AN ORDERING GENETIC ALGORITHM FOR ASSEMBLY PLANNING P. De Lit 1 , P. Latinne 2 , B. Rekiek 1 and A. Delchambre 1 1- Université Libre de Bruxelles, Department of Applied Mechanics 50, av. F. D. Roosevelt CP 165/41, B-1050 Brussels, Belgium Phone: +32-2-650.47.66, Fax: +32-2-650.27.10, e-mail: pdelit@ulb.ac.be 2- Université Libre de Bruxelles, Department of Artificial Intelligence (IRIDIA) 50, av. F. D. Roosevelt CP 194/06, B-1050 Brussels, Belgium Keywords: Assembly planning, ordering genetic algorithm, multi-criteria decision-aid. ABSTRACT Assembly sequence planning (ASP) aims to find an optimal sequence to assemble a product. We propose here to use the ordering genetic algorithms (OGA) to solve this problem. The approach is based on three main ideas. Firstly, the identification of the subassemblies in the sequence is done by keeping trace of the components membership to a set of parts all along the sequence. Secondly, precedence constraints are satisfied by an univocal transformation of the studied sequence into a valid one, thanks to “precedence values” changing through the sequence. Finally the evaluation of each sequence is realized thanks to a multi-criteria decision-aid method called Promethee II. The use of such a method avoids the aggregation of several technical criteria into a unique fitness value, and lets us compares the individuals of the population one to each other. 1 INTRODUCTION An assembly product is composed of N parts connected together. When the number of components grows, several combinations can be chosen to execute the assembly operations to build the final product. The assembly sequence planning aims to determine and evaluate the different ways to assemble the product from its elementary components. This problem is now well known and several approaches were proposed to generate evaluate and select the best plans [1, 2, 3]. As the problem is a combinatorial one, the number of valid plans may become very huge as the number of components grows. So authors tried to use metaheuristic methods to generate assembly plans. For example, Sebaaly and Fujimoto [4] proposed a complete method based on GAs, with an interesting idea consisting to transform any sequence to a valid one, thanks to a reference matrix α ij , which values are determined by the links between components and precedence constraints. Bonneville et al. [5] presented a GA to generate assembly trees, using syntactic operations on a LISP representation of assembly trees. Unfortunately these works do not clearly explain the cost function used. Most of the time when using genetic algorithms, a weight is given to the evaluation index of each criterion and these indexes are summed over the set of criterions. This aggregation makes the fitness of an individual easier to compute, but the parameters tuning suffers on a lack of coherence for the user. In the remainder of this paper we present an ordering genetic algorithm (OGA) to tackle the assembly sequence planning problem. The essential features of the approach are described together with a simple illustrative