solution of the element equations in part i · contents: textbook: examples: solution of the...
TRANSCRIPT
![Page 1: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/1.jpg)
Contents:
Textbook:
Examples:
Solution of theNonlinear FiniteElementEquations inStatic AnalysisPart I
• Short review of Newton-Raphson iteration for the root ofa single equation
• Newton-Raphson iteration for multiple degree offreedom systems
• Derivation of governing equations by Taylor seriesexpansion
• Initial stress, modified Newton-Raphson and full Newton-Raphson methods
• Demonstrative simple example
• Line searches
• The Broyden-Fletcher-Goldfarb-Shanno (BFGS) method
• Computations in the BFGS method as an effective scheme
• Flow charts of modified Newton-Raphson, BFGS, and fullNewton-Raphson methods
• Convergence criteria and tolerances
Sections 6.1,8.6,8.6.1,8.6.2,8.6.3
6.4, 8.25, 8.26
![Page 2: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/2.jpg)
kJ~ J>E'R\VE1) IN ,HE
"p~EVI OlA S LEe Tlif<ES
THE: F L EQuATION'>
r N Ttl IS LECTllRE WE:
(ONS,\1)E"R V A'R\OvS
TfCrlNIQUES OF
rTE1<~TION AN1)
(ONV ER5ENC. E
C.1<ITE"RIA
Topic Ten 10-3
Markerboard10-1
![Page 3: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/3.jpg)
104 Solution of Equations in Static Analysis - Part I
Transparency10-1
SOLUTION OF NONLINEAREQUATIONS
We want to solveHdtR _ HdtF = 0
externally applied nodal point forcesloads corresponding to internal
element stresses
• Loading is deformation-independent
T.L. formulation U.L. formulation
Transparency10-2
The procedures used are based on theNewton-Raphson method (commonlyused to find the roots of an equation).
A historical note:
• Newton gave a version of the methodin 1669.
• Raphson generalized and presentedthe method in 1690.
Both mathematicians used the sameconcept, and both algorithms gave thesame numerical results.
![Page 4: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/4.jpg)
Consider a single Newton-Raphsoniteration. We seek a root of f(x), givenan estimate to the root, say Xi-1, by
. _. _ f(Xi-1)x, - X'-1 f'(Xi-1)
Once Xi is obtained, Xi+1 may becomputed using
. _. _ f(Xi)X1+1 - XI f'(xj)
The process is repeated until the rootis obtained.
The formula used for a NewtonRaphson iteration may be derived usinga Taylor series expansion.
We can write, for any point Xi andneighboring point Xj-1,
f(Xi) = f(Xi-1) + f'(Xi-1)(Xi - Xi-1)
+ higher order terms
. f(Xi-1) + f'(Xi-1)(Xj - Xi-1)
Topic Ten 10-5
Transparency10-3
Transparency10-4
![Page 5: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/5.jpg)
1()..6 Solution of Equations in Static Analysis - Part I
Transparency10-5
Since we want a root of f(x) , we setthe Taylor series approximation of f(xi)to zero, and solve for Xi:
Transparency10-6
Mathematical example, given merely todemonstrate the Newton-Raphsoniteration algorithm:
Let f(x) = sin x , Xo = 2Using Newton-Raphson iterations, weobtain
o 2.01 4.1850398632 2.4678936753 3.2661862774 3.1409439125 3.141592654
error =
1.141.04.67.12
6.5 x 10-4 } quadratic
10-9 convergence
< is observed
![Page 6: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/6.jpg)
The approximations obtained usingNewton-Raphson iterations exhibitquadratic convergence, if theapproximations are "close" to the root.
Mathematically, if IEi- 1 1 -.:.. 10-m
then IEil' 10-2m
where Ei is the error in theapproximation Xi.
The convergence rate is seen to bequite rapid, once quadratic convergenceis obtained.
However, if the first approximation Xo is"far" from the root, Newton-Raphsoniterations may not converge to thedesired value.
