Research has been carried out to investigate the use of genetic algorithms (GAs) as a common solution technique for solving the range of problems that arise when designing and planning manufacturing operations. A variety of problem areas have been selected that are representative of the range of problem types found in manufacturing decision-making, i.e. assortment planning, aggregate planning, lot sizing within material requirements planning environments, line balancing and facilities layout. Part 1 of this paper reported how typical solutions for each problem area were coded in terms of a genetic algorithm structure and how suitable objective functions were constructed. In addition, comparisons of performance were carried out between GA solution methods and traditional solution methods. Part 2 of this paper now describes the GA experiments undertaken during the identi®cation of suitable GA operators and operator parameter values. These experiments have enabled underlying relationships between problem characteristics and performance of individual operator types and parameter values to be identi®ed. From this work a set of guidelines has been identi®ed for selecting appropriate genetic algorithm structures for speci®c types of operations management decision area.Keywords: genetic algorithms, manufacturing system design, assortment planning, aggregate planning, material requirements planning, assembly line balancing, facilities layout
ASSORTMENT PROBLEMIn Part 1 of this paper the assortment problem was coded into a format suitable for a genetic algorithm (GA) solution string using a single binary digit to represent each product speci®cation considered as a standard [1]. The decision the GA then had to make was simply to determine whether or not to include a particular standard in the product range.In order to determine the applicability of individual GA operators for providing solutions to the assortment problem, a range of experiments was performed as follows:1. The alternative selection operators investigated were roulette wheel, roulette wheel with elitism, tournament and truncated. The e ciency of each selection operator was identi®ed using experiments in which other operator types and their values remained constant: the number of replications was set at 10, single-point crossover was used with a probability value set at 0.6, the mutation probability rate was set at 0.0001 and the population size was set at 100. The results are shown in Fig. 1 and indicate that all types of selection operator investigated were able to ®nd the best solution. However, the e ciency in terms of the number of generations required varies between selection operator types, i.e. both the tournament and truncated selection operators allowed the GA to ®nd good solutions in fewer generations than the roulette wheel and the roulette wheel with elitism operators. The inclusion of the elitism option improved the performance of the roulette wheel operator by retaining in the next generation the best solution found in the current population. Th...