Papuga - Validation of fatigue prediction models and of fatigue solvers

Transcription

Validation of fatigue prediction models and of fatigue solvers
Jan Papuga
FADOFF consortium: http://www.fadoff.eu, [email protected]
FADOFF ‐ Fatigue Analysis Documentation OFFice, www.fadoff.eu, e‐mail: [email protected]
1
Multiaxial fatigue problem
 If more load channels interact a phase shift among their periodical loadings can occur
 The non‐zero phase shift between them can improve the fatigue properties (the maximums of the load are not concurrent)
 Also the combination of one static and one variable load channels – the common stress tensor reduction hypotheses fail
2
Criteria for Fatigue Limit Estimation
How to check them?

All fatigue criteria converted to the standard: D p  f 1

f‐1‐ fatigue limit in fully reversed axial loading
For an experimentally set multiaxial fatigue limit: D p  f 1

f‐1 – type of axial loading (push‐pull, bending, rotating bending) corresponds to the fatigue limit used in calculation Fatigue index error: FI   D P  f 1

f 1


 100%

3
One of the Simplest Solutions
 Signed von Mises stress
 Can be efficiently used also for loading with non‐
constant (or random) amplitude


1
 x   y 2   y   z 2   z   x 2  6  xy2   yz2   zx2
 (t ) 
2
PragTic  * (t )   (t )  signI 1 (t ) 
MSC.Nastran  * (t )   (t )  signmax( ,  ) 
1
3
(and PragTic)

a
m
Walker
4
..But the Results…
 Signed von Mises
 The difference between both signing variants is negligible overall  Optimum variant only for ductile materials and in‐phase loading with zero mean stress
VMI1
342 tests from 407
45%
40%
35%
30%
25%
20%
15%
10%
5%
37.5
32.5
27.5
22.5
17.5
12.5
7.5
2.5
‐2.5
‐7.5
‐13
‐18
‐23
‐28
‐33
0%
‐38
Relative occurence
 Problems
 Mean axial stress within multiaxial loading
 Mean torsion stress within multiaxial loading
FI [%]
5
MSC.Fatigue ‐ MAPS
 Provides the best overall results, though they are not very good
MAPS
335 tests from 407
45%
40%
40%
35%
FI [%]
37.5
32.5
27.5
22.5
17.5
12.5
7.5
2.5
‐2.5
‐7.5
37.5
32.5
27.5
22.5
17.5
12.5
7.5
2.5
‐2.5
‐7.5
0%
‐13
0%
‐18
5%
‐23
5%
‐28
10%
‐33
10%
‐13
15%
‐18
15%
20%
‐23
20%
25%
‐28
25%
30%
‐33
30%
‐38
Relative occurence
35%
‐38
Relative occurence
MAPS on {nMS‐IP; MS‐To; MS‐Ax}
173 tests from 190
45%
FI [%]
6
Damage Initiation on the Surface
Plane stress state :
(!not valid at contacts!)
surface
If x and y are principal stresses:
Maximum load according to
max (Tresca) hypothesis
(deviated by 45° from the
surface)
surface
xz plane
yz plane
Maximum load as regards the
maximum principal stress
(deviated by 90°from the
surface)
7
Multiaxial Fatigue Calculation
Common Assumptions

Induced damage ~ Crack ~ Plane

Load state description on that plane constitutes direct input for fatigue parameter

The shear parameters play usually the dominant role, normal parameters are secondary (but important anyhow)
Bannantine & Socie
8
Dang Van criterion
a DV  C a  b DV   H , max  f 1
 Critical plane criterion
 The most often used representative of multiaxial criteria
 Use of maximum hydrostatic stress does not seem to give acceptable results
 C: MS, Ax+Ax
 N‐C:
 nMS, OP
 MS,To
average: ‐0.1%
range: 92.9%
standard deviation: 12.2%
9
Fatigue limit solution today
 In fatigue solvers:
 FI – relative error between predicted and experimental fatigue limit
 FI =0 ideal
 FI >0 conservative
 FI <0 non‐conservative
10
Weakness of Dang Van method
 Hard to believe: multiaxial fatigue
relative difference between predicted
and experimental fatigue limit
no mean stress, out-of-phase loading
no mean stress, in-phase loading
 There is a significant difference in mean prediction values depending on the phase shift of individual load channels
11
Versions of Dang Van solution




