John F. Sowa wrote:
> The following point near the end of your note is critical:
>
> EB> Being careful, 'tolerance' and 'uncertainty' are entirely
> > different things, as I mentioned in a previous email. Tolerance
> > is the allowable variance from the specified quantity, where that
> > difference allows the measured object to meet the functional and
> > physical requirements; uncertainty is the amount by which the
> > measurement may be different from the actual magnitude of the
> > quantity. In complying with a tolerance, one has to ensure that
> > the worst case measurement error gives a value that is still
> > within the tolerance.
>
> This distinction is related to the difference between science
> and engineering. Uncertainty is inevitable in both. Tolerance
> is an engineering constraint on the permissible variance in the
> construction of some artifact.
> (01)
Tolerance is an external constraint. It is an "engineering constraint"
if you think that term applies to a legal requirement on food packaging
that the actual weight must differ from the stated weight by less than 1%. (02)
> EB> The problem is that the scientific usage and the commercial
> > usage are significantly different in this area.
>
> It depends entirely on what commercial application your talking
> about. Does your company make meat balls or ball bearings?
> Many commercial applications "push the envelope" to the extremes
> of science, and others can be done with stone-age technology.
> (03)
I agree. Most commercial usage, however, is not very scientific. Of
course, there are some highly scientific commercial applications, and
yes, they are very sensitive to measurement issues. (04)
> EB> I would prefer that the units of measure ontology not go deeply
> > into 'measurement' at all. My point was that 'measurement' is
> > the meat of the VIM, and that affects its way of modeling 'quantity'.
>
> If you're making meat balls, you're not going to use high-precision
> instruments. But if you're making silicon chips, all the issues
> of measurement are critical. Both applications are commercial.
>
> Even if you restrict the range of applications to a single domain,
> such as medicine, you have to accommodate everything from a nurse
> measuring a patient's height to high-precision instruments for
> microsurgery.
> (05)
Yes. So the ontology must be correct with respect to arbitrary
applications, which means we have to understand "tolerance". The
question is: how deeply do we go into "measurement" and "uncertainty"?
Measurement science is not about the requirements; it is about the
nature and quality of the measurements themselves. And that gives rise
to an idea like "no two chairs have the same height" because we are
talking about two different measurement actions. Where a commercial
application says "these chairs are of equal height", the measurement
scientist says: "the height measurements of the two chairs, using
procedure P, differ by less than X, inclusive of uncertainties." And an
ontology that insists on the latter phraseology is going to be less
useful to most applications. (06)
Put another way, let M be some measurement of a physical thing and let
t1 and t2 be distinct things in the domain of M. Is (= (M t1) (M t2))
a meaningful proposition? Does it evaluate to true or false?
Similarly, is (= (M t2) "5 m") a meaningful proposition? (07)
This is where the ontological rubber meets the "commercial" road. The
measurement scientist will be very concerned about how you define M, or
how you define "=". If M is the "nominal value of the measurement",
then it is a 'quantity value' stated to some "useful accuracy" (which is
context dependent), which is half the information the measurement
scientist needs. In that case, "=" is a number/unit comparison between
quantity values or "magnitudes", but it says very little about the
comparison of the measurements. (08)
A common scientific and engineering practice is to state a measured
value as a "scientific number", that is, to the least decimal place of
uncertainty. If that is what is being stated, then you have a
well-defined measurement value, and you can talk about the uncertainty
of the difference, which is not at all the same thing as "=". (09)
So if our ontology is going to be correct at both ends of this spectrum,
we need to be really careful about the terminology and the definitions,
and we will definitely be breaking some eggs. The alternative, as John
has proposed, is to develop two or more "microtheories" for different
"usage patterns". (010)
> EB> We use the value "180cm" in practice, as if there were a
> > 'quantity magnitude' to which it refers. It is not clear to me
> > that the VIM agrees that there is one.
>
> Ordinary language, even among engineers and scientists, typically
> omits any details that are not in the immediate focus of attention.
> But that doesn't mean that the speakers are ignoring the details.
