Scott A. Moses

Real-time due-date promising by build-to-order environments

A vast amount of literature exists on scheduling to meet due dates, but very little work considers how to set these due dates before scheduling the orders. A method is described for real-time promising of order due dates that is applicable to discrete build-to-order environments facing dynamic order arrivals. When computing a due date, the method considers: (1) dynamic time-phased availability of resources required for each operation of the order, (2) individual order-specific characteristics and (3) existing commitments to orders that arrived previously. Performance of the method surpasses that of due-date assignment methods previously examined in the literature and also those commonly used in practice. The median and standard deviation of absolute flow-time estimation error and of absolute lateness are chosen as the primary performance criteria because they capture both positive and negative error in flow-time estimation of each individual order. Computational results from large-scale simulation studies of realistic systems with 20 resources and up to 100 000 orders also indicate the method is highly scalable.

Real-time prediction of order flowtimes using support vector regression

In a make-to-order production system, a due date must be assigned to new orders that arrive dynamically, which requires predicting the order flowtime in real-time. This study develops a support vector regression model for real-time flowtime prediction in multi-resource, multi-product systems. Several combinations of kernel and loss functions are examined, and results indicate that the linear kernel and the @e-insensitive loss function yield the best generalization performance. The prediction error of the support vector regression model for three different multi-resource systems of varying complexity is compared to that of classic time series models (exponential smoothing and moving average) and to a feedforward artificial neural network. Results show that the support vector regression model has lower flowtime prediction error and is more robust. More accurately predicting flowtime using support vector regression will improve due-date performance and reduce expenses in make-to-order production environments.

A scalable data structure for real-time estimation of resource availability in build-to-order environments

This paper defines a highly scalable interval index structure called the Temporal Bin tree (TB-tree) that can be embedded in any resource planning application whose algorithms require efficiently estimating either the time that a resource will be available to process a specific task of known length or the net availability of a resource during a specified period of time. It is specifically engineered to meet the real-time response and space efficiency requirements of large-scale resource planning applications that are required for mass customization. Basically, the TB-tree is a binary tree structure that represents availability of a resource across a planning horizon. Representing intervals of availability hierarchically using a tree structure increases the efficiency of search for resource availability when the discretization of time is fine-grained or the planning horizon is long. The tree forms a backbone structure that does not require disruptive rebalancing during update operations, which would mitigate the ability of the tree to respond to queries in real time. Its specific implementation allows for random access at any level of the tree to further improve scalability. An application of planning to real-time promising of order due dates for custom built products provides the context for empirical evaluation. Results of analytical evaluations and simulation experiments clearly demonstrate the scalability of the TB-tree relative to existing index structures in terms of both time and space.

Due date assignment using feedback control with reinforcement learning

Good due date assignment for an order requires the calculation of a time buffer that will account for the uncertainties associated with the arrival of future orders in a dynamic environment. This paper presents a method that controls the size of this time buffer for a discrete manufacturing system. The applicability of the method to an unrestricted class of discrete manufacturing systems is preserved by the use of a feedback control paradigm, and control knowledge is acquired using reinforcement learning. The current trajectory of the state of the shop is considered so that due date performance is improved during transient conditions. Results of simulation experiments demonstrate the effectiveness of the approach.

The shifting bottleneck procedure for job-shops with parallel machines

The shifting bottleneck (SB) heuristic has been successfully applied to the job-shop scheduling problem. In this paper, the shifting bottleneck heuristic has been extended to solve job-shop problems with parallel machine workcentres. The efficient shifting bottleneck heuristic was developed to reduce the number of subproblems and provide tradeoff results between the computation time and the solution quality. The effect of bottleneck machine selection and reoptimisation procedures on the computational time is discussed. Moreover, a procedure is developed to achieve the desired makespan with near optimal number of machines of each machine type. The tradeoff between the makespan and the minimal number of machines required to achieve the makespan is also illustrated on few job-shop problems.