Overall improvement rather weak
Better form of the histogram in nMS,OP experiments degraded by the shift of the mean value
Distinctions in the suitability for brittle and extra‐
ductile materials
Distinction in changes for Ax+Ax and Ax+To groups
DV – original solution (1974) as a critical plane solution
DVnew – new version (1989) based on transformation of the load path to the Ilyushin deviatoric space (solution by Martin Nesládek)
12
Recent attempt to improve Dang Van
 Change induced by  influence of biaxiality ratio
 influence of load ratio at the moment of maximum I1
 Apparently, the improvement doesn’t involve the phase shift effect
Louis Augustins: An empirical multiaxial high cycle fatigue criterion for
automotive cylinder head design. Engineering Failure Analysis 28 (2013), pp.
264-274.
13
Fatigue limit solution today
 In research papers:
Well, the software developers could try also Liu & Mahadevan
and maybe should dare to implement Crossland in order to
keep the prediction as bad as now
14
Manson‐McKnight
 So simple, that it can be computed in MS Excel
 !The cycle has to be detected a priori!
 Amplitude and mean value of each stress component is evaluated:






a 
1
 x,a   y,a 2   y,a   z ,a 2   z ,a   x,a 2  6  xy2 ,a   yz2 ,a   zx2 ,a
2
 m*  signI1,d 
1
 x,m   y,m 2   y,m   z ,m 2   z ,m   x,m 2  6  xy2 ,m   yz2 ,m   zx2 ,m
2
 The mean equivalent value is signed according to the stress tensor invariant with biggest magnitude
15
Manson‐McKnight ‐ Results
 Not that bad
 Shift to conservative prediction results in many cases (To; nMS‐OP; Ax+To; brittle materials)
MMK, 404 tests from 407,
Dang Van, 403 tests from 407
45%
40%
35%
Relative occurence
 Ax+Ax with a phase shift – unsafely non‐
conservative (mean value FI=‐17.2%)
 In many other groups Dang Van better, but fails in MS cases
MMK
30%
Dang Van
25%
20%
15%
10%
5%
‐38
‐33
‐28
‐23
‐18
‐13
‐7.5
‐2.5
2.5
7.5
12.5
17.5
22.5
27.5
32.5
37.5
0%
FI [%]
16
MMK versus Dang Van
DV – Dang Van critical plane method (1974)
MMKF –Manson‐McKnight according to Filippini (2010)
 MMK not to be used for
 brittle materials
 MS,Ax+Ax, PS<>0
 out‐of‐phase loading
 MMK useful for pressure vessels
 MS,Ax+Ax,noPS
 The difference is not big overall!
17
Versions of Papuga PCr solution
PCr – original solution (2008) presented in IJF
PCrN – new version (2011) including mean shear stress effect
 New special formula includes both Nm and Cm mean stress parameters – a part of the Nm
effect moved to Cm effect
 It improves mean value of MS,To group
 Substantially improves scatter by MS,Ax and MS,Ax+Ax experiments!
 Minor worsening in MS‐
Ax,Ax+To group – a further study will be done before a final publishing
18
What Is FADOFF?
 Abbreviation
 Fatigue Analysis Documentation OFFice
 Consortium focused on
 Fatigue Analysis – prediction methods
 Documentation – pros and cons of individual methods
 Office – providing the information to interested companies and universities
19
Who Established FADOFF?
 In 2011‐2014, Technology Agency of the Czech Republic
supported a 4‐year project of:
CTU in Prague
Evektor spol. s r.o.
VZÚ Plzeň s.r.o.
VŠB ‐ Technical University Ostrava
 At this moment, a spin‐off company is prepared ‐ further elaborating of FADOFF results planned
20
FADOFF Topics
Extension of topics previously elaborated by me
 PragTic fatigue solver development
 Internet databases forming the knowledge base for fatigue calculations
 Their implementation and linking into one database unit
 Feeding by data
 Checking the quality of fatigue prediction methods and of commercial fatigue solvers
 Search for ways enabling long‐term sustainable continuation
21
Why all this stuff should be done?