Example: f(x) = sin x , Xo = 1.58
o 1.581 110.22920362 109.94871613 109.95574304 109.9557429] not the desired root
Topic Ten 10-7
Transparency10-7
Transparency10-8
![Page 7: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/7.jpg)
10-8 Solution of Equations in Static Analysis - Part I
Pictorially:
Transparency10-9
f
. '1sin x
I-'\I'-----+---~..........------x
Pictorially: Iteration 1
Transparency10-10
.5 slope f' (xo)
xo
f
I-'\I'-----+------:~--~::----x
![Page 8: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/8.jpg)
Topic Ten 10-9
Pictorially: Iteration 1Iteration 2
Transparency10-11
.5" slope f' (xo)
Xo
. 7sin x
I-'V'-------+-~,.----~------..:::,,_k__--- X
Pictorially: Iteration 1Iteration 2Iteration 3 Transparency
10-12
x,
.5" slope f'(xo)
slope f' (X2)
. 7Sin X
Xo
I-'V.'-------+-------"'Io.:-~k----~"'k_--- X
![Page 9: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/9.jpg)
10-10 Solution of Equations in Static Analysis -Part I
Transparency10-13
Pictorially: Iteration 1Iteration 2Iteration 3Iteration 4
Transparency10-14
Xo
Pictorially:
5' slope f'(xo)
slope f' (X2)lope f'(X3)
X3 Xl
fBad choice for Xo
f'(xo) = 0
I-JV''-----+-I----~------X
Xo Xo
![Page 10: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/10.jpg)
Topic Ten 10-11
Newton-Raphson iterations for multipledegrees of freedom
We would like to solvefeU) = HAtR - HAtF = Q
Transparency10-15
where now f is a vector (one row foreach degree of freedom). Forequilibrium, each row in f must equalzero.
To derive the iteration formula, wegeneralize our earlier derivation. Transparency
10-16
We writef(HAtUCi») = f(HAtUCi-1»)
+ [;6] (HAtUCi) - HAtUCi-1»)- l+~t1!(i-1)
.neglected to obtain aTaylor series approximation
+ higher order terms\
![Page 11: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/11.jpg)
10-12 Solution of Equations in Static Analysis - Part I
Transparency10-17
Since we want a root of f(U), we setthe Taylor series approximation off(t+ IitU(i») to zero.
Transparency10-18
or
[~}[:n +
af1 •.. af1
aU 1 aUn
afn ... afnaU 1 aUn
l+ t1tU(H) a squarematrix
![Page 12: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/12.jpg)
We now use
af I [-at+~ri'l0 [at+~tF(i-1)] IaO t+4Iy(H) =, ~~ It+4IU(H)- a1l t+4Iy(i-1)
because the loads aredeformation-independent
= _t+~tK(i-1)
--=--the tangent stiffness matrix
Important: HAtK(i-1) is symmetric because
• We used symmetric stress and strainmeasures in our governing equation.
• We interpolated the realdisplacements and the virtualdisplacements with exactly the samefunctions.
• We assumed that the loading wasdeformation-independent.
Thpic Ten 10-13
Transparency10-19
Transparency10-20
![Page 13: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/13.jpg)
10-14 Solution of Equations in Static Analysis - Part I
Transparency10-21
Transparency10-22
Our final result is
This is a set of simultaneous linearequations, which can be solved forau(i). Then
HatU(i) = HatU(i-1) + au(i)- - -
This iteration scheme is referred toas the full Newton-Raphson method(we update the stiffness matrix ineach iteration).
The full Newton-Raphson iterationshows mathematically quadraticconvergence when solving for theroot of an algebraic equation. In finiteelement analysis, a number of requirements must be fulfilled (for example,the updating of stresses, rotationsneed careful attention) to actuallyachieve quadratic convergence.
![Page 14: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/14.jpg)
We can depict the iteration process intwo equivalent ways:
Topic Ten 10-15
f = t+~tR _ t+~tF(i-1)
t+.:ltF(i-1)
u
Modifications:
load I t+.:ltK(i-1)sopet+~tR _
displacement
This is like a forcedeflection curve. We usethis representation henceforth.