Scalability and performance of computational structures for real-time order promising

This paper analyzes the scalability and performance of computational structures used in an order promising method designed for discrete build-to-order environments facing dynamic order arrivals. Main-memory database technology and supporting index structures are exploited to improve scalability of this data-intensive procedure so that orders can be promised in real time for industrial-sized systems. Two index structures, a linear structure and a tree structure, are compared. The tree structure improves response times and also has higher scalability at the expense of larger memory requirements compared to the linear structure. Performance of the method is evaluated under two manufacturing scenarios: a tandem line and an assembly shop. Computational results obtained from large-scale simulation studies of realistic systems with 60 resources show the method to be robust. Accuracy of the method in terms of median absolute lateness surpasses that of due date assignment methods previously examined in the literature.

Using simulation to evaluate buffer adjustment methods in order promising

Much literature exists for scheduling production, but there is little work on establishing the due dates that serve as the inputs to developing a production schedule. We call this order promising. This paper explores a simulation-based approach for evaluating methods for promising the delivery of orders based on dynamic buffer adjustment coupled with various methods to forecast the amount of buffer required. The primary objective of the paper is to frame the problem and suggest methods of analysis. Preliminary computational results are also presented.

Scalability of indexing structures in a production systems testbed for computational research

This paper analyzes the scalability of indexing structures in a production systems testbed for large-scale computational research on discrete build-to-order environments. The testbed consists of five integrated services that interoperate for order entry, order promising, production planning, execution, and control. The service-based architecture is Java-based, object-oriented, event-driven, memory-resident, and multi-threaded. Services in the testbed very frequently need to rapidly locate specified elements in their large data models during algorithmic computations, and in that regard a number of indexing structures have been designed by computer scientists with the purpose of increasing the efficiency of data access. We explore the tradeoff between improved application scalability and increased implementation complexity of indexing structures by comparing the B+-tree, T-tree, and R-tree indexing structures to a simple and widely used linear structure in the context of an application to real-time order promising. Scalability is evaluated by measuring space requirements and the computational time as a function of the size of the system.

Disruption management in a two-period three-tier electronics supply chain

Abstract We study strategies to manage demand disruptions in a three-tier electronics supply chain consisting of an Electronics Manufacturing Services provider, an Original Equipment Manufacturer (OEM), and a Retailer. We model price sensitivity of consumer demand with the two functions commonly used for this purpose, linear and exponential, and introduce disruptions in the demand function. We assume each supply chain member faces an increasing marginal unit cost function. Our decentralized supply chain setting is governed by a wholesale price contract. The OEM possesses greater bargaining power and therefore is the Stackelberg leader. A penalty cost incurred by the Retailer is introduced to capture the cost of deviation from the original plan. We find exact analytical solutions of the effectiveness of managing the disruption when the consumer demand function is linear, and we provide numerical examples as an illustration when the consumer demand function is either linear or exponential. We show that the original production quantity exhibits some robustness under disruptions in both centralized and decentralized supply chains, while the original optimal pricing does not. We show that supply chain managers should not automatically react to an individual disruption, in certain cases it is best to leave the production plan unchanged.

Scalable material assignment methods for build-to-order environments

This paper develops scalable and effective yet simple heuristic methods for material assignment in a large discrete Build-To-Order (BTO) environment. The material assignment function allocates available supplies of raw material to orders over a period of time. Despite popular belief to the contrary, Material Requirements Planning (MRP) does not perform material assignment. As a result, material supplies are by default consumed by orders on a First Come First Served (FCFS) basis and this allows low priority orders to obtain material instead of high priority orders. We develop heuristic methods that consider priority when performing material assignment and study their effectiveness using total weighted tardiness, total tardiness, fraction of orders that are tardy and computational time as performance measures. The material assignment problem is formulated as a transportation problem to obtain a lower bound for the heuristics in terms of total weighted tardiness. Simulation results show that one of the heuristics yields solutions whose total weighted tardiness is within about 0.25% of the optimal solution, yet is very scalable and able to quickly assign materials in realistically sized systems with many tens of thousands of orders.