson method

The above procedure is called load control and can be illustrated as


follows:

Increment load

Apply load increment

Solve with
Newton-Raphson

Repeat for next


increment

Institute of Structural Engineering Method of Finite Elements II 15


The Newton-Raphson method

In many cases the solution path is not monotonic

Phenomena such as snap-through and snap-back can be present

The Newton-Raphson method will either fail or not provide the


full path in those cases

Such phenomena are often of interest in practice e.g.when the


overcritical behavior of a structure needs to be known

Institute of Structural Engineering Method of Finite Elements II 16


Example

Solution path involving snap-through:

Institute of Structural Engineering Method of Finite Elements II 17


Example

Solution path involving snap-through:

Institute of Structural Engineering Method of Finite Elements II 17


Example

Solution path involving snap-through:

Institute of Structural Engineering Method of Finite Elements II 17


Example

Solution path involving snap-through:

Institute of Structural Engineering Method of Finite Elements II 17


Path-following methods

In this category of methods:

It is attempted to follow the full solution path including


snap-back and snap-through phenomena

The load factor is included as an additional unknown in the


nonlinear system of equations

An additional equation (constraint) is added in the system

The choice of this constraint leads to different methods

Institute of Structural Engineering Method of Finite Elements II 18


Path-following methods

After adding ∆λ as an unknown and the constraint as an equation


the new system of equations is obtained:
" # " #
R (u, λ) 0
=
g (u, λ) 0
where:
u is the vector of nodal unknowns
λ is the load factor
R is the residual of the Newton-Raphson method
g is the additional constraint

Institute of Structural Engineering Method of Finite Elements II 19


Path-following methods

The above system can be solved by employing the Newton-Raphson


method. The linearization of the system yields:
   
  ∂R ui , λi∂R ui , λi

" #
 R ui , λi + ∆u + ∆λ  0
∂u ∂λ =
 
i i i i 0
  
  ∂g u , λ ∂g u , λ 
g ui , λi + ∆u + ∆λ
∂u ∂λ

where:
∆u is the nodal displacement increment
∆λ is the load factor increment

Institute of Structural Engineering Method of Finite Elements II 20


Path-following methods

The above system can be solved by employing the Newton-Raphson


method. The linearization of the system yields:
" #" # " #
KT −fext ∆u −Ri
=
hT s ∆λ −g i

where:
∆u is the nodal displacement increment
∆λ is the load factor increment
∂g
h Is the gradient of g with respect to u: h =
∂u
∂g
s is the derivative of g with respect to λ: s =
∂λ

Institute of Structural Engineering Method of Finite Elements II 20


Path-following methods

The coefficient matrix in this system is not symmetric.

To exploit the symmetry and sparsity of KT the following


partitioning procedure is employed:

∆uI = KT −1 fext , ∆uII = −KT −1 Ri

g i + hT ∆uII
∆λ = − , ∆u = ∆λ∆uI + ∆uII
s + hT ∆uI

Institute of Structural Engineering Method of Finite Elements II 21


Path-following methods

The choice of the additional constraint g is crucial

Different alternatives exist

In some cases problem specific constraints are needed

Some widely used alternatives are demonstrated in the following

Institute of Structural Engineering Method of Finite Elements II 22


Path-following methods: Load control

Setting g = λ − λ̄ results in load control:

∂g ∂g
g = λ − λ̄ ⇒ h = = 0, s = =1
∂u ∂λ

g i = λi − λ̄

" #" # " #


KT −fext ∆u −Ri
=
0 1 ∆λ λ̄ − λi

 
∆λ = λ̄ − λi , ∆u = KT −1 λ̄fext − fint i

Institute of Structural Engineering Method of Finite Elements II 23


Path-following methods: Displacement control

Setting g = T · u − ū results in displacement control:

∂g ∂g
g = T · u − ū ⇒ h = = TT , s = =0
∂u ∂λ

g i = T · ui − ū

" #" # " #


KT −fext ∆u −Ri
=
T 0 ∆λ ū − T · ui

h i
In the above T = 0 0 . . . 1 . . . 0 0 where the entry 1
corresponds to a selected dof

Institute of Structural Engineering Method of Finite Elements II 24


Path-following methods: Displacement control

Displacement control in a path Displacement control in a path


involving snap-through involving snap-back

Institute of Structural Engineering Method of Finite Elements II 25


Path-following methods: Displacement control

Displacement control in a path


Displacement control in a path
involving snap-back
involving snap-through

Institute of Structural Engineering Method of Finite Elements II 25


Path-following methods: Displacement control

Displacement control in a path Displacement control in a path


involving snap-through involving snap-back

Institute of Structural Engineering Method of Finite Elements II 25


Path-following methods: Arc-length
q
Setting g = (u − u0 )T · (u − u0 ) + (λ − λ0 )2 − ∆s results in arc
length control (Crisfield 1981):
q
gi = (ui − u0 )T · (ui − u0 ) + (λi − λ0 )2 − ∆s

ui − u0 λi − λ0
 
∂g ∂g
h= = , s= =
∂u g ∂λ g

 
KT −fext " # " #
T   ∆u Ri
 u − u0
i i 0
λ − λ  ∆λ = −g i

g g

In the above, superscript 0 refers to the beginning of the step

Institute of Structural Engineering Method of Finite Elements II 26


Path-following methods: Arc-length

We observe that for i = 0 the system becomes:

 
KT −fext " # " #
T   ∆u −Ri
 u − u0
0 0 0
λ − λ  ∆λ = −g i

g g

Institute of Structural Engineering Method of Finite Elements II 27


Path-following methods: Arc-length

We observe that for i = 0 the system becomes:

" #" # " #


KT −fext ∆u −Ri
=
0 0 ∆λ −g i

Institute of Structural Engineering Method of Finite Elements II 27


Path-following methods: Arc-length

We observe that for i = 0 the system becomes:

" #" # " #


KT −fext ∆u −Ri
=
0 0 ∆λ −g i

The coefficient matrix is singular!

To obtain the load factor at the first iteration a predictor step is


introduced.

Institute of Structural Engineering Method of Finite Elements II 27


Path-following methods: Arc-length

For the predictor step the system is solved for the external loads:

∆up = KT −1 fext

Then the increment of the load factor is computed as:

∆s
∆λp = ±
k∆up k

Institute of Structural Engineering Method of Finite Elements II 28


Path-following methods: Arc-length

The sign of the increment is determined by the current stiffness


parameter:

fext T ∆u
κ=
∆uT ∆u

Institute of Structural Engineering Method of Finite Elements II 29


Path-following methods: Arc-length

Setting g = (∆up )T u − u1 + ∆λp λ − λ1 results in an


 

alternative method for arc length control (Riks 1972):


   
g i = (∆up )T ui − u1 + ∆λp λi − λ1

∂g ∂g
h= = ∆up , s = = ∆λp
∂u ∂λ

" #" # " #


KT −fext ∆u −Ri
=
(∆up )T ∆λp ∆λ −g i

In the above, superscript 1 refers to the predictor step

Institute of Structural Engineering Method of Finite Elements II 30


Path-following methods: Arc-length

Illustration of arc-length methods:

Crisfield: Riks:

Institute of Structural Engineering Method of Finite Elements II 31

Вам также может понравиться