مدل سازی دینامیکی شبکه های تولید سیستم های کار مستقل با کنترل ظرفیت های محلی

In this paper, a dynamic model is presented for production networks with a potentially large number of autonomous work systems, each having local capacity control. The model allows fundamental dynamic properties to be predicted using control-theoretic methods, together with the response of variables such as work-in-progress and lead-time for the network and its individual work systems. This is illustrated using industrial data. The behavior of one of the work systems in this network is analyzed further, and the results are compared with results obtained using a discrete event simulation model.

Production networks are emerging as a new type of cooperation between and within companies, requiring new techniques and methods for their operation and management [1]. Coordination of resource use is a key challenge in achieving short delivery times and delivery time reliability. These networks can exhibit unfavorable dynamic behavior as individual organizations respond to variations in orders in the absence of sufficient communication and collaboration, leading to recommendations that supply chains should be globally rather than locally controlled and that information sharing should be extensive [2] and [3]. However, the dynamic and structural complexity of these emerging networks inhibits collection of the information necessary for centralized planning and control, and decentralized coordination must be provided by logistic processes with autonomous capabilities [4]. Dynamic models will be an important tool in understanding the behavior of networks with decentralized control and enabling design of effective autonomous logistic processes. A production network with several autonomous work systems is depicted in Fig. 1. The behavior of such a network is affected by external and internal order flows, planning, internal disturbances, and the control laws used locally in the work systems to adjust resources for processing orders. In prior work, sharing of capacity information between work systems has been modeled [5] along with the benefits of alternative control laws and reducing delay in capacity changes [6]. Several authors have described both linear and nonlinear dynamical models for control of variables such as inventory levels and work in progress (WIP), including the use of pipeline flow concepts to represent lead times and production delays [7]. A closed-loop production planning and control concept has been employed with adaptive inventory control in decision support systems in a multi-product medical supplies market [8]. State-space models have been used for switching between a library of optimal controllers to adjust WIP in serial production systems in the presence of machine failures [9], and switching of control policies in response to market strategies has been investigated [10]. In this paper, the focus is on development of a discrete state-space dynamic model for production networks with an arbitrarily large number of work systems, illustrating the use of this generic model to predict performance, and comparing the results with results obtained using discrete event simulation.

The dynamic model that has been presented represents the fundamental behavior of a production network with an arbitrarily large number of autonomous work systems with local capacity control. It has been shown that transfer functions are readily obtained that represent the inter- and intra-work-system relationships between the outputs (orders input, orders processed, order output rates and WIP) and inputs (external input, work disturbances, planned WIP and capacity disturbances) of all work systems. In the production network example that was studied, WIP remained in the vicinity of the planned value, with variations due to fluctuations in external order inputs. Furthermore, lead times were held relatively constant with a simple proportional WIP-control law even with no exchange of information between work systems. One of the work systems in the network was analyzed further, and its response was characterized using results from both the dynamic model and a DES model, which also incorporated local capacity adjustment. Similar means and variations were predicted for WIP and lead time; however, greater differences are expected in future comparisons as DES models become more detailed and large numbers of work systems are incorporated. Tradeoffs need to be better understood between the complexity of detailed DES models that accurately model response in given scenarios and the potentially lower fidelity of control-theoretic dynamic models that can characterize fundamental dynamic properties. Supporting intuitive trial-and-error methodologies and theoretical methodologies, respectively, these approaches will be complementary tools in the design of autonomous logistic processes for production networks.