Transparency10-23
Transparency10-24
'OK ~U(i) = HAtR _ HAtF(i-1)- -
• 'T = 0: Initial stress method
• 'T = t: Modified Newton method
• Or, more effectively, we update thestiffness matrix at certain times only.
![Page 15: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/15.jpg)
10-16 Solution of Equations in Static Analysis - Part I
Transparency10-25
Transparency10-26
We note:
• The initial stress method and themodified Newton method are muchless expensive than the full Newtonmethod per iteration.
• However, many more iterations arenecessary to achieve the sameaccuracy.
• The initial stress method and themodified Newton method "icannot"exhibit quadratic convergence.
Example: One degree of freedom, two loadsteps
force
displacement
![Page 16: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/16.jpg)
Initial stress method: 'T = 0Example: One degree of freedom, two loadsteps
forceAll slopes oK
I displacementlU(l) lU(2) 2U(1) 2U(2) 2U(3) 2U(4)
Line searches:
We solve
and consider forming t+~tF(i) using
where we choose 13 so as to maket+~tR - t+~tF(i) small "in some sense".
Topic Ten 10-17
Transparency10-27
Transparency10-28
![Page 17: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/17.jpg)
10-18 Solution of Equations in Static Analysis - Part I
Aside:Transparency
10-29
Transparency10-30
If, for all possible U, the number
UT (H£ltR - H£ltF(i») = 0
then H£ltR _ H£ltF(i) = Q
any rowReason: consider of .u
UT = [0 0 0 1
This isolates one row ofH£ltR _ H£ltF(i)
During the line search, we choo·seU = ~O and seek ~ such that
~OT (H£ltR - H£ltE(i») = 0
a function of ~
since t+~tU(i) = t+~tU(i-1) + ~ ~O
In practice, we use
o 0]
~OT (H£ltR _ H£ltF(i»)~OT (HatR _ HatF(,-1») -< STOL
- - - a convergencetolerance
![Page 18: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/18.jpg)
BFGS (Broyden-Fletcher-GoldfarbShanno) method:
We define~(i) = H.l1tU(i) _ H.l1tU(i-1)
lei) = H.l1tF(i) _ H.l1tF(i-1)
and want a coefficient matrix such that(H.l1tK(i») ~(i) = lei)
Pictorially, for one degree of freedom,
Thpic 'Tha 10-19
Transparency10-31
load
t+dtU(I) displacement
Transparency10-32
![Page 19: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/19.jpg)
10·20 Solution of Equations in Static Analysis - Part I
Transparency10-33
Transparency10-34
• The BFGS method is an iterativealgorithm which produces successiveapproximations to an effectivestiffness matrix (actually, to itsinverse).
• A compromise between the fullNewton method and the modifiedNewton method
Step 1: Calculate direction ofdisplacement increment
~O(i) = (t+~tK-1)(i-1) (t+~tR _ t+~tF(i-1»
(Note: We do not calculate the inverseof the coefficient matrix; we usethe usual ~ 0 ~T factorization)
![Page 20: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/20.jpg)
Step 2: Line search
HAtU(i) = HAtU(i-1) + f3 ~O(i)
a function~of ~
~O(i)T (HAtR _ HAtF(i»)
~O(i)T (Hl1tR _ Hl1tF(' 1») < STOL
Hence we can now calculate ~(i) and "1(i).
Step 3: Calculation of the new "secant"matrix
(HAtK-1)(i) = A(i)T (HAtK-1)(i-1) A(i)
where
A(i) = I + V(i) W(i)T- - - -~(i) = vector, function of
~(i), "1(i), HAtK(i-1)
W(i) = vector, function of ~(i), "1(i)
See the textbook.
'lbpic Ten 10-21
Transparency10-35
Transparency10-36
![Page 21: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/21.jpg)
10-22 Solution of Equations in Static Analysis - Part I
Transparency10-37
Transparency10-38
Important:
• Only veqtor prodl:Jcts are needed toobtain ~(I) and W(I).
