ABSTRACT

This thesis addresses the  problem of computing the optimal parameters
for production  control policies in  the glass  manufacturing industry
and provides a framework of analysis related with the structure of the
production  policies. We  consider a  multi-product,  multi-stage, and
capacitated  discrete-time  production-inventory  system  with  random
yield. Random demand occurs in each period. The optimal parameters for
a given production control  policy are determined in order to minimize
the expected  costs or reach  a given  service level. Three  different
production  strategies  are discussed:  Make-to-Order (MTO),  Make-to-
Stock (MTS), and Delayed-Differentiation (DD).

We use real data (processing times, random yield factors, etc.) from a
glass  manufacturing  company,   providing  simultaneously  the  model
validation  and  the evaluation  of  the  relative  performance of the
different strategies.

The approach  used to analyze  this problem  will be  simulation based
optimization. Simulation will be used as a tool to obtain estimates of
the  objective  function  value  and  gradient  with  respect  to  the
parameters that describe  the control policy. The  gradient estimation
is based on Infinitesimal Perturbation Analysis.


KEYWORDS: Capacitated Inventory Systems; Alternative
          Production Strategies; Random Yield; Glass
          Manufacturing;  Infinitesimal Perturbation
          Analysis.