سیستم های چند عامل در برنامه ریزی تولید و کنترل: برنامه به منظور زمان بندی در خطوط مونتاژ چند منظوره مدل

This work deals with production smoothing, one of the keys of success of Just In Time and Lean Production. By levelling the load of the workstations, production smoothing allows a regular material flow, shorter manufacturing lead times, and lower work in process. Different solutions to the mixed-model assembly lines sequencing problem have been proposed in literature. In this paper, a Multi-Agent System is presented, which solves this problem according to the theory of autonomous agents. The experimental results show that this innovative approach has a good performance if compared with the traditional ones.

Short-term production planning techniques have known a long evolution starting from the Seventies. After a first era of optimisation dominion, heuristic approach tried to overcome its limitations, yet introducing others. Later, researchers’ attention shifted to other innovative paradigms, which are more effective in modern dynamic and complex contexts. In particular, one of these approaches is Artificial Intelligence (AI): nowadays autonomous agent theory, a product of AI, is one of the most interesting fields of research as far as production planning and control are concerned. The topic of this paper is an application of autonomous agent theory to a particular short-term production planning problem, sequencing of mixed-model lines, which has been studied for years in literature. After a brief overview of the evolution of short-term production planning techniques across last decades (Section 2), an introduction to autonomous agent theory and multi-agent architectures is reported in Section 3. Later, in Section 4, the problem of sequencing of mixed-model assembly lines is presented, together with the model usually adopted in literature in order to solve this problem. Once modelled, the problem can be solved through different techniques: in particular, the section presents the heuristic and the optimisation approaches, as they have been developed in literature; moreover, it proposes a multi-agent architecture which has been recently developed by the authors. The performance of this architecture has been tested and compared with other performing models: the experimental results, shown in Section 5, suggest the existence of significant margins of improvement on the performance of traditional approaches.

This paper has presented three different approaches to short-term production planning problem of mixed-model lines. As stated in the previous sections, the optimisation approach (i.e. [22]) is more suitable for static contexts, where the effort to determine the optimal solution (mainly due to algorithmic complexity, long processing time, etc.) is justified. On the other hand, heuristic algorithms are much more simple, in terms of complexity, and they require shortest processing time: so that they are more suitable for dynamic contexts, where frequent re-programming is needed. Last but not least, the autonomous agents approach is a different and innovative technique. The classical problem of short-term production planning of mixed-model lines is basically mono-objective and deterministic, and so the utilisation of innovative resolution techniques seem not justified (i.e. the agent-based approach does not improve the results of the Bautista et al. [22] algorithm). Nevertheless, the autonomous agents architecture which has been presented in this paper is interesting at least for the following two reasons: • the proposed architecture is characterised by a distributed control (various autonomous entities fill the system and each of them is endowed with intelligence and objectives); as a consequence, such a model is modular, that is new bonds or new degrees of freedom (for example, the possibility of working the same product on more than one assembly line indifferently) or new objectives (for example, the minimisation of setup time) can be easily introduced without heavy modifications of the architecture: for example, they can be faced with a redefinition of agents’ objective functions or with the addition of new agents to the system • the agent architecture shows a good performance in comparison with traditional algorithms and it can potentially find a better solution than that obtained by other heuristics, thanks to the introduction of co-operation, a learning process which reduces the myopia of this procedure in comparison with others. The state-of-art of research in autonomous agent theory has not yet proposed efficient and stable instruments for software development to make possible the application of this technique to real manufacturing contexts. As a consequence, only a huge experimental activity is possible at the moment, as the one is in progress at Politecnico di Milano. In any case, the experimental results obtained until now show that the autonomous agent technique has a great potential which will be surely exploited in the future through real applications, as soon as the necessary instruments will be developed.