• Only vector products are used tocalculate ~O(i).
Reason:
~O(i) = {(! + W(i-1) ~(i-1)T) •••
o+ W(1) ~(1)T) TK-1 0 + ~(1) W(1)T)
... a+ ~(i-1) W(i-1)T)} X
[t+atR _ t+atF(i -1 )]
![Page 22: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/22.jpg)
In summary
The following solution procedures aremost effective, depending·on theapplication.
1) Modified Newton-Raphson iterationwith line searchestK aO(i) = t+4tR _ t+4tF(i-1)- - - -
t+4tU(i) = t+4tU(i-1) + ~ aO(i)'---'
determined by theline search
2) BFGS method with line searches
3) Full Newton-Raphson iteration withor without line searches(full Newton-Raphson iteration withline searches is most powerful)
But, these methods cannot directly beused for post-buckling analyses.
'Ibpic Ten 1()"23
Transparency10-39
Transparency10-40
![Page 23: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/23.jpg)
10-24 Solution of Equations in Static Analysis - Part I
Modified Newton iteration with linehsearc es: It+Atu(O) = tu, t+AtF(O) = tF i = 11- -'
ICalculate tK1I i = i + 1 I ItK dO(i) = t+AtR - t+AtE(i-1>l
1t+AtU(i) = t+AtU(i-1) + 13 dO(i)l
f Is t+AtR - t+AtE(i-1) . 0Perform line search_to determine 13 with dO(i) small?
No !ves
Transparency10-41
BFGS method:Transparency
10-42It+Atu.(O) = tu , t+AtF(O) = tF, i = 11
~
ICalculate tKIli=t 11--------j
Update inverse ofsecant matrix ICalculate ~Q(i) I
t ~Perform line search Is t+AtR _ t+AtE(i-1) == Q
to evaluate t+AtU~i) • No with ~O(i) small?t+AtF(i)
~ Yes
![Page 24: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/24.jpg)
i=i+1
Full Newton iteration with line searches:
IH dt!J.(O) = t!J.. H dtE(O) = tE, i = 11
.~
ICalculate Hdt.ts(i-1)1
+IHdtK (i-1) liO(i) = HdtR _ HdtF(i-1)1t . - - - -.
IHdtU(i) = HdtU(i-1) + R liO(i)1 I- t- to'- !
Perform line search Is HdtR - t+dtE(i-1) ..:.. 0
to determine ~ No with liQ(i) small?
!ves
Convergence criteria:
• These measure how well the obtainedsolution satisfies equilibrium.
• We use1) Energy
2) Force (or moment)
3) Displacement
Topic Ten 10-25
Transparency10-43
Transparency10-44
![Page 25: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/25.jpg)
10-26 Solution of Equations in Static Analysis - Part I
Transparency10-45
On energy:
(Note: applied prior to line searching)
Transparency10-46
On forces:
IIt+dtR - t+dtF(i-1)1I2
RNORM <: RTOL" 4 -#
reference force(for moments, use RMNORM)
Typically, RTOL = 0.01RNORM = max IitRI12
'-"
considering only translationaldegrees of freedom
Note: lIal12 = ~~ (11k)2
![Page 26: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/26.jpg)
On displacements:
IlaO(i)112 < DTOlDNORM
~reference displacement(for rotations, use OMNORM)
Topic Ten 10-27
Transparency10-47
![Page 27: Solution of the Element Equations in Part I · Contents: Textbook: Examples: Solution of the Nonlinear Finite Element Equations in Static Analysis Part I • ShortreviewofNewton-Raphsoniterationfor](https://reader031.vdocuments.us/reader031/viewer/2022022508/5ace0a427f8b9a56098b470f/html5/thumbnails/27.jpg)
MIT OpenCourseWare http://ocw.mit.edu
Resource: Finite Element Procedures for Solids and Structures Klaus-Jürgen Bathe
The following may not correspond to a particular course on MIT OpenCourseWare, but has been provided by the author as an individual learning resource.
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.