Group technology by an ant system algorithm

Clustering of items to form meaningful groups is a theoretically challenging problem. Moreover, it has considerable practical value in manufacturing. Therefore, extensive research is conducted in this field and numerous techniques have been developed. The starting point of these techniques is usually part–machine incidence matrices. This data structure models the cell formation problem as well. Here, the aim is to get a block-diagonalized structure. This is the basic problem of group technology. The paper presents a novel and potent technique to solve this basic problem. The grouping problem is first represented as an artificial ant system. Then better and better groupings are obtained as semi-blind ants find their way by a communication-supported random search process. Finally, the proposed technique is compared with other AI methodologies, namely genetic algorithms, simulated annealing and tabu search. The main concern in this evaluation phase was to devise an environment appropriate for a fair assessment. For that reason, the stated techniques are formulated with the simplest possible configurations and parallel structures. Tests made using the well-known data sets from the literature revealed a remarkable outcome: ant systems perform better than the other AI techniques as far as an equal number of solution alternatives are concerned.