Operations Sequencing for a Multi-stage Production Inventory System Junmin Shi * Xiaohang Yue † and Yao Zhao ‡ July 2013 Abstract This paper studies operations sequencing for a multi-stage production inventory system with lead times under predictable (deterministic) yield losses in the presence of uncertain demand, with full or partial release of work-in-process (WIP) inventories, for pre- or post-operation cost structures, and under the total discounted or average cost criteria. We derive necessary and sufficient criteria for the optimal sequence of operations for each of these cases. While the criteria differ in their obtained necessary and sufficient conditions, they all imply the same principal: those operations with (1) lower yields, (2) lower processing costs, (3) longer lead times, and (4) lower inventory holding costs should be placed upstream in the system. Key words: operations sequencing, yield loss, multi-echelon inventory system, lead times. 1 Introduction Product and process design/redesign has been viewed as a powerful means to increase the flexibility of the process and to lower the supply chain costs (Lee and Tang 1998). In this paper, we study the operations sequencing deci- sion under yield losses. Our work is motivated by an operations sequencing problem observed in the mechanical industry. Specifically, the mechanical industry typically involves machining and heating operations, where ma- chining is done to form the shape of a product for functionality, and heating is done to improve surface integrity such as hardness and residual stress for high quality. Generally heating has a much lower yield than machining, as machining is a well-controlled process (i.e., dimensional tolerance up to 1/1000th of an inch is often obtainable * Department of Managerial Sciences, Robinson College of Business, Georgia State University, Atlanta, GA. [email protected]† Sheldon B. Lubar School of Business, University of Wisconsin-Milwaukee, Milwaukee, WI. [email protected]‡ Dept. of Supply Chain Management and Marketing Sciences, Rutgers University, Newark, NJ. [email protected]
22
Embed
Operations Sequencing for a Multi-stage Production ...jshi/Shi-Yue-Zhao-2013.pdf · Operations Sequencing for a Multi ... This paper studies operations sequencing for a multi-stage
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Operations Sequencing for a Multi-stage Production Inventory System
Junmin Shi∗ Xiaohang Yue† and Yao Zhao‡
July 2013
Abstract
This paper studies operations sequencing for a multi-stage production inventory system with lead times
under predictable (deterministic) yield losses in the presence of uncertain demand, with full or partial release of
work-in-process (WIP) inventories, for pre- or post-operation cost structures, and under the total discounted or
average cost criteria. We derive necessary and sufficient criteria for the optimal sequence of operations for each
of these cases. While the criteria differ in their obtained necessary and sufficient conditions, they all imply the
same principal: those operations with (1) lower yields, (2) lower processing costs, (3) longer lead times, and (4)
lower inventory holding costs should be placed upstream in the system.
Product and process design/redesign has been viewed as a powerful means to increase the flexibility of the process
and to lower the supply chain costs (Lee and Tang 1998). In this paper, we study the operations sequencing deci-
sion under yield losses. Our work is motivated by an operations sequencing problem observed in the mechanical
industry. Specifically, the mechanical industry typically involves machining and heating operations, where ma-
chining is done to form the shape of a product for functionality, and heating is done to improve surface integrity
such as hardness and residual stress for high quality. Generally heating has a much lower yield than machining,
as machining is a well-controlled process (i.e., dimensional tolerance up to 1/1000th of an inch is often obtainable
∗Department of Managerial Sciences, Robinson College of Business, Georgia State University, Atlanta, GA. [email protected]†Sheldon B. Lubar School of Business, University of Wisconsin-Milwaukee, Milwaukee, WI. [email protected]‡Dept. of Supply Chain Management and Marketing Sciences, Rutgers University, Newark, NJ. [email protected]
without much difficulty). On the other hand, heating, which involves heating and cooling rate control as well as
maintaining the desired temperature on specific sections of a part, is harder to control than machining. Traditional-
ly, machining precedes heating for surface hardening because machining a hard surface is difficult. However, with
the recent development of a new technology, hard turning, heating can now be done before machining in many
circumstances. Thus, it is of our interest to find out which sequence of operations is better in terms of operational
efficiency.
Yield loss is not unique to the mechanical industry, but common in electronic & computer, food processing and
chemical industries. The issues of operations sequencing under yield losses widely exist in practice as operations
in these industries can often be re-sequenced or reorganized. Examples can be found in Lee and Tang (1998), Bard
and Feo (1989) and Schraner and Hausman (1997). In this paper, we consider a multi-stage production-inventory
system under deterministic yield losses where work-in-process (WIPs) inventories are either fully or partially
released to downstream stages. Our objective is to determine the optimal operations sequence that minimizes the
system-wide production and supply chain cost.
