logo

European Institute for Statistics, Probability, Stochastic Operations Research and its Applications

About | Research | Events | People | Reports | Alumni | ContactHome


 
 

Stochastic models of manufacturing systems
June 24-25, 2010

PROGRAMME / ABSTRACTS

Thursday, June 24, 2010

09.30-09.45 Opening by Onno Boxma  

09.45-10.45

Keynote John Buzacott

What’s the Use of Stochastic Models of Manufacturing Systems?

10.45-11.15

Break

 

11.15-11.45

Hubert Missbauer

Order release planning based on stochastic models of manufacturing systems

11.45-12.15 Erwin v.d. Laan Managing uncertainty in hybrid manufacturing/remanufacturing inventory systems
12.15-12.45 Remco Bierbooms Performance Analysis of Production Lines with Fluid Flows and Finite Buffers

12.45-13.45

Lunch and Break

 

13.45-14.45

Keynote
George Shantikumar

Queueing Systems Modelling with Operational Statistics

14.45-15.15

Break

 

15.15-15.45

Fabian Wirth

Structure preserving model reduction for logistic systems
15.45-16.15 Dieter Armbruster Some issues for aggregate models of production systems
16.15-16.45 Casper Veeger Aggregate modeling in semiconductor manufacturing using effective process times
16.45-17.15 Nico van Dijk Lessons from an OT-ICU system for manufacturing
17.15-18.30 Drinks with posters  
19.00 Workshop  dinner

 

Friday, June 25, 2010

09.30-10.30

Keynote
John Fowler

Healthcare Systems Engineering

10.30-11.00

Break

 

11.00-11.30

Oliver Rose

An approach to model production systems with SysML

11.30-12.00

Ger Koole

Process optimization in hospitals

12.00-12.30 Çagdas Büyükkaramikli

Delivery Lead Time and Flexible Capacity Setting for Repair Shops with Homogenous Customers

12.30-13.30

Lunch and Break

 

13.30-14.30

Keynote Reha Uszoy

Production planning with resources subject to congestion

14.30-15.00

Break

 

15.00-15.30 Stéphane Dauzère-Pérès Modeling and optimizing operations in semiconductor manufacturing
15.30-16.00 Nicky van Foreest Order Acceptance Policies for Produce-To-Order Environments with
Family-Dependent Due-Dates
16.00-16.30 Michiel Jansen The effect of workload constraints in periodic order release models for manufacturing systems
16.30-17.00 Yoni Nazarathy The Variance of Production Counts over a Long Time Horizon

17.00-17.05

Closure

 

 

ABSTRACTS

Dieter Armbruster

Some issues for aggregate models of production systems

We have shown in recent years that describing production systems via a continuum flow leads to models on large scales that can be simulated fast. In this talk we will discuss three issues or extensions of these models: Control at the aggregate level, aggregate models for strongly time-varying inputs and aggregate production lines with finite buffers.

Presentation


Remco Bierbooms

Performance Analysis of Production Lines with Fluid Flows and Finite Buffers

In this talk we consider production lines consisting of a number of machines or machines in series and a finite buffer between each of the machines. The flow of products behaves like a fluid. Each machine suffers from breakdowns, which are for example caused by failures, changeovers, cleaning etcetera. We are typically interested in the throughput of such production lines. Our research is inspired by a production line at Heineken Den Bosch, where retour bottles are being filled and processed by 11 machines in series having different machine speeds. A conveyor belt between each pair of machines is being used for transportation and for buffering in case of machine breakdowns. The number of bottles handled per hour by the machines is very large, which makes it natural to treat the material flows as fluid rather than as discrete items. In the first part of the talk, we explain how to build a simulation model for the Heineken production line using factory floor data.
In the second part we construct an analytical model, which is substantially faster than simulation. The proposed model is an iterative method, where the production line is decomposed into subsystems, each subsystem consisting of an arrival machine, a departure machine and a buffer in between. We model the up- and down behavior of the arrival and departure machines as continuous-time Markov chains, including possible starvation of the arrival machine and possible blocking of the departure machine. The method is compared to simulation for a large test set and for the Heineken production line.