Not including implemented standards, the software producers 
do not present any more extensive verification of implemented methods

if they implement some method or standard, the potential deviation from it are not verified

It is common, that the producers disclaim any warranty with losses caused by the use of their software

How the end‐users know what they computed?

There is no independent authority that would check quality of different software solutions
22
Warranty disclaimer
 Researcher:
 Proposes a new criterion
 Proves its validity on limited data he has in hands
 His only (vague) responsibility is for these research results
 Solver developer:
 Selects and implements the method
 I do not know about any case, where further testing was sponsored by such a company with publicly available results
 Decides to what extent to release publicly details of the implementation (so that the competitors would not steal his ideas)
 Disclaims any responsibility for the use of the software
 End user ‐ engineer:
 Gets a very expensive tool in his hands
 Due to high price is forced to use it to maximum
 Does not have time enough to get through all the theoretical basis or validation studies (if there are any available)
 Would like to believe that the previous two persons were responsible
23
Conclusions
 The fatigue solvers are wonderful tools  simple to generate nice fatigue damage maps on a selected structural element
 But fatigue methods are empiric, fatigue is a weak link mechanism
 There is often present
 quite a big scatter of experimental results
 important uncertainty as regards load levels and courses, materials, ...
 The Help sections and the documentation of quality of used methods are poor
Users can get quickly to complete results without adequate knowledge what they mean
 The end users of the commercial fatigue solvers are the only persons responsible for results they get from it
Are they aware of it?
Should not they get at least maximum scope of information?
Do they get it?
24
Financial means in fatigue research
1960:
Industry
Industry
Industry
Industry
2010:
Industry
Research
Research
money
knowledge
technical means
Research
Research
Government
SW Developer
Industry
Industry
Industry