While yield losses are likely random, a predictable (or deterministic) yield can be a good approximation in
many practical applications. For instance, in many production systems of the mechanical industry, the variation of
yield rates in many operations, e.g., drilling, sawing, heating, etc., may have stable yield losses so that the yield
can be approximately treated as being predictable. In industries with batch productions, such as food processing
and chemical, the yield variability of a batch can be negligible due to the law of large numbers if the batch size is
sufficiently large. In such a case, it is reasonable (and often the practice) to assume a predictable yield loss.
Operations sequencing under deterministic yield losses in multi-echelon inventory systems is not only an
untapped problem in the literature but also a problem worthy of study. Although one can redefine the flow units
along a production line by multipliers representing the impact of yield losses, the flow units may be different for
different operations sequences, and thus the impact of the yield losses on the operations sequence, compounded
with production costs, lead times and inventory carrying costs of different operations, is yet known. This study is
particularly important and meaningful for today’s lean operation system design. We believe this study captures a
generic real-world situation in many industries, and our model makes practical sense in operations management.
This paper is related to the literature of operations re-sequencing/re-designing, production-inventory systems
with yield losses, R&D testing sequence, and deteriorating job scheduling (DJSP). We shall review each stream of
the literature below and point out the difference and our contribution.
In the operations re-sequencing/re-designing literature, Lee and Tang (1998) address the problem of variabil-
2
ity reduction by changing the sequence of operations, known as operations reversal. They build a probabilistic
model to derive conditions under which such a change is desirable. Kapuscinski and Tayur (1999) consider the
same problem but using standard deviation. Jain and Paul (2001) generalize the operations reversal model of Lee
and Tang (1998) to incorporate two characteristics of fashion goods markets: heterogeneity among customers and
unpredictability of customer preferences. Schraner and Hausman (1997) study the production operations sequenc-
ing problem for a single-product and multi-stage system using base-stock policies. They show that resequencing
decisions can be made using a cost-time profiling. Gupta and Krishnan (1998) consider a problem where the ob-
jective is to determine assembly sequences for a family of products. They provide an algorithm to generate generic
sub-assemblies by taking into account inventory costs at an aggregate level. Yan, Sriskandarajah, Sethi and Yue
(2002) study the impact of the process redesign (i.e., operation re-sequencing, operation merging) on the safety
stock levels in a supply chain. They demonstrate that process redesign could have a significant impact on the
safety stock investment. In this stream of literature, however, yield loss is not considered and the impact of yield
on operations sequencing is not studied.
Yield losses have been studied extensively in the literature of production-inventory systems, see Yano and Lee
(1995) for a detailed review. There have been two streams of literature on this subject. One stream focuses on
the single-stage production systems with yield losses (e.g., Henig and Gerchak 1990, Wang and Gerchak 1996,
Gurnani, Akella and Lehoczky 2000). The other stream (e.g., Yano 1986, Lee and Yano 1988, Wein 1992, and
Lee 1996) focuses on multi-stage but single-period systems with yield losses. The inventory control problem of a
general multi-echelon production-inventory system with random yield is well known as a notoriously hard problem
(Grosfeld-Nir, Anily and Ben-Zvi 2006) and remains to be unsolved in the literature. Grosfeld-Nir and Gerchak
(2004) discuss a produce-to-order system consisting of multiple machines (stages or work centers) in series, and
show that the machines should be arranged in an increasing order of a ratio, which suggests to put those operations
with higher processing-succuss rate and larger variable production cost downstream. However, their setting of the
model is different from ours. Specifically, they consider a produce-to-order system where inventory is unnecessary,
and assume multiple lotsizing and rigid demand, i.e., each demand should be satisfied fully.
In the R&D literature, a related problem is the “least-cost testing sequence”, in which a project needs to pass
multiple tests to be completed and can fail in each of them. In a simple single-project model, Boothroyd (1960)
shows that the tests with lower costs and higher failure rates should be conducted earlier. We refer the reader to
Schmidt and Grossmann (1996) for a more recent review of this literature. This literature typically focuses on
a single project. In contrast, our paper considers the operations sequencing problem in multi-echelon inventory
3
systems with yield losses which faces massive and random demand and must take both production and inventory
cost into account. Although the testing sequence problem is drastically different from the operations sequence
problem, their optimal (least cost) sequences surprisingly bear much similarity, e.g., lower cost and lower yield
tests (operations) should be conducted earlier. Of course, the operations sequencing problem is significantly more
complex than the testing sequence problem due to its dynamic nature, the lead time, the more complicated cost
structure and non-consecutive operations, which lead to many novel results.
