|
PROJECTS
AFFILIATED FACULTY
AFFILIATED LABORATORIES
EXTERNAL SUPPORT
- Electric Power Research Institute (EPRI)
- ARO
- AFOSR
- NSF
- AFRL
- HP
- Nokia Corporation
- Alcoa Research Center
- Foresight, Nu Thena Systems Inc.
AFFILIATIONS
- Analog Devices, Inc.
- MIT Laboratory for Information and Decision Systems (LIDS)
- Coordinated Science Laboratory, University of Illinois, Urbana-Champaign
- Sycamore Networks, Inc.
- Draper Laboratory
- Foresight, Nu Thena Systems Inc.
- Genuity Corporation
- Tellabs Operations Inc.
- Los Alamos National Laboratory
- Nokia Research Center
- HP
|
 |
« Back to areas
MANAGEMENT of COMPLEX PRODUCTION and SERVICE SYSTEMS PROJECTS
Management of Supply Chains: Production, Service, and Distribution Systems
- Supply Chain Production Planning in Time-Based-Competition Markets
Objective: Links/production systems collaborate in a supply chain to meet final demand. Each link is modeled as a production network forming a super-node in a super-network used to model the supply chain. Vertical coordination of supply chain supper-nodes is achieved in an iterative solution of (i) a master problem producing a tentative production schedule for each link, and (ii) the decentralized solutions of labor planning and performance evaluation sub-problems at each link. This closed-loop production-planning-approach maximizes the velocity of the supply chain by capturing the fact that lead times in each link are dynamic as they depend on the time varying production requirements.
Methodological Approach and Required Background: Hierarchical decomposition techniques are augmented through information on production lead times fed back by decentralized non-linear response surface estimation sub-problems. Lead times at each supply chain link during each period of the planning horizon are modeled as a function of production requirements, set-up delays, and detailed intra link scheduling and resource allocation. Mathematical programming techniques are used to solve the master problem. Analytical probabilistic models or Monte Carlo simulation models are used at the sub-problem level.
Affiliated Faculty:
Michael Caramanis, mcaraman@bu.edu
Ioannis Paschalidis, yannisp@bu.edu
J. Q. Hu, hqiang@bu.edu
James R. Perkins, perkins@bu.edu
- Control of Supply Chains
Objective: In the modern manufacturing environment no factory is an island. Companies consist of factories, suppliers, distributors, and customer service centers scattered around the globe. As a result, managing this global supply chain becomes increasingly important. Manufacturing is also becoming more customer oriented. Customers are more demanding, and require customized products delivered in a timely manner. Consequently, Quality of Service (QoS) becomes a predominant factor in acquiring and maintaining market share. Our primary research objective in this area is to devise effective policies to optimize the operation of supply chains in a way that addresses the needs of this new modern manufacturing environment. Key control decisions include: (a) idling decisions, which determine at each point in time whether the facilities located at different stages of the supply chain should be producing; (b) capacity allocation decisions, which determine which products should be processed by these facilities when they are producing; and (c) operational decisions, which relate to the optimization of the production process in each of these facilities. These latter decisions involve scheduling and routing in a multiclass queueing network (MQNET), which can be used to model the internal operation of such facilities. In parallel to our work in communication networks we utilize probabilistic constraints to capture Quality of Service (QoS). This is a significant departure from most of the work in the literature, which has focused on minimizing expected linear inventory and backorder costs. A more appropriate performance objective is to minimize expected (not necessarily linear) inventory costs subject to stockout probabilities being bounded above by given constants.
Methodological Approach and Required Background: In this framework, stockouts constitute events of interest and can be analyzed using Large Deviations techniques. Such techniques can handle autocorrelated demand and service processes, which allows us to model more realistic demand conditions than was before possible, and failure-prone production facilities. To make scheduling and routing decisions within production facilities we combine some earlier work on characterizing the region of achievable performance for MQNETs with some newer approximate dynamic programming techniques. Required student background includes Stochastic processes, Queueing systems, Optimization, and Stochastic Dynamic Programming.
Affiliated Faculty:
Ioannis Paschalidis, yannisp@bu.edu
Affiliated Laboratories/Center:
Control of Discrete Event Systems (CODES) Laboratory
Production Control of Manufacturing Systems (PCMS) Laboratory
Center for Information and Systems Engineering (CISE)
Planning Scheduling and Control of Complex Stochastic Systems
- Machine Scheduling of Stochastic Production Systems
Objective: The development of routing, set-up and machine loading policies that optimize the performance of a multiple workstation multiple part type processing production system. Performance measures to be optimized include system time (queueing + processing), inventory (Work in Process and Finished Goods), and Quality of Service (backlog, tardiness, on time service provisioning frequency).
Methodological Approach and Required Background: Stochastic Dynamic Programming, Queueing Systems, Markovian Decision Making.
Affiliated Faculty:
Michael Caramanis, mcaraman@bu.edu
Ioannis Paschalidis, yannisp@bu.edu
James R. Perkins, perkins@bu.edu
- Stochastic Fluid Models for Control and Optimization of Manufacturing Systems
Objective: Although naive fluid approximations may introduce significant deterioration in estimation accuracy of manufacturing system performance metrics, recent results have shown that they provide extremely accurate models for the purpose of control and optimization. A combination of such models with optimization algorithms leads to real-time adaptive control of manufacturing systems with little or no information about their stochastic characteristics required.
