Sie sind auf Seite 1von 3

about FEA

What a Manager should know


about Finite Element Analysis
Derek Pashley, a member of

at your manager should know


Nafems ETWG, takes a look at wh

Design Process

"Manager" 1
as in the title, means anyone from
immediate supervision upwards. If
your manager(s) needs educating
about FEA, show them a copy of
this as a start. If he/she/they are still
talking to you, get a copy of
Adams1.
Contributors to this article are
mainly Stress Engineers or similar.
Some points apply to any topic in
addition to stress or "structures"

Think about where FEA is


placed in your design process;
hand calcs and|or FEA before
CAD (for 3D definition) is smart.
(There's a use for CAD programs to
produce a preliminary design too.)
FEA has ramifications beyond CAE
input data comes from the real
world and components are made
there too. FEA and simulation arent
really CAD add-ons or a subset of
CAE. Its about understanding
everything youve done to date and
what you are about to do. People
who get this right are those who
see FEA as an extension of hand
calculation and design by numbers,
people who get this wrong think
the computer has a hotline to a
greater truth beyond our
comprehension.

Analysis by FEA after CAD, is


NOT design, it's autopsy (or
design fault-finding).
This is an important point: FEA is
often used too late to be effective
and just flags up problems rather
than solving them, but
customers, internal or
external, sometimes have to
have their need met, even
though it may diminish
the efficacy of the design
process.

CAD is for making


shapes CAD is
no more Design
than MS Word is
Twelfth Night. In less
hype-worthy times the
D used to mean
Drafting. Maybe by CAD
we mean finalised production
definitions and doing the FEA
after doing this is almost as daft
as doing it after youve made
stuff. Apologies for being off
subject of FEA.

about your analysis process.

If you must go round the


draw it, stress it change it
loop then at least use CAD,
particularly for 3D definition, with
FEA in mind in order to help mesh
generation. Otherwise you put a lot
of time in before the stress it bit.
Instant detail not only costs but it
almost always kills analysis projects
before they get going

Dimensional reduction and


feature removal of 3D models
is hard and expensive and
probably shows your design process
is not very sensible. Instant detail
not only costs but it almost always
kills analysis projects before they get
going.

Acceptance criteria must be


changed in accordance with
model style. Examples from
gas turbines: (1) shell models of
compressor blade cf 3D continua the latter resolve stress
concentrations where the blade
shape runs into the platform or
root; for shells the concentration is
not resolved and hence is built into
the allowable. (2) 2D axisymmetric
models of a turbine disc cf 3D
continua the latter resolve stress
concentrations at the root fixings;
for axisymmetry the concentration is
not resolved and hence is built into
the allowable.

7
8

It is false economy not to insist


on proper documentation of
an analysis. (This applies to
documenting programs as well).
Using different people for
meshing and analysis is asking
for trouble. When driving
anywhere keep steering and
changing gear in the same
department and dont get your
spouse to change gear and the kids

benchmark january10

21

about FEA

...it is easy to get carried away and finish up


with an analysis job that is too big for your kit.
to brake. This sounds stupid but
isnt any more stupid than splitting
meshing, solution and post
processing. The reason is that
analysis and post-processing often
mean feedback to a prevous stage
do the job properly.

9
10

Correlation with physical tests


is very useful, if not imperative.
Remember that test
measurements are approximate too.
... and sadly, in some
people's experience, the
higher up an organisation
you go the more they get into a
simulation means we dont have to
test mindset. It can mean less
testing however.

11

Analysis is not carried out


for sake of analysis. It can
be required to assess and
substantiate a design to
demonstrate that it meets its
functional and safety requirements.
Also an individual analysis can be
part of a design optimisation task or
a robustness assessment.

Outside the Design Process

12

FEA can also be useful in


post-mortems/failure
investigations; in such
cases data on the root cause may be
in short supply, emphasizing caution
is needed and collaboration with
other parties (point 32)

13

FEA can also be used in


in-service management,
e.g. Fatigue and|or creep
life estimation of nuclear plant or
flying gas turbines.

Analysis Definition

14

A fully converged
solution (that is, the
numerical error is
reduced to sensibly zero) is still only
an approximation to a solution for a
nominal design, because of
uncertainty in loading and the
representation of adjacent parts.

22

benchmark january10

What is designed is NOT


what is made because of
material variability,
dimensional tolerances, surface
finish even, etc. See Thacker[4].

15

21

FE Technology

Programs

16
17

22
23

The solution is
approximate.

Linear, continuum models


can be automated for
meshing and error
estimating = "automatic analysis".
But there are limits to "automatic
analysis" because you still have to
pose a meaningful analysis task that
approximates to the physics. Also it
is easy to get carried away and
finish up with an analysis job that is
too big for your kit.

