### Manufacturing Cell Design in a Multi-Criterion Environment,

Abstract

Manufacturing Cell Design problem has been a research area in the
cellular manufacturing context since 1970s. Prior to the 1990s,
the problem had been examined basically with respect to only a
single criterion, e.g.: minimizing intercellular parts movements,
maximizing similarity among the parts and/or machines in the cells,
minimizing imbalance of the work load among the cells. In the 1990s
and after, a number of papers have been published that model the
cell design problem as a Multi-Criterion Decision Making (MCDM)
problem, in particular as a Multiobjective Optimization Problem (MOP).
In this thesis a new multiobjective model has been developed to
deal with the exceptional elements in the design of a Cellular
Manufacturing System. The set of objectives include: minimizing
intercellular part movements, minimizing sum of machine duplication
and part subcontracting cost, minimizing machinery underutilization
and minimizing imbalance of the cells workloads. Due to the conflicts
among the objectives, attaining an ideal solution, i.e. a solution
that simultaneously optimizes all of the objectives, is impossible.
However a set of non-dominated (or Pareto Optimal) solutions could be
sought from which the decision maker will be able to select based on
his or her priorities. To this end, a multiobjective genetic algorithm
called XGA was developed and coded in Visual C++. The algorithm makes
use of the non-dominated sorting idea to rank individuals in the
population. An extension of Reminder Stochastic Sampling Without
Replacement (RSSWP) is developed wherein a novel probabilistic scheme
of elitism is applied. Niching is employed to keep diversity among
the individuals in the elite set. Stopping criteria of the algorithm
is devised so that takes into account convergence of the algorithm
to the Pareto-Optimal frontier along with the maximum number of
generations.
A number of cell design problems taken from the literature were
solved by the XGA and also by three reference algorithms, namely: VEGA,
NPGA and NSGA. The results obtained by the XGA show promising
improvements in three dimensions, i.e. quality, diversity and CPU
time compared with the reference algorithms.