🚀 go-pugleaf

This is by no means a trivial question.  You must first ask the class of FEA to
be considered:
 
    1)  rough analysis for simple scaling
    2)  detailed analysis for design verification

The latter is the most typical use of FEA an basically requires a unix type
workstation or above - the memory limitations of dos usually make dos solutions
prohibitive in 3D, which what is generally required for desing verification (our
typical models are on the order of 7000 to 40,000 nodes).

You must also consider whether you plan to do general nonlinear analysis, or only
linear analysis.  A quick review of cpu cycles utilized in a facility suggests
that even if a relatively small number of analyses are nonlinear, then virtaully
all cpu cycles are being used for nonlinear analysis, and you should  design your
facility as a nonlinear facility.

It is also imperative that you have  a good mesh generator and graphical post
processor.  The difference between a good mesh generator and an ok mesh generator
is about 20X in productivity.

Finally you will  need color hardcopy capability in order to distribute your
results.
 
Typical linear codes (all of whcih work at least ok) are:
 
     ANSYS
     NASTRAN

Typical nonlinear codes are:
 
     ABAQUS
     ADINA

You might also look at Marc if it solves special needs of yours.

Commericial pre/post processors include PATRAN (I recommend rev 2.5 for now) and
SUPERTAB from SDRC.

You should also consider one of the many color postscript printers (we have
several from seiko and Tektronix).

Finally you want to get the fastest risc worklstation you can get.  We have
tested the HP9000/730, SGI R4000 based systems, and other of the new 64 bit
floating point chips.  Get yourself 2 1GB dfisc drives if you are doing this for
real.

A typical system ight cost the following (rough costs):
  
     1)  FEA program $5K - $10K year

     2)  Mesh generator $10K yr

     3)  Workstation:  $25K
 
     4)  color postscript printer $5K to $10K

This is nearly a 1/3 cost reduction and 10,000 performance improvement over the
system I put together is 1983!


-- 

Regards.
 
Don Rolph  a722756@pan.mc.ti.com WD3 MS10-13 (508)-699-1263

In article <1992Dec14.181725@roper.mc.ti.com>, a722756@roper.mc.ti.com (W. Donald Rolph) writes:
|> 
|> This is by no means a trivial question.  You must first ask the class of FEA to
.......
|>  
|> It is also imperative that you have  a good mesh generator and graphical post
|> processor.  The difference between a good mesh generator and an ok mesh generator
|> is about 20X in productivity.
........
|> 
|> A typical system ight cost the following (rough costs):
|>   
|>      1)  FEA program $5K - $10K year
|> 
|>      2)  Mesh generator $10K yr
|> 
|>      3)  Workstation:  $25K
|>  
|>      4)  color postscript printer $5K to $10K
|> 
|> This is nearly a 1/3 cost reduction and 10,000 performance improvement over the
|> system I put together is 1983!
|> 
|> 
|> -- 
|> 
|> Regards.
|>  
|> Don Rolph  a722756@pan.mc.ti.com WD3 MS10-13 (508)-699-1263

Thanks for a very informative post. Here's another question. What are
the latest in Mesh Generation techniques? 

The last time I dabbed in FEM, the finite-octree (w Delaunay triangulation) 
appeared the most promising but there were some concerns about robustness. 
I know that companies like PDA engineering (developers of PATRAN) were 
looking at "paving". What is the current status of these two techniques, 
and are there any new ones out there (AI approaches etc.)?

Raman Bhatia (bhatia@bigcat.bellcore.com)

In article <1992Dec14.212505.27326@cs.wayne.edu> landman@splat.physics.wayne.edu writes:
>From: landman@splat.physics.wayne.edu
>Subject: FEA help!
>Date: Mon, 14 Dec 1992 21:25:05 GMT
>
>Hello Net.People:
>
>
>    I have a question about FEA, specifically software and hardware
>platforms.  A friend is looking into the possibility of purchasing 
>hardware and software to perform FEA calculations for his company.
>I was wondering what software exists (commercial and PD) and what
>platforms are recommended.
>
> Thanks
>
>Joe Landman
>
>Landman@splat.physics.wayne.edu
>landman@hal.physics.wayne.edu
>
There are quite a few commercial FEA packages available, but I've never 
heard of a PD one.  Maybe because of the effort involved to develep any non-
trivial FEA code is too great.  To chose a suitable package, you may need to 
specify the purpose of the analysis.  A package suitable for stress analysis 
may not be suitable for metal forming at all.  There are so-called "general 
packages", such as LUSAS from FEA UK limited.  PAFEC and ABAQUS are other 
examples.