18

Beams, shells etc. models


are much harder to
develop and interpret,
but cheaper and quicker to solve.
A big opportunity in CAE is to link
CAD models to these idealisations
and get the cart/horse combination
in the right order.

19

"10 element models"


can provide more design
information than 2
million tetrahedra. The Object of
Computing is Insight[2]. Big tet
models may have their place though
... if they cost little to create and
analyse or you can't get the design
information any other way.

20

If (mesh size < crystal


dimension) then it's a
WOMBAT (waste of
money, brains and time) because
the usual homogeneity assumption
fails to be reasonable.
Of course if your model is one of
actual crystals and perhaps you are
trying to track crack growth, or
something similar, then it might be
OK.

if (mesh size < length |


area over which you
need the stress
averaged>>) then it's another
WOMBAT. I could be wrong in some
cases but please think about it!

Programs usually have


errors - testing is needed,
unless all you want is
pretty pictures. See[3].
Graphical displays can be
deceptive. e.g. magnified
displacements can
mislead when looking at
interferences, gaps and contacts.

24

Post processing often


gives pictures of
response smoothed over
the mesh; this gives prettier
pictures, but unaveraged stresses (or
whatever) are more informative.

25
26

Licences can be
expensive ...

... Or too cheap.


Meaning cheap and
nasty, or cheap and bugridden, or cheap and
undocumented or cheap and
unsupported, or cheap and people
dont take it seriously, leading to the
CAD add on stress checker = spell
checker view.

27

Recognise that FEA can


cost. Executing programs
sometimes needs
machine grunt and space. Making
do with kit that is not up to the job
is another WOMBAT.

People

28

People need education in


the technology and
maybe training in the use
of individual programs. You have to
invest in people and it takes time;
training bolted-on to a sales
package is rarely enough.

about FEA

31
32

Managers and analysts


should understand the
need for 'Verification'
and 'Validation', see[5].
Encourage analysts to
work closely with those
from other disciplines
such as Materials, Plant Engineering,
Non-Destructive Testing, The
Customer even since this will
minimise the risk of missing
something important.

33

Your people need to


know and work within
their limits and get
advice if going outside their- what
Vince Adams calls personal
problem solving environment - and
learn to widen it.

34

Managers/Staff should
define their competency
and plan their
development to fill gaps and ensure
succession plans are robust.

35

Encourage analysts to go
to "User Group"
meetings; personal
contact with other users can be
useful. Some say that going to
"User Group" meetings is essential.
See also personal problem solving
environment above.

29

Your competition has


access to similar tools.
It's how you use them
that matters, which really means
that the people matter more than
the program.

30

If you have a decent


education and can use
one of the "traditional"
codes, you don't really need
expensive training in how to read a
user's guide for the other traditional
programs.
This doesn't mean that vendor
training is not useful, just be careful
where you spend this year's training
budget.(Vendors, and others, may
try to make money by providing
expensive training). This is a
contentious point.

Your
competition
has access to
similar tools.
It's how you
use them that
matters...

Acknowledgment
This article has been produced with
help from Laurence Marks, in
particular, and the NAFEMS ETWG
(specifically Adib Becker, Trevor
Hellen, Nawal Prinja. Andy Morris
and Mark Chillery). However, any
errors are mine, so if you want to
argue or just complain, contact me
at dandh.pashley@talktalk.net.

References
1) Adams V How to Manage Finite
Element Analysis in the Design
Process, Glasgow; NAFEMS 2006.
2) Hamming R Numerical methods for
Scientists and Engineers New York,
NY;McGraw Hill 1973
3) Pashley D G F.E. Software testing,
Benchmark Magazine, NAFEMS,
April 2009.
4) Thacker B H Why Do Probabilistic
Finite Element Analysis, Glasgow,
NAFEMS, 2008.
5) Performance Test Code Committee
60, Verification and Validation in
Computational Solid Mechanics,
ASME V&V 10-2006,
www.asme.org
6) Marks L Tips and Work Around for
CAD generated models.. Glasgow,
NAFEMS, Revised and Republished August 2008

AUTHOR INFORMATION
Derek Pashley I NAFEMS ETWG
dandh.pashley@talktalk.net

De re k Pa shle y sp
en t
Ch ri st m as an d
Ne w Ye ar
vi si ti ng fr ie nd
s in A us tral ia
ag ai n.
Hi s Ne w Ye ars
re so lu ti on is to
st op pl ac e
droppi ng . Hi s se
co nd Ne w Ye ar
s
re so lu ti on is to
st op sh ow ing of
f.

36

Fred Bloggs N.R.A.


(NAFEMS Registered
Analyst) is worth more
than Derek Pashley Ph.D. (Unless
you have rigorously interviewed
me!!) Because NAFEMS Registered
Analyst is about FEA; a Ph.D. could
be about anything.

benchmark january10

23

Das könnte Ihnen auch gefallen