Search

We study the scheduling of production in a multi-item, failure-prone machine with setup times, with the objective of minimising long-run average inventory and backlog costs. We make two contributio...

In several manufacturing systems found in the automotive, food and semiconductor industries, product quality or value deterioration due to excessive residence time in the system are observed. This phenomenon generates defective or low value products, thus undermining the performance of these systems. In this paper, we develop an exact analytical method for calculating the lead time distribution of closed loop systems and present an heuristic to predict the peaks of this distribution. Numerical results show previously uninvestigated behavior of closed loop manufacturing systems under lead time constraints and provide insights on the design of these complex systems.

In this paper, we study a manufacturing system with downward substitution in sales, i.e., selling superior products as inferior ones when the demand for the latter cannot be met immediately, in order to reduce the backlog cost. However, downward substitution in sales also causes profit loss. Therefore, controlling a manufacturing system with downward substitution in sales in order to minimize total production cost is a problem worthy of investigation. In this paper, we focus on the production and sale control problem for a two-product-type manufacturing system with downward substitution in sales and time-delay between the releasing of raw materials and the output of final goods. The system is decomposed into two subsystems. In the first subsystem, the state variables are total production surpluses (i.e., the sum of work-in-process and final goods surpluses), and a hedging point policy with a new structure is constructed to control the raw material releasing rates of the two product-types. In the second subsystem, the state variables are the final goods surpluses, and a new switching curve policy is constructed to control the downward substitution rate in sales. Numerical experiments show that downward substitution in sales contributes to the reduction of total production cost, especially when the time-delay is large.

Buffer space allocation is an important step in production line design. In this paper, we focus on maximising the profit rate of a line subject to a production rate constraint. We describe a newly observed property of production line optimisation. The property is that the production rate constraint, if it is effective, allows an original line to be decoupled into several short lines for optimisation. An approximation method is developed from this property. Instead of optimising a long line, the method divides it into several short lines, optimises them separately and combines their optimal buffer distributions to find the optimal or near optimal buffer distribution of the original line. The method greatly improves the computation efficiency for solving buffer allocation problem for long lines, while ensuring the accuracy of the optimal buffer distribution. A heuristic explanation is proposed. Numerical experiments are provided to show the accuracy and efficiency of the method. The effect of the number and...

The special issue on advances in stochastic models of manufacturing and service operations presents state-of-the art research results in the area of stochastic models for the analysis, design, coordination, and control of manufacturing and service system operations. Although the title includes both “manufacturing” and “service” operations, the main emphasis is on manufacturing. The term “service system operations” is intended to refer mainly to functions that are supportive of manufacturing system operations. The volume includes thirteen state-of-the art articles that can be classified in three categories: stochastic modeling and analysis for design and optimization of manufacturing systems; production planning, management, and control; and analytical approaches to supply chain management.

We examine cumulative net energy use and cumulative net CO2 emissions associated with the development of photovoltaics (PVs) on a global scale. The analysis is focused on the performance of five countries with the largest installed PV capacities—Italy, Japan, Germany, Spain, and the United States—and on the aggregate values for the world (23 countries). The historical record shows that, during the past 19 years of development, the installed base has grown to 64 GW, with an average annual growth rate of almost 40%. During that period the manufacturing and use of photovoltaics has led to a cumulative net consumption of approximately 286 PJ of energy, and cumulative net emissions of 34 Mt of CO2 as a result of a considerable payback time. While energy/CO2 payback time is not unique to PV systems, it plays a larger role in the development of new energy systems than other low-carbon systems. PV energy/CO2 payback time decreases with the following measures: installation of PVs in locations with a large PV potential and high CO2 emissions of the electricity replaced, manufacturing PVs at locations with low CO2 emissions/kW h of electricity used in the production, recycling PVs, and increasing PV conversion efficiency. The analysis is therefore extended into the future for three scenarios with different maximum capacities of photovoltaics (20%, 50%, and 100% of total electricity production). In these scenarios, cumulative net CO2 emissions can be reduced by 4%, 9%, and 18%, respectively, over the long term (by the year 2050). Short-term CO2 increases during growth versus long-term CO2 reduction present a trade-off in developmental growth strategies.

This article focuses on analytical models of two-machine one-buffer Markov lines including waste production. The aim is to compute the probability of producing good parts, referred to as effective efficiency, when waste production is related to stoppages of the first machine. This problem is common in industrial fields where parts are generated by a continuous process; e.g., in high-speed beverage packaging lines. Two innovative models including waste production are presented: the WP-Basic Model extends the model of a basic two-machine–one-buffer transfer line; the WP-RP Model extends the model of a two-machine–one-buffer transfer line with a restart policy operating on the first machine (i.e., when the first machine is blocked because the buffer is filled, it is not allowed to resume production until the buffer becomes empty). The two existing models are improved by distinguishing, at any time step the first machine is operational, whether it is producing a good or a bad part. The probabilities of the sy...