The choice of hardware depends both on the availability of the suitable 
software for that platform and the nature of the analysis.  For elastic 
stress analysis, or heat conduction problems, a 486 PC would generally be 
sufficient, but if it involves large non-linear deformations I would 
recommend more powerful workstations, such as Apollos.

Hope it helps.

Weiping Hu

In article <1992Dec15.225304.11571@porthos.cc.bellcore.com>, bhatia@bigcat.bellcore.com (Raman Bhatia) writes:
|> In article <1992Dec14.181725@roper.mc.ti.com>, a722756@roper.mc.ti.com (W. Donald Rolph) writes:
|> |> 
|> |> This is by no means a trivial question.  You must first ask the class of FEA to
|> .......
|> |>  
|> |> It is also imperative that you have  a good mesh generator and graphical post
|> |> processor.  The difference between a good mesh generator and an ok mesh generator
|> |> is about 20X in productivity.
|> ........
|> |> 
|> |> A typical system ight cost the following (rough costs):
|> |>   
|> |>      1)  FEA program $5K - $10K year
|> |> 
|> |>      2)  Mesh generator $10K yr
|> |> 
|> |>      3)  Workstation:  $25K
|> |>  
|> |>      4)  color postscript printer $5K to $10K
|> |> 
|> |> This is nearly a 1/3 cost reduction and 10,000 performance improvement over the
|> |> system I put together is 1983!
|> |> 
|> |> 
|> |> -- 
|> |> 
|> |> Regards.
|> |>  
|> |> Don Rolph  a722756@pan.mc.ti.com WD3 MS10-13 (508)-699-1263
|> 
|> Thanks for a very informative post. Here's another question. What are
|> the latest in Mesh Generation techniques? 
|> 
|> The last time I dabbed in FEM, the finite-octree (w Delaunay triangulation) 
|> appeared the most promising but there were some concerns about robustness. 
|> I know that companies like PDA engineering (developers of PATRAN) were 
|> looking at "paving". What is the current status of these two techniques, 
|> and are there any new ones out there (AI approaches etc.)?
|> 
|> Raman Bhatia (bhatia@bigcat.bellcore.com)

I have an interesting proof that show that the triangularization schemes all
violate controllability theory at any given point in the model.  The effect is
that in 3D, with typcial dimensions in orthoganoal directions which have
significantly different scales (1.4 in thick by 5 inches wide by 20 inches long),
the triangularizations schemes produce highly unsatisfactory meshes for many
classes of problems, particularly nonlinear problems.  It is interesting to note
the CFd codes are gradually adopting Laplacian mesh generation schemes, schemes
which were popular as early as 1960 for fea.  It should be noted that CFD can be
very suseptible to poor meshes, and perhaps there is a message here!  

I have simplistic proof which suggests that Laplacian mapping produces the
"optimal mesh" fgor certain classes of problems, based on the givven mesh density
provided as input.  I also have s ample program which generates hexahedral meshes
off an arbitray boundary rep trimmed nurb solid model, demonstrating that the
technogy is feasible, and the input is not overly burdensome.
-- 

Regards.
 
Don Rolph  a722756@pan.mc.ti.com WD3 MS10-13 (508)-699-1263

In article <1992Dec14.212505.27326@cs.wayne.edu> landman@splat.physics.wayne.edu writes:
>       I have a question about FEA, specifically software and hardware
>   platforms.  A friend is looking into the possibility of purchasing 
>   hardware and software to perform FEA calculations for his company.
>   I was wondering what software exists (commercial and PD) and what
>   platforms are recommended.

I have recently been asking the same question. Two avenues which may 
be worth following up are:

The NAG finite element library (FE03). This is a collection of 
subroutines from which one may assemble an FE program. Well-annotated 
(FORTRAN) source code is included. In addition there is an applications 
library (with source code) which uses the basic routines. The idea is 
solve ones own FE problem by suitably altering the template of the 
application programs. As a bonus one learns a lot about finite element
programming. The manuals which come with the library are excellent and
constitute a good introduction to finite elements. The library is quite 
cheap by commercial standards (about US$500). FAX numbers for UK, US 
and Germany are +44 865 310139, +1 708 9712706, +49 89 3207396.

A second source of FE routines is the book by Zienkiewicz and Taylor
_The Finite Element Method, vol 1_ (McGraw-Hill 1989). This book contains
an extensive chapter on FE programming, and incudes the source code to a 
suite of programs called PCFEAP. The code can be compiled to run on PC's
or on Unix machines. The FORTRAN 66 is of the spaghetti variety, but I
have hopes of understanding it one day!

Both these programs naturally lack the sophisticated front and rear ends
of the commercial packages. On the other hand, if one has a non-standard 
application then this may be the only way to go. It requires some effort
but is worthwhile in the end!

Roger Young