Presentation


Çagdas Büyükkaramikli

Delivery Lead Time and Flexible Capacity Setting for Repair Shops with Homogenous Customers

In this paper, we study a system consisting of a repair shop and its homogenous customers. The customers operate machines that are prone to failure. Upon a failure of a machine, a substitute machine is hired at the market at a certain cost and the failed machine is sent to the shop for repair. We assume that the substitute machine is immediately available and is hired for a specific amount of time, which is decided by the customer. If the time until repair exceeds the hiring duration, the customer looses revenue due to the machine unavailability. The shop’s capacity affects the time until repair and the unit repair costs. Under centralized decision making, system wide optimal shop capacity and the hiring duration can be found. However, in practice, these decisions are taken separately and most of the time the shop capacity decision is unbeknownst to the customers. Therefore, first we design the coordination mechanism under fixed shop capacity and find the iterative decision making process that achieves the system wide minimum cost per repair cycle. Then, in the second part we extend the system by allowing the repair shop to adapt its capacity and to admit failed machines only at equidistant points in time. Time between consecutive points is called a period and it delays the delivery of the failed machine to the shop and creates burstiness in the arrival stream to the shop. We model the unit capacity cost and failed machine transportation costs as well as the delay in machine delivery as a function of period length and with the coordination mechanism developed in the first part; investigate the improvements in total costs in this periodic admission system. The more flexible use of shop’s capacity and the efficiency in the transportation of the failed machines make the periodic admission policy an interesting alternative.

Presentation


John Buzacott, York University, Canada

What’s the Use of Stochastic Models of Manufacturing Systems?

This talk addresses the question of why stochastic models are useful and relevant in designing and operating manufacturing systems. Investing in advanced manufacturing systems requires a major commitment by the firm. While simulation is a well established tool for design, stochastic models provide valuable guidance in the rapid comparison of key features of different alternatives. They can help designers avoid major mistakes because they expand the designer’s intuitive understanding of what determines system performance. The talk will outline the range of manufacturing systems and issues for which stochastic models exist and illustrate their use for understanding these issues. The talk will also show how stochastic models can provide insight on important issues relating to the overall manufacturing strategy of the firm.

Presentation


Stéphane Dauzère-Pérès, Ecole des Mines de Saint-Etienne, France

Modeling and optimizing operations in semiconductor manufacturing

This talk aims at covering various research work conducted at different decision levels with semiconductor manufacturers in the last six years. A first global overview of the main characteristics of semiconductor manufacturing will be presented, together with the three problem types discussed in the sequel. The first problem is related to capacity planning, and in particular qualification management. We will show how we model flexibility, and its impact on manufacturing performances. The second problem is related to detailed scheduling. The presentation will focus on a specific workshop, and summarizes how scheduling decisions were modeled and optimized. The last problem is related to the detailed simulation model of a semiconductor manufacturing facility, including automated transportation, production and storage. The features of the model will be presented, together with the current work on optimizing key parameters. The talk will end with more recent research avenues, and provide some perspectives on the related challenges.

Presentation


Nico van Dijk, University of Amsterdam, The Netherlands

Joint work with Nikky Kortbeek, University of Twente

Lessons from an OT-ICU system for manufacturing

Motivated by the interaction between an Operating Theatre (OT) and an Intensive Care Unit
(ICU) in hospitals, a simple finite production line structure is considered. A closed form approximation and performance bound are concluded. Next to the practical value for
hospitals by itself the results contain a number of lessons to analyze unsolvable
production line systems.


Nicky van Foreest, University of Groningen, The Netherlands

Order Acceptance Policies for Produce-To-Order Environments with Family-Dependent Due-Dates

In this talk we consider a produce-to-order production environment in which a single bottleneck machine produces one product family at a time and is subject to significant switch-over times when the product family changes. Orders are to be delivered on-date to customers and the lead times of orders are family dependent. An acceptance and scheduling policy determines whether an arriving order is to be accepted into a production schedule, or rejected upon arrival. The aim of the production system is to find a policy that maximizes the long run expected reward per arriving order. We use a new Markov Decision Process (MDP) based approach to find a threshold type of policy that is near optimal and easy to implement. First, by means of an MDP approach we find an optimal policy which is shown to be difficult to implement. Second, a new approach is used to analyze the structure of the optimal policy and to find a threshold type of policy that is easy to implement and nearly optimal under a wide range of parameter settings, including product family asymmetries in arrival rate, job size, job reward, and due-date.

Presentation


John Fowler, Arizona State University , USA

Healthcare Systems Engineering

Modern industrial engineering, systems engineering, operations management, and operations research methods hold significant promise for health care systems and quality of care research. Among the most promising methods are queuing theory, optimization and process simulation. This presentation will utilize cases to demonstrate application of engineering and operations management principles and tools to improve health care systems.