Interest in recycling has surged due to increasing material costs, environmental concerns over material production and disposal, and laws designed to improve material recycling rates. In response, recycling systems are becoming more complex as increasing material recovery is required from products with complicated material mixtures such as waste electrical and electronic equipment (WEEE) and ELVs. To increase performance and process complex material mixtures, separation systems are typically organized as highly integrated multistage systems. However, the problem of estimating the performance and designing multistage separation systems has rarely been tackled from a systems engineering perspective, resulting in poor integration and suboptimal configuration of industrial multistage separation systems. This paper presents a new approach to modeling, analyzing, and designing multistage separation systems to meet specified performance goals in terms of recovery/grade. Results can be used to generate maps of optimal system configurations for different requirements. The industrial benefits are illustrated by a real case study.

In some kinds of manufacturing, the time a part may spend in a buffer between successive operations (such as cleaning and baking in semiconductor fabrication) is limited. Parts that wait longer than this time must be reworked or discarded due to the risk of quality degradation. This constraint is important in several industries, especially the semiconductor industry. To study this issue, we present an analytical formulation for the part waiting time distribution in a Buzacott two-machine one-buffer transfer line. The numerical solution is tested with Little's law. Numerical experiments are provided to illustrate the accuracy of the solution as it is compared with simulation.

This paper deals with our previously introduced Hedging Zone Policy (HZP), which is a simple-to-implement, well-behaved heuristic for scheduling machine setup changes in real time. In the HZP, a pair of parameters for each product type defines a set in surplus space called the hedging zone, and a different changeover rule is implemented depending on the location of the production surplus state with respect to this zone. When the surplus state lies outside of the hedging zone, a fixed rank ordering or prioritization is used for deciding which item to produce next. When the surplus state is inside the hedging zone, the policy may either produce the current part type at its demand rate until more demand builds up, or change over to the item that is furthest behind (in an appropriate sense). After providing some background and overview of the HZP, we present new results concerning its stability properties. In particular, we show that if a deterministic counterpart of the stochastic system satisfies an easy-to-evaluate condition, the stochastic system with exponentially distributed failure and repair times is rate stable.

We consider a two-stage manufacturing system composed of a batch processor and its upstream processor. Jobs exit the upstream processor and join a queue in front of the batch processor, where they wait to be processed. The batch processor has a finite capacity Q, and its processing time is independent of the number of jobs loaded into the batch processor. In certain manufacturing systems (including semiconductor wafer fabrication), a processing time window exists from the time instance the job exits the upstream processor till the time instance it enters the batch processor. If a job is not processed before reaching the end of its processing time window, job rework or validation is required. We model this drawback by assigning a reward R for each successfully processed job by the upstream processor, and a penalty C for each job that reaches the end of its processing time window without being processed by the batch processor. We initially assume an infinite job source in front of the serial processor and also assume that the batch processor is operated under a threshold policy. We provide a method for controlling the production of the serial processor, considering the processing time window between the upstream processor and the downstream batch processor. We then show how the serial processor control policy can be modified when the serial processor also experiences intermittent job arrival.

We examine the problem of minimizing the mean cycle time of a batch processor with incompatible job families and future job arrivals. Batch processors frequently found in wafer fabrication include oxidation and diffusion ovens. When the problem is deterministic with a fixed number of job arrivals, the problem is NP-hard; an optimal dynamic program formulation is developed and empirically tested on small problem instances. For larger problem instances with possibly infinite job arrivals, an online heuristic based on model predictive control (MPC) is proposed. The proposed heuristic has two parameters: the number of job families considered f, and the number of future job arrivals L. Statistical analysis of simulation results shows that the MPC-based heuristic has significantly lower mean cycle time than a benchmark look-ahead method under the assumption that the arrival time is deterministic and job families of consecutive job arrivals are uncorrelated. These observations also hold true when the predicted arrival times can be erroneous, indicating that the suggested MPC-based heuristic is as robust as the benchmark with respect to errors in forecasted arrival times. When the job families of future job arrivals are positively correlated, the mean cycle time of jobs passing through the batch processor is significantly reduced, regardless of control policy. We infer that the introduction of correlation generates less improvement for policies that foresee events longer into the future. For this reason, when the correlation is high, the MPC-based heuristic, which considers events that occur farther into the future, may have worse performance than the benchmark. The results suggest two ways to improve the performance of the batch processor. First, we can use a heuristic that has better mean performance, which may entail much larger computational costs. Another option is to continue using a simple look-ahead policy at the batch processor, and focus attention on controlling the upstream processors, such that the batch processor arrival process exhibits correlated job families between consecutive arrivals. This introduction of correlation causes a significant reduction of cycle time regardless of system parameter and batch processor control policy.