Methodological Approach and Required Background: Discrete event simulation is used to study and compare stochastic fluid models and their actual discrete event system counterparts. Perturbation Analysis methods are used to develop sensitivity estimators.
Affiliated Faculty:
Christos Cassandras, cgc@bu.edu
Michael Caramanis, mcaraman@bu.edu
Ioannis Paschalidis, yannisp@bu.edu
-
Integrating Manufacturing Processes and Operational Control (Hybrid Systems)
Objective: Hybrid systems combine a lower-level component with time-driven dynamics (describing the physical state of the system) with a higher-level component with event-driven dynamics (describing the changes in the operating modes of the system). This is a natural framework for manufacturing processes: The physical characteristics of production parts undergo changes at various operations described by time-driven models (e.g., differential equations), while the timing control of operations is described by event-driven models (e.g., timed automata). Accordingly, manufactured parts are characterized by physical states (e.g., temperature, geometry) subject to time-driven dynamics, and by temporal states (e.g., operation start and stop times) subject to event-driven dynamics. The tradeoff between the "quality" of parts and various timing requirements on part delivery leads to optimal control problems aimed at jointly optimizing the performance of both system components.
Methodological Approach and Required Background: Basic parametric and dynamic optimization techniques are used to formulate and solve control and optimization problems in the hybrid system setting described above.
Affiliated Faculty:
Christos Cassandras, cgc@bu.edu
-
Air Traffic Control
Affiliated Faculty:
Christos Cassandras, cgc@bu.edu
Affiliated Laboratories/Center:
Control of Discrete Event Systems (CODES) Laboratory
Production Control of Manufacturing Systems (PCMS) Laboratory
Center for Information and Systems Engineering (CISE)
Performance and Sensitivity Evaluation of Stochastic Systems
- Monte Carlo Simulation of Performance and Sensitivity of Fluid Model Approximations
Objective: Performance evaluation of stochastic production systems with blocking and dynamic dispatch policies is generally not analytically tractable, rendering Monte Carlo Simulation the sole analysis option. In this context, fluid model approximation has many advantages. It enhances computational efficiency, retains key dynamics properties, and finally facilitates sensitivity estimation since it is amenable to Infinitesimal Perturbation Analysis (IPA) techniques. Nevertheless, naive fluid approximations introduce significant deterioration in estimation accuracy, and distort certain key aspects of dynamic behavior associated with system delay propagation when the system is in the vicinity of empty or full buffers. Enhancements are therefore desirable to the extent that they can retain the benefits of fluid approximation while improving the accuracy of performance statistics, including moments of trajectories in the vicinity of empty/full buffer states.
Methodological Approach and Required Background: Event driven Monte Carlo simulation algorithms of fluid model approximations are considered. Queueing delay dynamics modeling is enhanced by imposing explicit constraints on effective production rates in terms of finite capacity input and output buffer levels. We introduce differential equation constraints of the form and where is production rate at time t of machine i, , the level of the input and output buffer respectively of machine i at time t, Ni the capacity of the output buffer, and Ci the production rate capacity of machine i. Development of event driven simulation and IPA algorithms for the enhanced fluid approximation models is the main objective of this research effort.
Affiliated Faculty:
Michael Caramanis, mcaraman@bu.edu
Christos Cassandras, cgc@bu.edu
Ioannis Paschalidis, yannisp@bu.edu
-
Quick Response Manufacturing
Objective: Quick Response Manufacturing (QRM) is an enterprise-wide strategy whose chief goal is to reduce lead times. In contrast to "lean" manufacturing, QRM is primarily intended for manufacturing environments and supply chains with low volume operations and large product mix and variability. The application of QRM in practice requires a combination of effective modeling tools with emphasis on controlling manufacturing system components that can reduce average lead times and work in process. An example is controlling lot sizes to exploit the tradeoff between long production runs and setup times required to switch between products.
Methodological Approach and Required Background: Queueing models and related software are used to capture the dynamic behavior of potentially very complex manufacturing environments. Stochastic optimization methods are invoked to formulate and solve problems intended to minimize average lead times subject to specific constraints. Discrete event simulation is used for testing the algorithms developed.
Affiliated Faculty:
Christos Cassandras, cgc@bu.edu
Affiliated Laboratories:
Control of Discrete Event Systems (CODES) Laboratory
Production Control of Manufacturing Systems (PCMS) Laboratory
Dynamic Resource Allocation
- New Product Development (NPD) Portfolio and Pipeline Management
Objective: The development of methods for optimal selection of NPD projects to form an NPD portfolio, for dynamic management of projects at review gates, and for pipeline resource allocation. The project involves (1) modeling decision processes in portfolio selection and periodic review using stochastic models that incorporate uncertainty in the development process and the market; (2) defining an appropriate notion of risk management and quality of information used for decision making; (3) and determining optimal decisions. Allocation of human resources to NPD projects in the pipeline raises new scheduling issues due to the nature of resources and tasks and addressing these forms another component of the project.
Methodological Approach and Required Background: Stochastic modeling, Markov Decision Processes, Dynamic Programming, Optimal Control, Optimization.
Affiliated Faculty:
Pirooz Vakili, vakili@bu.edu
James R. Perkins, perkins@bu.edu
-
Cooperative Control of Autonomous Vehicles
Affiliated Faculty:
Christos Cassandras, cgc@bu.edu
Affiliated Laboratories/Center:
Control of Discrete Event Systems (CODES) Laboratory
Production Control of Manufacturing Systems (PCMS) Laboratory
Center for Information and Systems Engineering (CISE)
Back to top
|
|
