novacast - Stanford University

advertisement
■f
NOVACAST
Specialists in Foundry Technology
Industrial Computer Software
Expert Systems
Prof. Edward Feigeinbaum
Ronneby 870728
Dep. Computer Science
Stanford University
CA 94305
Dear Prof. Feigenbaum,
!_ pleasure to meet ou in London together with
It
the n?
"Inductive interest group" meeting.
W
D.Michie at
I promised to send you some information
about successful expert
systems that we or our clients have
developed
here
in
My company NOVACAST is a consultant company,
established 1979
Our speciality is metallurgy and foundry
technology. Our first
"expert system" was a
program
called
"Novacast
Foundry
Technology" that we started to market 1980. That program
was
written in basic and contained knowledge about the
foundry
process to be used by methods engineers
when selecting casting
processes. At that time we were not
aware of the "expert system"
concept
the program was thus designed from zero and contained
our knowledge represented as a mixture of normal calculations
and
then rules. The program was the first of its kind and was
well recieved. Since then the program have been improved
and
extended several times and is nowadays mainly delivered
for the
IBM PCrs and is in use at more than 200 foundries
worldwide.
The foundry process in very complex with more than
2000 variables
affecting the end result. A lot of the knowledge
is still
scientifically unknown and normally a lot of trial and error
is
used. In the knowledge acquisition phase we have
been able to
reveal interactions and to make practical knowledge explicit by
using inductive methods. Thus in our own systems
we have mainly
used the inductive "example learning method" as
a problem solver
and hypotesis tester.
We have made several other system for
foundries such as:
-
SANDMASTER
plants in
-
an expert
foundries.
system
for process control of green sand
-
NOVASHOT
an expert system for gating design in high pressure
diecasting of aluminum and sink alloys.
Recently we have launched a system for analysis
of casting
defect. The system is build with SuperExpert Professional
which
is a new ITL product where we
have been involved in the
development. The Casting Defect Analyser finds the primary
cause
for the defect and suggests a remedy.
In 1986 I had the
opportunuty to present a paper about our expert systems
for
foundries at the annual conference of American
Foundrymen
Association 1986 in Milwauki.
We feel that our industrial experience
and general knowledge
problem solving" have been very
about
valuable when developing
expert systems. Thus we have in the last two years expanded
our
activities to also include consultancy for building expert
systems for clients in other sectors.
NOVACAST AB
FRIDHEMSVAGENB " SOFT CENTER
S-372 00 RONNEBY
■
SWEDEN
TEL: 0457-718 00
BG: 544-9855
■
TLX: 43583
Novacast
Page 2
toother system that is in use is
our
advice or, how to send letters and packages "MAILWHIZ" which gives
by
uses all the possibilities and "crepholes" mail. The system
in the Swedish Ma__
Regulations in order to find
the cheapest way depend,"
?h_
and alS Sh WS *at
We
at miLmzz is ca
P^le of reducing our m__._A.ats
LT%
"f
It TL
o!
docLentHre
needed
° °
alternatives if _
F COTSiStentl
* *-«-.
was orginally
made in Expert-Ease but
0^
system
smart way
"smart
wav The
al_
is
now upgraded to SuperExpert and is beeing
marketed.
mSL^'V^?^"" * "
methods
-i-rinductSn. _nnL_tive
P
t
we have been able to reveleal rules that
where not known
th
haVe 3 V6ry hi
Predict!"
STOCKMASTER has been in daily use since
January 1987
Traversum
albSenS bSen b Ught by two Swedishtonks. at TraVerSUm
6
lnCreased from
60 to around 90 % for one
of the
STO^TThasl^
/\
Tt _tft" T °
users!
°
Enclosed please find a paper about
documentation about our products.
are
We
__c«_^
*
STOCKMASTER and some other
otner
you probably know, the representative
for ITL in
We have been using the
ITL
inductive
tools for our own needs since 1984.
During that time we have
the opportunity to tackle a variety
of problems, which has given
in
our
.
Our customers represent a wide variety
of
as
Scandinavia and Switzerland.
"stomerr"^
hid
*"*
supporting'
"* "*
branches
U
B_nk
manufacturing
_ha_ma_r aCar' m Stef ""V*"*"**.
Urln9 MrCraft«*^"
act-ing.
SstitutrshlpySdsHtc
"
'
Our
Credit
experience is that
inductive shells like Exnert- *■»__
SuperExpert and Extran can be regarded
not only
rf l >^l-olvers"
P
ncra rs for
or knowledge acquisition. Inductive
shells are
8
t0 kn WledgS
1
tradition.!
"r.lHenStols"
t£i_
*° f
"r.le
blsed^ystet ?
T «_Srt
31.
„" °"
°
"
-» "
"*
Several companies are using the induction
mechanism in solvi"
complex problems where traditional
mathematical
or
methods have failed. A typical case is when
a lot of test data is
available, some of it as logical attributes
and where the
of relations vary within the problem which
might also be
statistical
compoled
"
_
Z_
?°?J?"
