Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

CE221 - Nonlinear Structural Analysis Lecture 7 Filip C.

Filippou
Prepared by Filp C. FILIPPOU
Newton-Raphson Solution Strategy
Recall nonlinear equilibrium equations to be solved P − Pr ( U ) = Pu ( U ) = 0
where the force unbalance vector is the different between applied and resisting forces and the loading is assumed
to be displacement independent (exceptions: fluid pressure under large displacements, inertial effects, etc.)
For given loading P develop a linear incremental solution with Newton-Raphson strategy, i.e.

∂Pu
Pu ( U ) ≈ Pu ( U 0 ) + ( U − U0 ) = 0 where U0 is an initial "guess" for the solution
∂U 0 −1
⎛ ∂P ⎞
with this linearization the initial guess can be improved by applying successively: U = U 0 − ⎜ u ⎟ Pu ( U 0 )
⎝ ∂U 0 ⎠
noting that the partial derivative of the force unbalance vector with respect to the
displacement vector is the opposite of the tangent stiffness matrix we can write U = U 0 + K 0−1 Pu ( U 0 )

successive application of this process is denoted with a subscript for iteration U j = U j −1 + K −j −11 Pu ( U j −1 )
for j=1.. no_iter
Thus, given an initial "guess" for the solution U0 we proceed as follows for each iteration j

1. determine the resisting force and current stiffness for the displacement value, i.e. U j −1 → K j −1 , Pr j-1
this step is known as state determination for the structure; it breaks down into several steps for models with
complex structural elements, as will be shown in the following

2. determine force unbalance and displacement correction (


ΔU = K −j −11 P − Pr j -1
)
3. update displacements (solution estimate) and go back to step 1 U j = U j−1 + ΔU
This process results in the graph on the next page for a given applied force; the corresponding Matlab script
is also included, noting that the assignment operator U = U+ΔU instructs the program to replace the value in
the memory slot reserved for U by the value U+ΔU
2
CE221 - Nonlinear Structural Analysis Lecture 7 Filip C. Filippou
Convergence check Prepared by Filp C. FILIPPOU

For checking the convergence of the solution there are several options, each with its advantages and limitations

Option A: check the unbalance force, i.e. the difference between applied and resisting force Pu ( U ) = P − Pr ( U )

since Pu is a vector we use the Euclidean norm in the form Pu = Pu T Pu


and we check convergence by comparing against the norm of the applied force vector with a
specified tolerance η of the order of 10-8 Note that we use P, i.e. λ Pref in the tolerance check
Pu ≤ η P

Option B: check the work increment against its initial value; this is known as energy norm

the initial work increment is ΔW0 = ΔP T ΔU 0


Pu j
where ΔP is the load vector increment and ΔU0 the
corresponding displacement (vector) increment

ΔP the work increment during iteration j is ΔW j = Pu T ΔU j


j

convergence can then be checked with ΔW j ≤ η ΔW0

with a specified tolerance η of the order of 10-16

ΔU 0 ΔU j

7
CE221 - Nonlinear Structural Analysis Lecture 7 Filip C. Filippou

Load control Prepared by Filp C. FILIPPOU

in the preceding examples the load factor increment Δλ was assumed constant for all load steps and invariable during
the equilibrium iterations; this special case does not account for changes in the stiffness matrix of the structure; it is
advisable to use smaller load factor increments when the structure softens and larger when the structure hardens; equally
important, we should reduce the load factor during iterations, if we wish to capture softening behavior. The adjustment
of the load factor increment is known as load control strategy.

Load control during incrementation


this means that the load factor increment Δλ is adjusted at every load step depending on some criterion, it is subsequently
held constant during the equilibrium iteration phase; a suitable parameter for controlling the load factor increment during
load incrementation is the current stiffness parameter Sp proposed by Bergan et al. (1978).

with the definition of the stiffness parameter


for load step k

S pk = PrefT U tk = PrefT ( K k −1Pref )


Pref
we define the load increment for load step k
γ
⎛S ⎞
Δλk = Δλ0 ⎜ p 0 ⎟
⎝ S pk ⎠
Pref U tk where S p 0 = PrefT U t 0 = PrefT ( K 0 −1Pref )

the exponent γ ranges from 0.8 to 1.5, but


a value of 1 is most suitable

U t0 8
CE221 - Nonlinear Structural Analysis Lecture 7 Filip C. Filippou
Load control during iteration Prepared by Filp C. FILIPPOU

for tracing the softening post-peak response of structures we need to relax the requirement that the load factor remain
constant during equilibrium iterations; in fact, any other type of "constraint" can be used for determining the load factor
adjustment; one such constraint that is relatively simple to see is the requirement that the displacement increment of a
particular dof does not change during equilibrium iterations; we illustrate this in the following figure

λ let us assume that the particular displacement dof that we are trying
to control is n (e.g. the top of a multi-story building). Then according
to the NR procedure we have during iteration j

j −1
(
ΔU = K −1 λ j Pref − Pr j -1
) (see equation on page 2 with applied load P)

in this case the applied load is adjusted at every iteration, so the load factor
carries a subscript to denote this fact; the load factor at iteration j is equal
to the load factor at the preceding iteration plus a load factor increment, i.e.
λ j = λ j −1 + Δλ j
substituting into the above equation we get: ΔU = K −j−11 ⎡ λ j −1 + Δλ j Pref − Pr j-1 ⎤
⎣( ) ⎦
with Δλj the load factor increment to be determined

we denote with Pu j−1 = λ j −1Pref − Pr j-1 the unbalance force of the


completed iteration
Un
and with ΔU r = K −1 Pu
j -1 j −1 j −1
the corresponding increment

the above equation then becomes ΔU = Δλ j K −j−11 Pref + ΔU r

and recalling the definition of the "tangential displacement" Ut we have ΔU = Δλ j U t + ΔU r


j -1 j -1

setting now the increment of dof n to zero during iterations, as shown in the above figure, gives ΔU n = Δλ j U tn + ΔU rn = 0 j -1 j -1

ΔU rn
and solving for Δλj we get Δλ j = − j -1
this is typical for the type of equation we obtain for the load factor increment
U tn j -1 10
CE221 - Nonlinear Structural Analysis Lecture 7 Filip C. Filippou
Load control during iteration (cont'd) Prepared by Filp C. FILIPPOU

load control formulas during equilibrium iterations were obtained from the following constraints


1. minimum of displacement increment norm, i.e. from
∂Δ λ j
( ΔU TΔU ) = 0
U Ttj-1 Δ U r j-1
recalling that ΔU = Δλ j U t + ΔU r we get a linear equation for the increment Δλ j = −
j -1 j -1
U Ttj-1 U t j -1

T
Pref ΔU r j-1
2. constant external work, i.e
T
Pref ΔU = 0 we get a linear equation for the increment Δλ j = − T
Pref U t j-1
we note that the above constraint can be interpreted as the
displacement increment being normal to the reference load vector

3 .constant displacement at selected dof n as discussed in preceding page we get a ΔU rn


linear equation for the increment Δλ j = − j -1

U tn j -1

4 . arc-length control strategies

with quadratic equation for the increment (Crisfield)

linearization of nonlinear arc-length equation (Ramm and Riks-Wempner)

consult Crisfield's Volume 1, Chapter 9 or IJNME paper by Clarke and Hancock

11

You might also like