Algoritmo genético-tabu para a programação reativa da produção em um sistema de manufatura com recursos compartilhados

Abstract

Manufacturing systems with shared resources have been designed to meet the market needs, which require high quality products, low cost and the guarantee of the conditions agreed with the customer. The production scheduling in a manufacturing system is a complex task due to its combinatorial nature. Several studies show the use of search methods, such as Genetic Algorithms (GAs) and Tabu Search (TS), applied to the refinement of the production scheduling problem. The GAs present as disadvantages, the premature convergence, the lack of search intensification mechanisms in promising regions and also the lack of mechanisms which maintain the diversification of the population. In order to make GAs more effective in their search, by avoiding the premature convergence and ensuring the population diversity, some mechanisms are added to them so as to adjust their parameters during the search process, the so called Adaptative Genetic Algorithms (AGAs). However, these mechanisms do not guarantee the search intensification in the promising regions found. On the other hand, the TS presents mechanisms of search intensification and diversification, although its computational time depends on how optimal its initial solution is (solution by which the search process starts). In order to overcome the limitations of the traditional search methods, the Hybrid Algorithms (HAs) have been developed. They consist of the association of one method with another so that one helps the other in its deficiencies . This dissertation proposes the development of a Genetic-Tabu Algorithm (GTA) applied to the problem of the production reactive scheduling in a manufacturing system with shared resources, in order to ensure good compromise between makespan values and feedback time. The TS will be a functionality added to the AG and to the AGA, in other words, it will be a procedure to refine the individual(s) of the initial position and also to refine the individual obtained by the search methods. Tests have been conducted to determine which selection method (roulette or tournament) is more adequate for the definition of the neighborhood structure and also for the definition of the time to apply the mutation operator. Besides, other tests have been conducted by using different ways to calculate the makespan; one of them proposed in this dissertation and the others by Deriz (2007) and by Sanches (2008).