Presentation


Michiel Jansen,  TU/e IE&IS

The effect of workload constraints in periodic order release models for manufacturing systems

We study the relation between workload constraints and planned lead times in periodic order release models for manufacturing systems. We consider manufacturing systems where periodic throughput is limited and subject to uncertainties. Workload constraints are necessary to ensure that the manufacturing system can meet the planned lead time. Due to the periodic interaction with the manufacturing system, workload constraints may also lead to additional idling of resources which reduces the effective capacity of the manufacturing system. We analyze a stylized model of a manufacturing system with a single server and two queues in series: an admission queue and a work-in-progress (WIP) queue. Periodically, jobs from the admission queue are released to the WIP queue such that the number of jobs in WIP and in service does not exceed the workload constraint. We present a simple formula for the maximum utilization rate of such a system (one that permits stable queue-lengths) and show that it is generally less than one. We characterize the queue-length distribution by its generating function and give its first two moments. We also propose a numerical procedure for obtaining the steady state probabilities of the underlying DTMC. We compare these results to a model of a manufacturing system that does not include a workload constraint and observe that a workload constraint particularly leads to a significant increase in expected queue-length if it is close to or less than the expected potential throughput in a period (which is the setting that is oftentimes chosen in mathematical programming models for production planning). Furthermore, using these results, we illustrate the relation between workload constraints and planned lead times for various cost functions involving inventory and tardiness costs and conclude that a planned lead time of two periods is often the best choice.We study the relation between workload constraints and planned lead times in periodic order release models for manufacturing systems. We consider manufacturing systems where periodic throughput is limited and subject to uncertainties. Workload constraints are necessary to ensure that the manufacturing system can meet the planned lead time. Due to the periodic interaction with the manufacturing system, workload constraints may also lead to additional idling of resources which reduces the effective capacity of the manufacturing system. We analyze a stylized model of a manufacturing system with a single server and two queues in series: an admission queue and a work-in-progress (WIP) queue.  Periodically, jobs from the admission queue are released to the WIP queue such that the number of jobs in WIP and in service does not exceed the workload constraint. We present a simple formula for the maximum utilization rate of such a system (one that permits stable queue-lengths) and show that it is generally less than one. We characterize the queue-length distribution by its generating function and give its first two moments. We also propose a numerical procedure for obtaining the steady state probabilities of the underlying DTMC. We compare these results to a model of a manufacturing system that does not include a workload constraint and observe that a workload constraint particularly leads to a significant increase in expected queue-length if it is close to or less than the expected potential throughput in a period (which is the setting that is oftentimes chosen in mathematical programming models for production planning). Furthermore, using these results, we illustrate the relation between workload constraints and planned lead times for various cost functions involving inventory and tardiness costs and conclude that a planned lead time of two periods is often the best choice.


Ger Koole, VU University Amsterdam, The Netherlands

Process optimization in hospitals

In this talk we give an overview of health care logistics, with a focus on the differences with manufacturing. We present a number of cases, concerning nursing wards, shared resources such as scanners, and sojourn time reduction in an emergency department.

Presentation


Erwin van der Laan, Erasmus University, The Netherlands

Managing uncertainty in hybrid manufacturing/remanufacturing inventory systems

Product recovery has gained considerable interest as of late as it can contribute to both economical and ecological performance. It is well-known that product recovery comes with considerable uncertainties with respect to quantity, timing and quality of recoverable products.
This talk focuses on the implications of those uncertainties in managing hybrid manufacturing/remanufacturing inventory systems. In such a system, recoverables can be remanufactured into a product that is as good as new and therefore can be resold on the market of new products. Not all recoverables, however, can be remanufactured as this depends on their quality. The fraction of demand that cannot be satisfied through remanufactured products are serviced through the manufacture of new products. Therefore, a careful balancing act in terms of serviceable and remanufacturable inventories needs to be in place in order to optimize inventory related costs.

Presentation


Hubert Missbauer, The University of Innsbruck, Austria

Order release planning based on stochastic models of manufacturing systems

Planning and control systems for manufacturing systems and supply chains are often structured hierarchically. The top (upper) level performs medium-term planning of the material flow that coordinates the production units (e.g., component manufacturing, assembly), and the base (lower) level performs detailed scheduling of the work orders within the production units. The essential interface between these levels is order release. Therefore production planning models that support the coordination task of the top level usually are order release planning models.
Order release planning models require reliable flow time estimates in order to obtain realistic planned lead times for the work orders. Queueing-theoretical insights indicate that flow times are highly load-dependent, which must be considered in order release planning models. Simultaneous determination of order releases and (variable) lead times for a specified planning horizon is an important research topic.

Starting from a practical case, the presentation briefly describes past research efforts by the author and then critically evaluates clearing function models as one approach to solve this problem. Then it describes two alternative ways to improve clearing function models based on the theory of transient queueing systems. It discusses research issues that aim at designing order release planning models with load-dependent lead times that are theoretically consistent.

