Subasta combinatoria para la programación dinámica en sistemas de fabricación distribuidos

Resumen

Los métodos de programación de operaciones tradicionales, estáticos y basados en arquitecturas centralizadas o jerárquicas, no son suficientemente flexibles para adaptarse al dinamismo y complejidad de los sistemas de fabricación actuales. Por ello, la investigación en técnicas dinámicas de programación y control on-line está creciendo rápidamente. En este ar tículo se presenta una técnica de programación dinámica on-line basada en modelos de mercado e implementada sobre un sistema multiagente. Esta propuesta, además de dinámica, es distribuida no jerárquica, lo que aporta al sistema las características deseadas.

Palabras claves: programación y control de la producción, programación on-line, sistemas de fabricación basados en agentes, subastas combinatorias.


Combinatorial auction for dynamic scheduling in distributed manufacturing systems

Abstract: The traditional static scheduling methods, based on hierarchical and centralized architectures, are not flexible enough to self-adapt to the dynamism and complexity of today’s manufacturing systems. For this reason, new proposals to improve the scheduling and control of agile manufacturing systems constantly appear. The auction based allocation methods as well as the software paradigm of multiagent systems, which offers new techniques to face complex unsolved problems, can help to find promising solutions in manufacturing systems. Traditionally, scheduling problems have been solved offline by a centralized decision-maker that use a global optimisation model. We propose to include in the system several decision-makers modelled as agents instead. We consider two kinds of agents: order agents and machines agents. Each order agent represents a product that is characterized by its operations, precedence relationships and due date. The goal of each order agent is to find machines that can perform the required operations and hence completing successfully the order. Each production order creates its own schedule (local schedule). An auction mechanism ensures that local schedules are nearly compatible (several orders don’t use the same machine at the same moment) and globally efficient. Every agent in the system can communicate with other agents through the exchange of messages. The interaction mechanism is ruled by means of a combinatorial auction where a theoretical basis is provided for structuring message sequencing, bid evaluation, and price updating. Our research contributes to the auction technique in manufacturing scheduling and control in two basic aspects: (1) we apply the auction mechanism for a routing flexible environment (an operation can be performed in several machines with a differing efficiency), (2) we propose an implementation that can schedule online, updating real-time information: planning horizon, changes in orders, changes in machine availability and capabilities. We include explicitly the option of reallocate resources in real time when a new order arrives to the system. In order to test the features of this approach we display some computational results. Preliminary results show efficient performance in dynamic scenarios, but there are still many matters to investigate. Future works will be devoted to test the proposed approach on more case studies or even on real cases. We can add complexity to the structure of the system, and we must improve some aspects of the auction mechanism such as convergence and stability.

Keywords: manufacturing programming and control, on-line scheduling, agent based manufacturing systems, combinatorial auctions.