مواد توزیع شده و متمرکز برنامه ریزی حمل و نقل: مقایسه و نتایج حاصل از یک مطالعه شبیه سازی

Part of a larger research that employs decentralized holonic modelling techniques in manufacturing planning and control, this work proposes a holonic-based material handling system and contrasts the centralized and distributed scheduling approaches for the allocation of material handling operations to the available system resources. To justify the use of the decentralized holonic approach and assess its performance compared to conventional scheduling systems, a series of evaluation tests and a simulation study are carried out. As illustrated by the results obtained from the simulation study, the decentralized holonic approach is capable of delivering competitive feasible solutions in, practically, real-time.

The increasing influence of global economy is changing the conventional approach to managing manufacturing companies. Real-time reaction to changes in shop-floor operations, quick and quality response in satisfying customer requests, and reconfigurability in both hardware equipment and software modules are already viewed as essential characteristics for next generation manufacturing systems. The decentralized modular architecture, the cooperation and coordination mechanisms, and the real-time response capabilities of holonic systems make them a viable solution to achieve the above-mentioned characteristics [1]. An overview of the literature in holonic manufacturing scheduling showed that there are several approaches for designing the mechanisms for cooperation and coordination among the entities in the holonic architecture. Considering multiple manufacturing cells, Gou et al. [2] presented a coordination solution based on the pricing concept of market economy using a Lagrangian relaxation methodology with results reported as near-optimal in a timely fashion. Using a holonic decomposition framework for an entire supply chain, Walker et al. [3] developed a job-shop scheduling approach based on the formation of dynamic virtual clusters around a resource-scheduling dynamic mediator agent, the particular order and the potential resources to perform the order tasks. When compared to results given by scheduling heuristics and benchmark solutions, the holonic scheduling performance is encouraging. By combining evolutionary computation and dynamic programming, Sugimura et al. [4] proposed a real-time scheduling procedure to determine machining schedules through autonomous decision-making and cooperation among part holons and equipment holons. The solution provided a method to integrate process planning and scheduling system while providing both sequencing and scheduling of machines in the manufacturing system studied. A real-time control architecture viewed from the system, software, and functional architectures points of view is presented by Wang et al. [5] and [6], followed by an event driven real-time distributed control system developed via using a combination of intelligent agents and IEC 61499 function blocks. This work is part of a larger research that employs decentralized holonic modelling techniques in manufacturing planning and control for the purpose of obtaining better system performance [7], [8], [9], [10] and [11]. The main objective of the research so far, is to develop real-time feasible schedules for material handling (MH) resources working in stochastic manufacturing environments. Because of their rigid architecture, the existing MH systems are difficult to respond to the requirements set on future manufacturing. In this research, the holonic modelling framework is employed in the design of a decentralized control system used for scheduling MH operations in manufacturing cell environments. The designed Holonic-Material Handling System (H-MHS) uses specific internal holon evaluation and allocation procedures and inter-holon coordination mechanisms. To justify the use of decentralized holonic approaches for manufacturing control and assess the performance of the H-MHS in comparison to conventional systems, a series of evaluation tests and a simulation study are carried out. Optimal and heuristic search algorithms that serve as the basis for the MH conventional control approach are developed for this purpose. Section 2 gives a brief description of the H-MHS and its operation, and contrasts the two scheduling approaches that are the subject of this study. Section 3 depicts the characteristics and the design of experiments for the simulation study, while Section 4 presents the results obtained from the software implementation of the scheduling and simulation algorithms. Finally, Section 5 provides the conclusions coming from this study and future research directions. The characteristics tested in the experimental part comprise: the quality of the solution delivered, the real-time scheduling ability, which includes the real-time response to changes in production orders. By comparing the results given by the two alternative system configurations, the performance of the proposed decentralized holonic system can be evaluated.

As illustrated by the results obtained from running the preliminary tests, the proposed holonic system is capable of accommodating new arriving jobs and delivers good solutions in real-time. The simulation study results are validating the pilot test runs and give reliable information for the performance of the proposed holonic system and its centralized scheduling extension incorporated in the GVH's internal architecture. There is a very small difference in the results obtained by running the holonic algorithms compared to the optimal solution. Moreover, considering the real-time response, confirmed by the time to obtain a feasible solution which is always less than 1 min, the results given by the holonic approach show a significant potential. Using simple algorithms that never lead to combinatorial explosion, the performance of the holonic approach will always have the strength of a real-time response. Even the optimality of the solution cannot be guaranteed, using the procedures described in this research, the solution can be compared with the LB values, so its difference from the theoretical best possible solution can be calculated. Considering the large number of real-world scheduling problems for which searching for an optimal solution leads to combinatorial explosion, the lack of optimality of the holonic control approach is not an important issue overall. Future work will focus on the integration of the decentralized holonic MH operation scheduling and the holonic scheduling of processing operations in the same job-shop type of environment. Besides new arriving jobs, randomly occurring breakdowns in both processing machines and MH resources will be part of the working environment. In addition to the performance measures tested in this study, new performance measures, such as the real-time re-scheduling capability for the jobs affected by the breakdowns will be considered. The fundamental difference between decentralized manufacturing control systems, such as holonic control systems or other systems modelled using an agent-based framework, and traditional manufacturing control systems comes from the distributed decision-making (multiple small decision spaces) existing in decentralized systems as opposed to the centralized or hierarchical control (single large decision space) in the case of conventional scheduling systems. The distributed decision-making for the job assignment problem translates in simple algorithms having reduced computational complexity that can be solved practically in real-time for any potential real-world case.