Task assignment with imprecise information for real-time operation in a supply chain
Introduction
A supply chain can be defined as a network of autonomous or semi-autonomous business entities collectively responsible for procurement, manufacturing, and distribution activities associated with one or more families of related products [13]. Coordinating the activities of the various entities in a supply chain is important for improving performance in terms of multiple objectives such as on-time delivery, quality assurance, and cost minimization. A typical supply chain is shown in Fig. 1. It shows the upstream entities which are the suppliers and the downstream entities which are the consumers. As is evident, a supply chain is a distributed system.
Usually, there are three phases of decision making in supply chain management depending on the frequency of each decision and the time frame over which a decision phase has an impact [4]. These are the supply chain strategy phase, the planning phase, and the operation phase. In supply chain operation, the time horizon is weekly or daily, and companies make decisions regarding individual customer orders. Here, firms allocate individual orders to inventory or production, set a date that an order is to be filled, allocate an order to a particular shipping mode, set delivery schedules of trucks, etc. For many situations, it is desirable to re-optimize the schedules each time a new order is being considered [11]. This would enable details for scheduling the new order to be efficiently integrated with all other unscheduled orders. A new order may have a deadline associated with it and hence the whole operation has to be modeled with soft real-time constraints. This is referred to as real-time operation in a supply chain.
Multiagent systems consist of multiple autonomous or semi-autonomous agents where knowledge and control is distributed and agents communicate through messages [3]. Since supply chain operation is concerned with coordination among multiple decision makers, a multiagent framework for modeling it is a natural choice. In the current scenario, the consumer demand is ever changing and there are fluctuations in resource costs and availability. Hence companies must respond rapidly to maintain consumer satisfaction and profits. As these changes are occurring at increasing speeds, scales and complexity unmanageable by humans, the need for automated techniques becomes acute [14].
Let us consider a part of a supply chain with a focal organization H0 and its upstream entities (the suppliers) as shown in Fig. 2. H0 may not have all the required resources to completely perform all the subtasks of a task submitted to it. Those subtasks, which cannot be performed at H0, have to be assigned to its suppliers quickly in real-time operational environments. As depicted, in a practical situation, the suppliers of H0 may belong to more than one supply chain. Hence H0 does not have a complete control of the scheduling activities of its suppliers. At the same time, more than one supplier may have the resources to perform a subtask. H0 has to make a decision to choose one of the suppliers to which a subtask needs to be delegated. A similar task assignment decision has to be done by all the suppliers for the tasks delegated to them.
Usually there are multiple objectives (or criteria) like minimizing cost and time, maximizing quality, etc. associated with a task. Moreover, different preference ratings may be assigned to each of the criteria. These preference ratings, which are the relative importances for each of the criteria, may be assigned by the focal organization, downstream entities, or the end-users themselves. The problem of the assignment of a task to a node (we will henceforth use the common term node for an organization, a company, or a firm in a supply chain) is important since a good assignment would make it feasible to satisfy the criteria. Thus, task assignment in a supply chain is a multiobjective decision making problem.
In practice, it is convenient to assign ordinal values to the preference ratings. Linguistic terms like low, moderate, very high, etc. could also be used. There is imprecision implicit in the system under consideration. The first source of the imprecision is in the subjective specification of the preference ratings using ordinal values. Since we are considering a dynamic distributed system, another source of imprecision arises. A node does not have a complete view of the entire chain and hence does not have the global system state. Hence the global state has to be collected before the task assignment. Since the utilization of the resources and hence the schedules at various nodes change continuously, it is not possible to get a precise system state. Hence the decision making needs to be done allowing for tolerance for the imprecision.
In this paper, a model for the supply chain as a multiagent system is presented first with an agent located at each of the nodes. We design a real-time scheduler for the agents that would integrate schedules of new orders having soft real-time deadlines with the already existing schedules. To deal with the imprecision mentioned, a fuzzy set approach [16] is used in the context of multiobjective decision making for task assignment. We adapt the multiobjective decision process using fuzzy sets for the task assignment problem in a supply chain management system. The proposed technique is evaluated against other commonly used heuristic approaches for task assignment. It is seen that for a large range of task arrival rates the fuzzy technique has superior performance. However, its performance deteriorates for very high arrival rates.
To overcome this limitation, we next design a hybrid algorithm which augments the multiobjective decision process using fuzzy sets. This approach does not suffer from performance degradation even at very high task arrival rates. The hybrid algorithm is compared with other techniques and it is found that it out-performs all of them for all arrival rates.
An example where the proposed approach could be applied is the apparel industry. In this industry, outsourcing for various kinds of clothes is common and an apparel manufacturing company typically has many suppliers for the same types of clothes, each having different quality, cost, and lead times. A supplier may be part of different supply chains catering to more than one apparel manufacturing companies. A task here corresponds to the making of a trouser, suit, etc. Due to the ever changing market conditions, the manufacturer has to operate with real-time constraints to be competitive.
