Page 1 of 1

Operations optimisation in a batch job shop

Posted: Thu Aug 29, 2002 1:28 am
by "OShea, Laurence"
Hello,

I am currently performing a simulation analysis with regard to operations
research and manufacturing optimisation. The subject company in the study
manufactures and screens high reliability discrete semiconductors to
military standards.

Processing within the company can be described as a batch job-shop with
batch sizes ranging from 500 to 15K devices. Operation re-entry is very
prevalent and relatively high levels of rework also add to the sometimes
unpredictable processing times and steps. I have modeled the company from
initial parts receiving through to shipping. Due to the processing methods,
i.e. batch job-shop, planners and supervisors have great autonomy when it
comes to sequencing work orders through the work floor. I believe, from
examining some System Dynamics studies and literature, that this method may
be very beneficial, and possibly necessary, in achieving a very accurate and
representative model of the company.

I am writing this email in the hope that you may be able to give me some
further information, or point me in a direction where this area, or similar,
has been examined using System Dynamics methodologies. I would appreciate
any comments or suggestions that may help me determine the correct approach
to resolve and represent this situation using this method.

Thanking you in advance for your help.

Yours sincerely,
Lar OShea
From: "OShea, Laurence" <loshea@microsemi.com>

Operations optimisation in a batch job shop

Posted: Sat Aug 31, 2002 6:02 pm
by "Ray on EV1"
It looks like a key in understanding comes from the statement:
" processing methods, i.e. batch job-shop, planners and supervisors have
great autonomy".

A particular perspective would be to address the issue of local vs. global
optimization. Great autonomy might suggest each cell addresses an
objective function with a local optimum view. The problem space could then
be defined as the span from this collection of local views to an overall,
global optimization.

This could be addressed as combinatorial optimization problem - which cells
should share objective functions and to what degree of overlap. For
example, cells in adjacent work-piece sequence slots may share objectives to
assure some degree of cooperation. Similarly, cells sharing common
resources may benefit from such cooperation.

One of the concerns here is the optimization computational costs increasing
geometrically as the number of cells share objectives - the dimensions of
the optimization space increases. Not only do the computational cost go up,
but the modelers comprehension of the complexity of the local minima
placements become taxed.

This type of issue has been studied for decades under the subjects of
centralized / distributed control. There are many systems where it can be
shown that the original optimization can be achieved equally well with
either centralized or distributed control. But the political environment
will drive the choice to one or the other.

My favorite example is vehicular traffic control in a city. Centralized
control can help coordinate the use of information from different parts of
the city to optimize each vehicles trajectory from a centralized
controller. The same results can be obtained by providing that information
to each vehicle and letting the vehicle make all the control decisions.

Raymond T. Joseph, PE
rtjoseph@ev1.net
Aarden Control Engineering and Science