Mensurix

[Stephen Birkett]

Bernhard wrote:
>You are right, since it is a commercial program, 
>i don´t publish exactly what i am doing, but i 
>hope the information i gave here is at least 
>more than nothing.

Well, actually, what you've given doesn't do 
anything to describe the model which is the basis 
of your software. You've merely listed some 
mathematical devices used in generating equations 
and solving them. There seems to be some 
confusion here between model concepts (physical 
representation), their mathematical 
representation, solution methods, and software 
development, which are all different and can be 
described independently. A lot more of these 
details need to be revealed before a model can be 
considered a serious design tool, as opposed to 
something akin to a "sim-city" computer game. 
This is not say it is the latter, just that 
without the information we're in the dark. The 
onus is on the seller of a commercial design tool 
to convince the customer, and that cannot be done 
with smoke and mirrors.

>The model is derived by higher order 
>differentials, solved with usual difference 
>schemes. All what is necessary to build such a 
>model can be found in Klaus Jürgen Bathe´s book 
>"Finite Element Methods".

FEM is simply a generic technique for continuous 
systems. Giving that as the basis for a model is 
like saying you've used Newton's equations of 
motion to solve a dynamics problem, or Snell's 
Law to solve an optics problem. Not very 
informative. In order to communicate what's 
actually been done in the model you have to 
explain how these techniques have been used to 
represent the physics, what assumptions have been 
made, how the parameters have been derived, how 
the results are tested, and so on.

Actually, how long does the model take to do one 
run? If FEM is used as the computational basis it 
must take quite a while to analyse even one 
scenario, unless the software model just picks 
from a library of results of a few days worth of 
pre-generated FEM simulations. FEM models run 
real slow if they are complex enough to be 
realistic.

>Hysteresis is not a phenomene discovered by 
>Stulov, it is relatively common used in finite 
>element models.

Of course hysteresis wasn't discovered by Stulov. 
It's a fundamental physical phenomenon. And it 
has nothing to do with FEM or any other modelling 
technique. Stulov was (one of) the first to 
include hysteresis in a hammer model.

>But have you ever seen Yamaha open their black 
>boxes when using physical modeling in their 
>synthesizers or Bill Gates giving the source 
>code for his Windows Kernel?

This analogy is not valid. The success of 
Yamaha's physical modelling can be determined by 
the end result. Does the customer like the 
synthesized tones? That's what they're buying. 
They could care less how those tones are 
generated in order to assess the value of the 
product.  Ditto source code. Either software 
works properly and efficiently or it doesn't. 
Again the results can be used directly for 
assessment. In the case of an applied model that 
is supposed to predict the real behaviour of a 
physical system customer assessment must be based 
on a validation process for which transparency is 
essential. I'm not convinced that you've been 
through this lengthy process, or else you would 
not hestitate to reveal the results and use them 
in the advertising, even if you choose for 
whatever reason not to publish the model details 
themselves for critical scientific assessment by 
experts. Anything less than that and the model is 
essentially useless and its commercial 
distribution is simply a  software marketing 
exercise. The $ value/worth of a predictive model 
is derived from confidence in the predictions and 
how they are derived. This can be explained 
without revealing source code or exact details of 
every equation. There is always room to conceal 
some of the critical information that's needed to 
reproduce a model, thereby protecting commercial 
interests. [This I know from long experience with 
commerical models.]

As I said last time, if the model is really 
capable of accurate predictions of the factors 
claimed, then it is new and valuable. The details 
behind it would be interesting and should be 
published, along with the results of your testing.

Stephen
-- 
Dr Stephen Birkett, Associate Professor
Department of Systems Design Engineering
University of Waterloo, Waterloo ON Canada N2L 3G1
Director, Waterloo Piano Systems Group
Associate Member, Piano Technician's Guild

E3 Room 3158
tel: 519-888-4567 Ext. 3792
fax: 519-746-4791
PianoTech Lab Room E3-3160 Ext. 7115
mailto: sbirkett[at]real.uwaterloo.ca
http://real.uwaterloo.ca/~sbirkett