The paper is organized as follows. In Section 2, we review the related works. The supply chain is modeled using a multiagent framework in Section 3. The real-time scheduler used in the agents is explained in Section 4. In Section 5, the task assignment in a supply chain is formulated as a multiobjective decision making problem. Simulation studies are presented in Section 6. In Section 7, the hybrid algorithm is proposed. Section 8 concludes the paper.
Section snippets
Related works
Modeling of the supply chain using a multiagent framework has been done in the past. In [10], a reconfigurable, multi-level, agent based planning and scheduling architecture called Multi-Agent Supply Chain cOordi-nation Tool (MASCOT) has been designed. It is aimed at supporting dynamic, effective, and efficient coordination among supply chain partners for various activities like production, transportations, and other strategic decisions. It discusses new coordination protocols for better
Agent based system model of the supply chain
A supply chain can be represented as a directed acyclic graph. Let us define a set UpS(N) that includes all those nodes which are in the immediate upstream (i.e. immediate suppliers) of a node N. Note that the set UpS(N) may be empty and that the sets UpS(N1) and UpS(N2) for two nodes N1 and N2 need not be disjoint. Let H0 be the focal node under consideration. Let UpS(H0) = {H1, H2, …, Hn}. The nodes communicate only through messages and the communication delay for sending or receiving
The real-time scheduler at each FU
This section presents the scheduler at an agent F of a FU which controls a resource R having a duration Δ.
The scheduler is invoked periodically with the period δ × Δ (δ is a constant). At each invocation, the scheduler first finds out all the enabled tasks from QI. The enabled tasks are those for which all incoming enabling commitments are known. The enabling time for a task is the maximum of all the incoming enabling commitments. The expected worst case start time (EWCST) of all the enabled
Task assignment under imprecision
In this section, we describe how the task assignment in a supply chain, which is a distributed system, can be formulated as a multiobjective decision making problem. In a distributed system, task assignment has to be done when the global system state is not available at any single node. The node where the decision is taken has to collect the global state first. There is imprecision in the state collection itself since we are considering a dynamic system. Since the resource utilization changes
Performance evaluation
In the experiments carried out below, we have simulated a single focal node H0 and twelve nodes in its upstream entities. The supply chain used for simulation is shown in Fig. 6. The resources are distributed randomly across all nodes in the network. We randomly generate tasks and the preference ratings associated with every task. The task arrival at H0 is modeled as a Poisson process with the arrival rate X. When a task execution is over, we calculate a performance measure which is the
The hybrid algorithm
In order to avoid the performance degradation of the fuzzy technique at very high arrival rates, we design a hybrid algorithm for task assignment that reduces the dependence on the surplus values at very high arrival rates. It has been shown in [8] that the Guarantee Ratio-which is the ratio of the tasks that finish within their deadlines to the total number of input tasks-decreases drastically at very high arrival rates. We use this fact to design the hybrid algorithm which is described below:
- (1)
Conclusions
In this paper, a supply chain is modeled using a multiagent framework and an architecture for each of the agents is presented. We present a real-time scheduler that can accept new tasks with soft real-time deadlines. The schedules for the new tasks are integrated with the schedules of the already existing tasks in the system. The problem of task assignment for the real-time operation in a supply chain is then addressed. Multiple criteria like quality, cost, and duration may be associated with
References (16)
Fuzzy sets
Inform. Control
(1965)- J. Andersson, A Survey of Multiobjective Optimization in Engineering Design, Technical Report No. LiTH-IKP-R-1097,...
- et al.
Decision making in a fuzzy environment
Manage. Sci.
(1970) - et al.
Readings in Distributed Artificial Intelligence
(1988) - S. Chopra, P. Meindl, Supply Chain Management: Strategy, Planning, and Operation, Pearson Education Inc.,...
Multiobjective Optimization Using Evolutionary Algorithms
(2001)- K.S. Decker, Environment Centered Analysis and Design of Coordination Mechanisms, Ph.D. thesis, University of...
- U. Deshpande, A. Gupta, A. Basu, Adaptive Fault Tolerant Hospital Resource Scheduling, in: R. Meersman, Z. Tari, et....
Cited by (14)
A review of soft computing applications in supply chain management
2010, Applied Soft Computing JournalA modified two-tuple FLC model for evaluating the performance of SCM: By the Six Sigma DMAIC process
2007, Applied Soft Computing JournalFuzzy fixed charge solid transportation problem and algorithm
2007, Applied Soft Computing JournalAn overview of fuzzy techniques in supply chain management: Bibliometrics, methodologies, applications and future directions
2021, Technological and Economic Development of EconomyOptimal task assignment based on incentive design in the virtual enterprise: A Principal-agent Perspective
2019, ACM International Conference Proceeding SeriesAn evolutionary algorithm for supply chain network design with assembly line balancing
2017, Neural Computing and Applications