6XP6rt
add his knowledge
interactively
hUS flnd the mles The
V
inductive
shell will
in that case act as an "intelligence
amplifier"
Jn?w
finding and proving hypotheses.
-
°
Ca
and
assist
assist
in
in
Novacast
Page 3
Here are some examples:
1. A manufacturer of high alloyed
steel tubes had a problem with
an application where a customer experienced problems
when welding
tubes for heat exchangers. The problem was that gas
blows and
cracks sometimes occured in the welds. A lot of test data
was
available and about 10 factors were considered as being
of
interest. Among these were Voltage, Amps, Type
of electrode
Welding speed. By entering examples using class values
OK, blow
and crack in Expert-Ease the conditions for
defect-free welds was
revealed by analysing the induced rule. It was found that only 3
factors had to be kept within certain limits in
order to obtain
good welds.
2. A research institute has made a system using
Expert-Ease that
can predict the liquidus and solidus temperatures for
steel
alloys
The system is used when trying to develop new alloys and
allows
engineers
the
to simulate the effect of various
compositions. It replaces to a large
extent test melts and
thermal analysis.
The system was developed by entering values for alloys
with known
compositions
and
their
resulting
liquidus
and
solidus
.
temperatures
3. A
.
system
institute to
tool steels.
has
also
been
developed by
the
same research
optimize chemical analysis and heat
treatment for
4. Optimal parameters for laser hardening
of tool steels have
been identified by using Expert-Ease. A system has been developed
where the settings for the laser are advized depending
on type of
alloy, hardening depth ,etc.
The engineers who developed the system had started
using
traditional methods but without success. The final
solution
revealed by Expert-Ease was astonishingly simple and to
the
engineers an "aha! "-event. The
rule was converted to a "laser
hardening diagram" and is now used daily.
5. An advisory system has been built by a tool steel
manufacturer
to select the optimal type of tool steel in each particular
case.
The expert system asks the user questions such as "Maximum
operating temperature?", "Number of cycles per hour?", "Type of
plastic to be injected?" etc. and can
select from among the 15
different alloys.
6. A margarine manufacturer has made
a system that predicts the
percentage of solids in margarine mixtures at
4 different
temperatures. The system considers the effect of 15 components.
This is a very complex problem as all the components are
soluble
in each other and as the resulting solid phase
is not possible to
predict using normal math methods.
(An analogy in metallurgy
would be to construct a phase diagram with 15 components !
)
By being able to predict the resulting solid phase
it is now
possible to in advance add "modifiers" and
mixing
.
thus save
time
Novacast
Page 4
7.
A manufacturer of organic lubricant oils has used Expert-Ease
is a complex
task where no successful methods have been published. In this
case a useful approach has been developed by using Expert-Ease
and analysing the induced rules. Three different systems have
been developed, namely:
as a research tool in formulating emulsions. This
a. Investigation of the influence of 7 important variables
on the stability of emulsions.
b. Identifying optimal circumstances and selecting emulgators.
From viewing the rule it has been possible to draw "maps"
showing stable areas for the emulsions.
c. A system has been developed where the performance of
various emulgators can be studied.
This company has also used Expert-Ease in order to extract
knowledge from published articles. It has for instance been
possible to extract "hidden" knowledge from experimental
tests
published in the traditional "row and column" format and to
make
it easily understandable.
8. A cancer researcher has used Expert-Ease in order to
evaluate
the prognostic value of various tests. Using a complete set of
data from 75 patients it was revealed that the same prognostic
results as with the traditional expensive tests could be achieved
by checking easily observed factors such as type of tumor, size
and location.
A system has also been developed to find the optimal
of chemo therapy and radiation treatment.
combination
9.
A company that manufactures ventilation and filter equipment
has built a system that acts as a "risk analyser" for new
projects. Ther object is to trace "problems" before even quoting
a new project which often runs about 10 to 30 million
Pounds so
that correct allowance in both time and money can be made from
the start to avoid economical losses and unfulfilled time plans
etc.
10. A Swedish bank has made an investment advisory system and a
system for credit application and policy. They are also working
on a "legal advisor".
-
11. Our representative in Finland - Wasamation AB
have been
using Extran and in cooperation with doctors at ABO university
been able to find combinations of symptoms that have a very high
prognostic value for so called "sudden death of newly born
babies". The results are soon to be published in a scientific
paper. This we believe is a good example as statistical methods
did not reveal any significant correlation but the induction did.
We have found this to be true in many cases where
several
descrete function are overlapping in a problem and where the
limits are unknown. Thus induction can be used as a "hypotesis
finder". Once the rules and the limits are known traditional
methods can be used for further refinement.
(■
Novacast
Page 5
r
Several other systems are in use in Scandinavia. We
estimate that
around 50 systems of practical value have been developed on this
market. We have just recently started an application club and we
hope to be able to catalogue the various applications
in more
detail
.
Please let me know if you need any further information.
Sincerely.
Rudolf Val. Sillen
President Novacast AB
cc: Prof. Donald Michie
- ITL
Encl. Various papers : Stockmaster, Building
PC-housted
systems, Novacast broschure, Print Screens from Casting
expert
Defect.
Download