Research
Research
Government
Less money for research
More money for technical development
Loss of responsibility to the funds provider
A lot of an unused “trash” in the research
Industry shifts to a “mob”, the research is far from diversiform
25
… a personal note
“…The weld challenge thing has created some interest. People don’t really know
how to analyse fusion welds for the ground vehicle industry durability. The most
interest is usually expressed by the fatigue software suppliers – nCode, FeSafe,
MagnaSteyr. Of these Magna is probably the most willing to explore new
paths. The users don’t really want to do their own research anymore. They expect
nCode etc to "build-in" any new concept. Unfortunately this has stifled research
by others. The major fatigue research labs in N.America are dying off. …”
From personal communication with Al Conle, 2009
2010:
Industry
SW Developer
Industry
Industry
Industry
Research
Research
Government
26
Room for Verification Authority
 To bridge the gap between research and commercial application
 Need for verification of  methods implemented in SW
 the implementation ways themselves
 new calculation methods where a great potential of commercial implementation exists
27
NAFEMS case
„By the late 1970's and early 1980's, as computing power became more widely available,
increasingly industry was starting to solve practical engineering problems using finite
element analysis techniques.
There was however considerable concern that the accuracy of the methods, and software
implementations, required to be verified in order to allow the results to be effectively used.“
from: http://www.nafems.org/about/about_nafems/history/
 NAFEMS (National Agency for Finite Element Methods and Standards) established in 1983 „To promote the safe and reliable use of finite element and related technology“
 Funded for 7 years by UK government
 Then switched to a non‐profit organization funded by its activities and members
 Shift in the focus
 Originally – Benchmarks to test the fatigue solvers
 Now – Continuous education of FE‐analysts
28
FatLim database (2007)
 Database of fatigue limit experiments
 www.pragtic.com/experiments.php, www.pragtic.com/fatlim.php
 only smooth, unnotched specimens => focus on the multiaxial solution
 now there are more than 450 experimental fatigue limits
 407 experiments presented within my paper on multiaxial fatigue limit criteria in International Journal of Fatigue, 2011
 The web also publishes results of fatigue limit predictions
 Contains a selector to select special load conditions and to check how the gathered fatigue criteria can follow the experimental trends
29
FatLim ‐ Database of fatigue limits
 huge number of
fatigue limits
under complex loading
 www.pragtic.com/e
xperiments.php
 Selector:
30
Reasons for FatLim Internet Database
1.
proof of a suitability of my own method (see Int J Fatigue 1/2008)
2.
ensuring that only verified error‐free experimental data are used
3.
comparing different existing methods and sort them according to their results in different categories
4.
providing a tool, which would tell what quality of results you can expect for your type of loading and the material used
5.
overcoming problematic points noted in the next slides
31
Common problems – “Soft” Data
Evolution rule I: The sets of experiments which “meet wishes” of
the methods are preferred
40
30
20
50
Experimenty: kompletní současná FatLim databáze
(celkem 428 exp.)
Kritérium: Findley
40
30
20
10
10
0
0
-42,5
-37,5
-32,5
-27,5
-22,5
-17,5
-12,5
-7,5
-2,5
2,5
7,5
12,5
17,5
22,5
27,5
32,5
37,5
42,5
47,5
Experimenty: NKh, NKm, NKc (data Nishihara &
Kaw amoto - celkem 40 exp.)
Kritérium: Findley
-47,5
-42,5
-37,5
-32,5
-27,5
-22,5
-17,5
-12,5
-7,5
-2,5
2,5
7,5
12,5
17,5
22,5
27,5
32,5
37,5
42,5
47,5
-47,5
četnost
výskytu [%]
Occurrence
50
60
Occurrence
četnost
výskytu [%]
60
 FI [%]
 FI [%]
32
Common problem – lack of check
Usual testing sets
(data from IJF and FFEMS)
 the sets based on Papadopoulos’ proposal are those dominantly used
 [1] Nishihara & Kawamoto
 [33] Heidenreich et al.
 [35] Froustey & Lasserre
 [5] Lempp
33
Lack of check ‐ Nishihara & Kawamoto
 if asking for at least 4 completed experiments
 only 3 fatigue limits for hard steel are applicable
 even here the regression for both material curves f‐1 and t‐1 is doubtful (R2 too low, because of being computed for low number of data points)
34
Common Problems – „Softened“ Data
Evolution rule II: If some experiment in the set of experiments does not suit you,
remove it.
Repeated reproduction of data from secondary sources prevents detection of such
cases and also hides the originator of the softening
NINIC, D.; STARK, H. L.:
2007, pp. 533-548.
Case
Case
FI801
FI801
FI802
FI802
FI803
FI803
FI804
FI804
FI805
FI805
FI806
FI806
FI807
FI807
FI808
FI808
FI809
FI809
FI810
FI810
FI811
FI811
Remark
Remark
PB
PB
PB
PB
PB
PB
PB
PB
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
PB+To
A multiaxial fatigue damage function. Int. Journal of Fatigue 29,
aa [MPa] mm [MPa]
157,9
143,4
138,6
131,0
161,4
124,8
124,1
115,2
103,4
102,8
79,4
82,8
206,9
310,3
448,3
0,0
0,0
82,8
206,9
310,3
0,0
0,0
a [MPa]
0,0
0,0
0,0
0,0
33,4
62,4
62,1
57,6
51,7
75,2
95,9
m [MPa]
0,0
0,0
0,0
0,0
0,0
0,0
41,4
103,4
155,2
0,0
0,0
deg
FI
0
0
0
0
0
0
0
0
0
0
0
12,0%
28,8%
43,4%
58,3%
-0,5%
-7,2%
11,2%
33,6%
43,5%
-8,7%
-0,9%
FINDLEY, W.N.: Combined-stress fatigue strength of 76S-T61 aluminum alloy with superimposed
mean stresses and corrections for yielding. NACA Technical Note 2924, May 1953.
35
FADOFF –> Database tools