Presentation


Yoni Nazarathy, EURANDOM, The Netherlands

The Variance of Production Counts over a Long Time Horizon

Consider a production process. Once the throughput is known, knowledge of the variability can allow managers to allocate storage and transportation resources more effectively. This has led some researchers over the past years to develop computational models for assessing the variability of complex production systems. Since the variance of the number of items produced during the time interval [0,t] typically grows linearly for large t, a natural quantity of interest is the asymptotic variance rate - this is the asymptotic slope of the variance of the number of items produced during [0,t].
Quite surprisingly, the asymptotic variance of more elementary models such as the single server queue has not received attention until recently. For such models it is quite clear that the asymptotic variance rate is determined by the arrivals for under-loaded systems and is determined by the services for over-loaded systems. In the case of critically loaded systems both the arrivals and services play a role. In this case, we have discovered that the asymptotic variance rate is often reduced by a factor of about 30% compared to the average of the arrival and service processes!!! We call this phenomena BRAVO - Balancing Reduces Asymptotic Variance of Outputs. While it is so far proved only for single station queuing models, we believe that this phenomenon holds in great generality and should thus be made known to managers of production systems and supply chains.
This presentation is based on some joint works with Ahmad Al-Hanbali, Yoav Kerner, Michel Mandjes, Ward Whitt and Gideon Weiss.

Presentation


Oliver Rose, Technische Universität Dresden, Germany

An approach to model production systems with SysML

In this talk, we present an approach for developing a simulation-tool-independent description of production systems and how to convert such a general model into simulation-tool-specific models. Our aim is to develop production models by means of SysML (Systems Modeling Language) and to build converters from SysML models to a large variety of simulation tools.
To understand the specifics of modeling production systems we interviewed experts, studied current literature and conducted a market analysis of simulation modeling tools. Models for production systems can be divided into a structural, a behavioral and a control part. Based on this architecture, we develop a general model description for discrete processes in production which permits to create comprehensive production scenarios.
In addition, we tested whether SysML is appropriate to build our general model. We present first thoughts on how to improve the usability of SysML for system engineers. We decided to develop our own modeling tool which can be customized based on the requirements of system engineers and planners for production systems. After outlining the theoretical concept for building production models with SysML, we discuss a practical approach for automated model generation for a few simulators from SysML models.


J. George Shanthikumar, Purdue University West Lafayette

Queueing Systems Modeling with Operational Statistics

Presentation


Reha Uzsoy, North Carolina State University

Production planning with resources subject to congestion

While the use of optimization models for production planning dates back to at least the 1950s, the ability to represent the nonlinear relationships between work releases and critical performance measures such as lead times has remained a longstanding problem. We review alternative approaches that have been used in the literature, and then suggest a new approach using nonlinear clearing functions that represent the relationship between the expected throughput of a production resource in a planning period, and the expected work in process inventory level at the resource over that period. Extensive computational experiments show that the proposed approach produces production plans which are more realistic in terms of the production system's ability to execute them. We compare the performance of clearing function models to those from iterative LP-simulation methods, and discuss a number of extensions including incorporation of stochastic demand.

Presentation


Casper Veeger, Tu/e, The Netherlands

Aggregate modeling in semiconductor manufacturing using effective process times

In semiconductor manufacturing, model-based performance analysis is becoming increasingly important due to growing competition and high capital investment costs. A detailed simulation model of a manufacturing system may be helpful in performance improvement activities, but requires considerable development and maintenance effort. Alternatively, an analytical model can be used. Analytical models are fast to evaluate though incorporation of all relevant factoryfloor aspects is difficult. To reduce the number of factory-floor aspects that have to be modeled explicitly, we investigate the use of an aggregate model to represent the manufacturing system. The aggregate model is a simple analytical or discrete-event simulation model with only a few parameters, such as the mean and the coefficient of variation of an aggregated process time distribution. The aggregate process time lumps together all the relevant aspects of the considered system. We refer to this aggregate process time as the Effective Process Time (EPT). In the presentation, the concept of EPT-based aggregate modeling will be explained first. Next, we present an EPT-based aggregate model that we have recently developed. The model is a single-server representation of a workstation. The EPT distribution of the single server is estimated from arrival and departure events measured at the workstation being modeled. In addition, the order in which lots travel through the system is incorporated in the aggregate model. Finally, it is discussed whether the single-server aggregate model can be used to model entire networks.

Presentation


Fabian Wirth, University of Würzburg, Germany

Structure preserving model reduction for logistic systems

In modern logistics applications large scale networks are increasingly important, for instance due to the emergence of transnational supply chains. Due to their size analysis or simulation of such networks becomes expensive to such an extent that is is desirable to derive approximate models with lower complexity, that retain important structural properties of the complex network.
In this talk we will discuss techniques of model reduction that help in preserving important structural components of a given logistics network while reducing the complexity to arrive at models that are amenable to analysis. The techniques rely on results from queueing theory, which provides the reference framework for the modelling of logistics networks and ideas from graph theory related to ranking schemes.

Presentation


Last updated 12-11-2010,
By PK

  P.O. Box 513, 5600 MB  Eindhoven, The Netherlands
tel. +31 40 2478100  fax +31 40 2478190  
  e-mail: office@eurandom.tue.nl