>
> If and when the details are significant, the engineers are certainly
> aware of them. (011)
John has obviously led a charmed life in working with engineers. In my
experience, one engineer in three is aware of all the scientific details
that are significant to what he is doing. In the particular case of
measurements, design engineers are aware of tolerances, but presumptive
about the accuracy of the manufacturing measurements, which means they
often design things that cannot be built, and specify tolerances that
cannot be reliably achieved. Manufacturing process engineers are aware
of uncertainty at a certain level -- they understand the expected
accuracy of a specific measurement procedure with a specific
instrument. Some are aware of the environmental influences on
measurement behaviors, and most have problems with uncertainty propagation. (012)
Yes, there are engineers who have had to learn measurement science in
detail and apply it to their fabrication and test facilities, and these
guys are extremely valuable in precision manufacturing applications.
(And as the world economic picture changes, precision manufacturing has
become a much larger percentage of G8 manufacturing.) And yes, we want
to support these applications. (How) can we do it without alienating
the engineers with less demanding applications (and less knowledge and
awareness)? (013)
> The VIM document, for example, was largely written
> by and for engineers. (014)
I would say rather that it was written by measurement scientists (if the
NIST participants are any indication), partly in order to educate
engineers, among others. The major objective was to standardize
terminology in the field of measurements, for scientists, engineers,
regulators, technical writers, etc. The meaning of a term in a
measurement specification or a measurement report must be understood the
same by both the author and all readers for many scientific,
manufacturing, and trade applications. And the VIM intentionally
quashes the notion that any measurement is exact -- every specification
has tolerance; every report has uncertainty. (015)
> To return to a theme I've discussed many times before:
>
> 1. Even for a particular field or industry, there is a wide
> range of applications that treat the same subjects and topics
> at many different levels of detail and precision.
>
> 2. The details necessary at the most precise level are usually
> omitted in broader discussions or contexts.
>
> 3. Therefore, there is no such thing as a one-size-fits-all
> ontology, even for a particular field or enterprise.
>
> 4. But all the people in the field or enterprise have to
> talk to one another, at least at some broad level.
>
> 5. Therefore, the upper levels of the ontology will resemble
> terminologies with very few detailed axioms. But there
> will have to be families of microtheories that axiomatize
> those details for each of the tasks and applications.
>
OK. We agree on this. Now, what exactly does this mean with respect
the units of measure ontology? (016)
At the highest level, we have David Leal's draft VIM model, and the
DOLCE model, and some others we haven't explored yet. Are the concepts
and definitions generally applicable? Should some of the definitions be
changed to make them generally applicable (as distinct from inconsistent
with some detailed theories)? Which ones? Are the definitions
axiomatizable? Which axioms should be omitted at the highest level? (017)
I'm still uncertain that the VIM view of the quantity space is
consistent with the usage by pilots and carpenters and cooks, as well as
energy regulators and semi-conductor process engineers, and I don't know
whether we should care. Once we agree that specifications and
measurements are not exact, what does a 'quantity value' mean? And what
is a "magnitude"? (Which takes us back to the question of whether the
'quantity kind' "length" refers to a subclass of 'particular quantity'
or to a subclass of 'quantity magnitude'.) (018)
-Ed (019)
--
Edward J. Barkmeyer Email: edbark@xxxxxxxx
National Institute of Standards & Technology
Manufacturing Systems Integration Division
100 Bureau Drive, Stop 8263 Tel: +1 301-975-3528
Gaithersburg, MD 20899-8263 FAX: +1 301-975-4694 (020)
"The opinions expressed above do not reflect consensus of NIST,
and have not been reviewed by any Government authority." (021)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/uom-ontology-std/
Subscribe: mailto:uom-ontology-std-join@xxxxxxxxxxxxxxxx
Config/Unsubscribe: http://ontolog.cim3.net/mailman/listinfo/uom-ontology-std/
Shared Files: http://ontolog.cim3.net/file/work/UoM/
Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?UoM_Ontology_Standard (022)
|