Databases transferred www.pragtic.com ‐> www.fadoff.eu
Storing and managing data necessary for validation

RefDB – Database of papers and books related to fatigue

MatDB – Database of material properties

FinLiv – Database of experiments

FatLim – Database of fatigue limit experiments
Providing additional services

Experts

Labs
36
Job File for FatLim ‐> PragTic

PragTic fatigue solver implemented a simple solution to check the fatigue limit estimation methods



different load regimes
different materials
on all experimental data items by a single click

Available to public as Job File on www.pragtic.com/FL_job.php

Anybody can download PragTic freeware, this complete job file and check the implemented methods
But this verification concerns fatigue limit calculation methods implemented in PragTic only

37
FADOFF –> Fatigue Limit (Dummy Model)


New solution prepared within FADOFF
APDL script in ANSYS to prepare a single fictitious (dummy) model, where on elements and nodes are




predefined material properties history of stress tensor components
325 experimental items now
Effect:

Any user of a commercial fatigue solver can
 use it as an input for preparing fatigue prediction
 solve it by the methods implemented in his fatigue solver
 check the prediction capability of his fatigue solver
38
Some output…
 Two commercial fatigue solvers (CFS) available for testing
 Two more should be checked this year
 CFS#1 (Dang Van)
 ¾ of results agree very well with PragTic (Dang Van), the rest does not
 no similar markers for failure, results tend to be too unsafe
 will be communicated with the developer
39
Evaluation of Dummy Model
 Very limited focus of validation (no stress gradients, no notches, HCF only)
 Describes quality of the damage parameter (conversion from stress tensor history to the equivalent stress history)
 Neglects all the other aspects (roughness, size effect, quality of load history data,…)
 OK – but it’s adoption, testing phase and evaluation is very fast
40
Shear mean stress effect
 Papadopoulos, IJF 1997:
 There is no mean shear stress effect in HCF
41
Mean stress effect
1.8
1.6
Effect of mean stresses
AMP-To, MS-Ax
AMP-To, MS-To
Lineární (AMP-Ax, MS-To)
Lineární (AMP-To, MS-Ax)
AMP-Ax, MS-To
AMP-Ax, MS-Ax
Lineární (AMP-To, MS-To)
Lineární (AMP-Ax, MS-Ax)
 a /f -1; a/t -1 [-]
1.4
1.2
1
0.8
0.6
0.4
0.2
0
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
 m /  y ;  m /  y [-]
42
Versions of Papuga PCr solution
PCr – original solution (2008) presented in IJF
PCrN – new version (2011) including mean shear stress effect
 New special formula includes both Nm and Cm mean stress parameters – a part of the Nm
effect moved to Cm effect
 It improves mean value of MS,To group
 Substantially improves scatter by MS,Ax and MS,Ax+Ax experiments!
 Minor worsening in MS‐
Ax,Ax+To group – a further study will be done before a final publishing
43
FADOFF –> FinLiv

