Gerhard Werner enquired about reverse modeling. Apparently his query =
relates to a complex manufacturing system. I would like to offer two =
possible solutions:
a) The first proposed solution might be rather simplistic, but then, the =
most effective solutions are often quite simplistic:
Most manufacturing systems are devised along an "ideal way" (compare =
maximum capacity according to the manufacturer versus reallife =
experience). One way of identifying variances would be to run two =
models, one reflecting the "ideal" system, and the other reflecting the =
real system. A simple program on a spreadsheet would then identify then =
variances above a predetermined tolerance level, thus pointing to the =
problem areas.
b) The second proposal is to model the system in terms of Professor =
Goldratts Theory of Constraints. Resource utilization can be modeled as =
stocks, or, alternatively, (more in line with Professor Goldratts =
theory), inventory should be modeled as stocks. Using Theory of =
Constraint principles, the bottlenecks can then be identified.
Regards,
Schalk Jacobs
From: "Schalk Jacobs" <busexcel@lantic.co.za>
ReverseModelling

 Junior Member
 Posts: 14
 Joined: Fri Mar 29, 2002 3:39 am
ReverseModelling
We gave some thought to the reverse modeling problem about an year ago  though
we have done done much with our "insights".
A function has an inverse only if it is onetoone (and is injective (?) but that is not important
for our models) . In a similar vein imagine a simulation run to be a transition through a state
space. You can reverse this run thorough the statespace unambiguously only if the forward
state space map is onetoone. Then flipping the direction of state transitions will give you a
reverse simulation. If this condition is not met, an alternative would be to do some sort
of likelihood estmation using multipple forward directional runs and estimate the reverse
likelihoods from them. For a given set of input conditions in the reverse run (which are the
output conditions in the forward run) you get multiple output conditions (which are the input
conditions in the forward run) with some likelihood estimates. Which output condition you end
up with in a given reverse run of course depends on how you resolve the manytoone
state transitions of the forward model.
From: M V Nagendra Prasad <nagendra@cstar.ac.com>
we have done done much with our "insights".
A function has an inverse only if it is onetoone (and is injective (?) but that is not important
for our models) . In a similar vein imagine a simulation run to be a transition through a state
space. You can reverse this run thorough the statespace unambiguously only if the forward
state space map is onetoone. Then flipping the direction of state transitions will give you a
reverse simulation. If this condition is not met, an alternative would be to do some sort
of likelihood estmation using multipple forward directional runs and estimate the reverse
likelihoods from them. For a given set of input conditions in the reverse run (which are the
output conditions in the forward run) you get multiple output conditions (which are the input
conditions in the forward run) with some likelihood estimates. Which output condition you end
up with in a given reverse run of course depends on how you resolve the manytoone
state transitions of the forward model.
From: M V Nagendra Prasad <nagendra@cstar.ac.com>