The deteriorating job scheduling problem (DJSP) is to schedule a set of jobs in which the processing times of
the jobs are not constant but increasing over time, i.e., deteriorating. For example, the temperature of an ingot,
while waiting to enter the rolling machine, drops below a certain level, requiring the ingot to be reheated before
rolling. Another example is fire-fighting, i.e., the time and effort required to control a fire increases if there is a
delay in the start of the fire-fighting effort. DJSP has been extensively studied in literature. For some recent work,
please see Browne and Yechiali (1990), Cheng and Ding (1998), Lee (2004), Cheng, Wu and Lee (2008), Wang
and Cheng (2008), Wu, Shiau and Lee (2008). The DJSP literature focuses on a scheduling problem with the
objective of minimizing the makespan, total completion time, and/or maximum lateness where yield losses and
inventory control issues are not considered.
This paper complements these streams of literature by studying operations sequencing issue in a multi-echelon
production-inventory system under yield losses, and quantify the impact of yield, lead time, production and inven-
tory costs on the optimal operations sequence in a variety of circumstances:
• Pre- v.s. Post-operation cost structure: There are two ways to account for the costs incurred at an op-
eration subject to yield losses (Henig and Gerchak 1990): pre-operation cost structure calculates the pro-
cessing/production costs based on the input batch size to an operation while post-operation cost structure
calculates these costs contingent on the output batch size from an operation.
• Full release v.s. partial release of work-in-progress (WIPs) inventories. Under full release of WIPs, we
assume that all the WIPs are fully released to the downstream stage for processing without being held in
inventory. In contrast, under partial release of WIPs, some WIPs may be deliberately held at each stage for
some time before releasing to the downstream. In this case, the inventory holding cost of these WIPs must
be considered while sequencing the operations.
In addition, we consider interchanging consecutive and non-consecutive operations, and the total discounted and
long-run average cost criteria.
4
The objective of operations sequencing/re-sequencing is to take advantage of flexibility in process design to
improve supply chain efficiency (Schraner and Hausman 1997). Our research contributes to the literature in the
following aspects:
• First, we complement the extant operations management literature by studying the unexplored issue of op-
erations sequencing in a dynamic multi-stage supply chain with yield losses. Specifically, we characterize
the optimal sequence of operations that minimizes the total discounted cost or the long-run average cost.
• Second, our study reveals useful insights for sequencing operations: (i) it is more economical to move
operations with lower yields, lower processing costs, longer lead times, and lower holding costs upstream
if the sequence of operations can be altered. (ii) The optimal sequence of any two operations may depend
on the characteristics of the operations in between them. (iii) Priority should be given to improve the yield
in downstream operations if the investments required for the same yield improvement across operations are
comparable. All these insights would provide important implications and guidelines for lean system design
in supply chain and manufacturing practice. We believe that this study represents a step forward in the
science of flexible/lean manufacturing system design.
In the remainder of this paper, we shall consider the system with full release of WIPs under the pre-operation
cost structure in §2, the same system under the post-operation cost structure in §3, and the system with partial
release of WIPs in §4. Finally, we conclude the paper in §5.
2 Full Release of WIPs and Pre-Operation Cost Structure
We consider a multi-stage system (either a supply chain or a manufacturing system) consisting of a series of op-
erations each of which is reviewed and controlled periodically. We assume that the operations have effectively
unlimited processing capacity. Each operation has a deterministic yield loss and a lead time, and incurs a variable
processing cost. The demand for the finished goods of this system is random in each period. The system repre-
sents any valid way of production and processing including work centers at various locations, a combination of
production, warehousing and transportation.
The system works according to the following sequence of events: At the beginning of a period, a batch size
of components (i.e., raw materials) is released to the upmost operation of this system. After an appropriate delay
(due to the review period and operational lead time), the upmost operation is done on this batch and the resulting
work-in-process (WIPs) after the associated yield loss is fully released to the next operation (partial release of
5
WIPs is discussed in §4). The same sequence of events takes place at subsequent operations until the last one
where the finished products are stored in inventory to satisfy the demand. Unsatisfied demand is fully backlogged
and will be fulfilled in the future periods.
The series of operations is indexed by m = 1, 2, . . . ,M where operation m immediately precedes operation
m + 1 for all m < M, unless we explicitly mention otherwise. We consider a time horizon of N periods indexed
by n = 1, . . . ,N. Throughout this paper, we apply subscript (superscript) to denote operations (time periods,
respectively). For example, the yield and lead time of operation m are respectively denoted by Rm ∈ (0, 1] and
Lm ≥ 0, and the random demand in period n is denoted by D(n).
Let Q(n)m be the WIPs at stage m at the beginning of period n, Q(n) be the total input of raw materials into the
system at the beginning of period n, and y(n) be the inventory level of finished goods at the end of period n after
satisfying the demand in this period. At the beginning of period n, we observe y(n−1) and Q(n)m . At the end of period
n, demand D(n) is realized and any unmet demand is backordered. The finished goods inventory level and WIPs
after each operation are periodically updated as follows,