Keeping experimental data in a standardized format

notched and unnotched specimens

S‐N, ‐N curves, fatigue limits

complex load histories, more load channels available

interpolation functions

export to FEA (Ansys), PragTic to maximize efficiency in validating procedures
FinLiv for Excel
csv file
csv file
APDL macro
FinLiv DB csv file
(geometry)
FE-model
(material
properties,
loads)
PragTic
Ansys FEA
Fatigue results
result files
(geometry,
stresses, strains)
44
FinLiv – preprocessing of fatigue analysis
Dummy Model
xls file
FinLiv on‐line
csv file
APDL
Specimen
FEA solver
Load data
FinLiv.VBA
FEA results
PragTic and other fatigue solver
Show
Excel file
Updated csv file
Selection of section points
Summary of fatigue curves
Regression
analysis
Representation of a fatigue curve points
Representation of a fatigue curve points
…
45
FinLiv.VBA – Part I
Test group
 Set of experiments on a material with the same thermomechanical treatment
Fatigue curve ID
and type
Command buttons
 Back Up – sends the data to a separate csv file
 Show – generates xls file with separate
sheets for separate fatigue curves
Fatigue specimens
 Standard fatigue specimens are of several common types
 They are classified, parameters assigned
 APDL Specimen ‐ generates an input file for the automated build of an FE‐models
46
FinLiv.VBA – Part II
 Example of load definition
 Integrating also random sequences
47
FinLiv on‐line
48
Reference database
49
Material database
50
FADOFF – other parts




Experts
Labs
Contracts
Marketplace for fatigue curves
51
FinLiv – preprocessing of fatigue analysis
 Automated evaluating notched component is more complicated compared to Dummy Model
APDL
Specimen
FinLiv.VBA
Load
data
FEA solver
FEA results
PragTic and other fatigue solver
 Now ready only for PragTic
 Can be extended also to other SW equipped with macro‐language
52
Description of specimens
 Various flat, bar and tube specimens with parametrized notches
 Bar and tube specimens usually allow preparation of axisymmetric specimens
 ANSYS allows application of non‐axisymmetric loads to axisymmetric models
 Example:
SPC2 –Solid bar of circular cross-section, circumferential V-groove
2 variants of SPC2; (VERSION A – TYPE I,II)
v4
v3
v1
v2
v5
53
BCs and run
 FinLiv generates a text file describing
Y
 Geometry parameters
 Load parameters
X
Z
 APDL macros allow to set boundary conditions and run the analysis
 4 different load modes – tension, torsion, bending, pressurizing
 multiple axisymmetric and flat topologies
You can get the FE-model,
load it
and solve it
within 1 minute
54
FinLiv‐>Show
 csv data used to generate formatted Excel data sheets
 Active buttons for running regression analysis
55
Key features
Human genome report
 Private
 Belongs to the individual human, who paid to get it
 Cost continually decreases (1000$ in 2014)
 Standardized item ‐ a big effort spent on making it cheaper
 The person gives the consent while getting back the data about the potential risks discovered in his/her genome
Results of fatigue experiments
 Private
 Belongs to the company, which paid to get it
 Costs doesn’t seem to be decreased soon
 Things can get complicated (multiaxial loading, thermal load, …) and thus more expensive
 Data published only in rare cases (PhD studies, grant projects), a lot of them remain undisclosed
56
Our goal
 Building a unique system




to save fatigue experimental data
to make them searchable
to make them accessible
to simplify their next use
 Finding a way  to fund it
 to extend it
57
Experimental program
2124T851 aluminium alloy



All specimens made from one plate
Load combinations: Tension, Torsion, Plane bending
Specimens
 unnotched hollow specimens and bars
 notched bar specimens
58
Experiments on 2124T851
59
Micronotch effect
60
Micronotch effect
61
Micronotch: stress distribution change
62
Acknowledgement
The support of the Technology Agency of the Czech Republic within TA01011274 project is gratefully acknowledged
63

Papuga - Validation of fatigue prediction models and of fatigue solvers

Transcription

Similar documents

7pm Where

Coupled Dynamic-Fatigue Analysis of a Haulpak 730E

FEMFAT software - ECS

FTL Presentation 9 - Civil Aviation Authority

EDAN AFE - SmartCap

Orbitol Plus

creature feature - Pinnacle Entertainment Group

Fatigue Assessment

The Eye: Physiology, Diseases, and Educational Implications