ISSS 1998 Conference, July 23, 1998 -- (Rough) Transcriptions

Numerical simulation approaches for modeling industrial systems of matter, energy and information [Reid Bailey, ISSS 1998 Paper Session, July 23/98]

These notes are a rough transcription, prepared as each individual presenter and/or commentator spoke at the ISSS 1998 conference. Gaps and errors have likely occurred. For more accurate citations, please consult the original presenters. These notes have been contributed to the ISSS by David Ing, of the IBM Advanced Business Institute (sabi@systemicbusiness.org).

[Paper session, July 22/98, 2:35 p.m.]

Reid Bailey, Georgia Tech

Topics:

Motivation
Story
Types of models
Selecting the right model type.

Why model?

Want to replace a real system with a representation of that system (so that you can learn about it)
Why not interact with a real system? Cost, difficult, destructive, or too many players.
Types of representations (tree from Gordon, 1978):

Not interested in physical models.
Math models are either static or dynamic.
Analytical models have closed-form solutions.
Thus, math and dynamic.

Story:

A company trying to work within an industrial ecosystem: how do they work as a larger system?

Would like to do what-if, before committing for real.
Need to fit a tool to the problem.

The tools (beyond the initiation phase):

Discrete continuous models

Instantaneous completion of event, e.g. hot dog vendor sells hot dog, or manufacturing line.
Time skips over events.

Deterministic stochastic (uncertainty) models

Based on differential equations, which approximation through independent variables.
e.g. draining of a lake.

Hybrid / Mixture: both elements in the same model.

Comparison of discrete / continuous / mixed.

Incoming parts, wait, paint, and then into another bin.
Modeled as discrete shows oscillation around a trend, whereas a continuous model is a straight line, hybrid is joined between levels.

Deterministic:

For giving inputs, outputs always the same.

Uncertain:

For given inputs, behaviour is not fully determined.
Stochastic isn't quite the same.

		represent numbers with
Fuzzy set	more qualitative than quant	membership function
Bayesian statistics	only an estimated distribution can be determined	estimated distribution
Stochastic	exact distribution known	exact distribution
Interval arithmetic		upper and lower bound

Select model type:

Weigh off ...

purpose of the model (much more important, and unfortunately subjective) vs.
characteristics of the system being modeled.

Is the purpose in the scope of the model, and what accuracy is required?

	Discrete	Mixed	Continuous
Deterministic	microscopic, discrete, low accuracy	micro, mixed low	micro, continuous, low OR macro, all, low
Uncertainty	micro, discrete, high	micro, mixed, high	micro, continuous, high OR

Conclusions

Fill the toolbox first, then select tools
Purpose is paramount
Difference between discrete and continuous is small, with end behaviours the same.
More to do in the future ...
Can also model with other methods.

Questions

Define industrial ecosystem

Looking at wholistic approaches to enterprises, where not connecting outflow from one industry into another.

Examples:

Have been looking at continuous, deterministic working on case studies of a power plant, heating system for a town.

Navigate ...
... to the 1998 ISSS Conference (Rough) Transcription Main Page
... to the ISSS home page at www.isss.org.