Академический Документы
Профессиональный Документы
Культура Документы
For over a decade, the Kellogg Company has used its planning
system (KPS), a large-scale, multiperiod linear program, to
guide production and distribution decisions for its cereal and
convenience foods business. An operational version of KPS, at
a weekly level of detail, helps determine where products are
produced and how finished products and in-process products
are shipped between plants and distribution centers. A tactical
version of KPS, at a monthly level of detail, helps to establish
plant budgets and make capacity-expansion and consolidation
decisions. Operational KPS reduced production, inventory,
and distribution costs by an estimated $4.5 million in 1995.
Tactical KPS recently guided a consolidation of production ca-
pacity with a projected savings of $35 to $40 million per year.
INTERFACES 31:6 2
KELLOGG
adding a new production facility to ex- tributing over 600 skus at about 27 loca-
pand capacity and extend the product line, tions (plants, co-packers, or DCs) with a
and it used KPS to compare the overall total of roughly 90 production lines (in-
cost implications of locating that facility in cluding coaters, puffing towers, and cook-
one existing plant versus another. ers) and 180 packaging lines. Optimizing
We installed KPS in 1990 and developed this many decisions is clearly a formidable
it further over several years. Initially, we task. The production, inventory, and dis-
used only systemwide average costs for tribution activities behind getting a pack-
each plant, basing optimal solutions on age of Kellogg’s Variety Pak to market
differentials in transportation costs and on helps to illustrate.
available production capacities, rather Kellogg’s Variety Pak, sku 05337, con-
than on differentials in manufacturing tains 10 small boxes of different cereals,
costs at the various production sites. This for example Corn Flakes, Rice Krispies,
helped smooth the transition from the and Froot Loops. The individual boxes (in-
then-current, decentralized production- dividuals) can be produced and packed at
planning process to a more centralized
process guided by KPS. In particular, us- Kellogg wants to ship fresh
ing true costs, KPS would have shifted products.
overall production patterns dramatically,
and actually carrying out this shift would Kellogg plants in Battle Creek, Omaha,
have been impractical. By 1994, however, Lancaster, or Memphis. But not all plants
we had introduced true manufacturing produce all individuals. All of the individ-
costs into the operational model, and this uals at one plant compete for packaging
produced savings of $4.5 million in 1995 capacity because they are packed in the
[Scott 1999]. Currently, KPS is in use same size box. Many of the basic produc-
weekly for operational planning and al- tion decisions are independent; for exam-
most daily for dealing with tactical issues. ple, Mini-Wheats and Corn Pops share
Kellogg’s Products and Operations neither materials nor production facilities.
Kellogg operates five plants in the On the other hand, Frosted Flakes are es-
United States and Canada: Battle Creek, sentially coated Corn Flakes, and produc-
Michigan; Memphis, Tennessee; Omaha, tion constraints dictate that not all Corn
Nebraska; Lancaster, Pennsylvania; and Flakes in a production run can be coated.
London, Ontario. It has seven core DCs in Thus, the production of these two prod-
such areas as Los Angeles and Chicago, ucts must be synchronized.
and roughly 15 co-packers that contract to Variety Paks are assembled only in Bat-
produce or pack some of Kellogg’s prod- tle Creek, so Kellogg must coordinate
ucts. Customer demands are seen at the prior production, packaging, and shipping
DCs and at four of the Kellogg plants. In from the other three plants with the final
the cereal business alone, the firm coordi- assembly operations in Battle Creek. Each
nates the production of about 80 products, individual is an intermediate product or
while packaging, inventorying, and dis- constituent sku, which may be inventoried
November–December 2001 3
BROWN ET AL.
INTERFACES 31:6 4
KELLOGG
placed in inventory or are used to meet the p⬘ ⳱ Los Angeles DC. For the most
demand assigned to the plant (Kellogg part, we model a shipment leaving in
plants can act as their own DCs) or are week t to arrive in week t Ⳮ 1.
shipped to another plant or DC. All cus- A DC may be viewed simply as a Kel-
tomer demand is aggregated by sku and logg plant with no production or packag-
location, that is, plant or DC. ing facilities, and thus a DC incorporates
For each week of the 30-week planning only inventory and shipping variables.
horizon, the decision variables associated KPS currently models Kellogg’s own
with a plant are the following: plants and distribution centers and about
MAKEh,p,t —Production of products on a 15 co-packers. Co-packers are non-Kellogg
processing line and other facilities that production facilities under contract to pro-
form a production process, for instance, duce and package products designed by
the klbs (pounds ⳯ 103) of product 015 Kellogg and bearing the Kellogg label or
(Frosted Flakes) produced on processing to produce constituents for Kellogg prod-
ucts that undergo final processing at Kel-
Kellogg runs KPS each logg plants. (Kellogg performs no co-
Sunday morning to guide packing for other companies.) A co-packer
production decisions in week has no exogenous demand assigned to it.
Within a plant, basic constraints for each
2 and beyond.
week require that the system:
line LL01 and frosted with sugar on coater (C1) Does not exceed processing line
LC01 at the p ⳱ Lancaster plant in week . capacities;
The product 015 and two production fa- (C2) Does not exceed packaging line
cilities LL01 and LC01 form the produc- capacities;
tion process h. Every product requires pro- (C3) Packages all products produced in
cessing on one processing line but may a week into skus during that week (these
also consume capacity on other facilities, are flow-balance constraints between pro-
such as a coater. cessing and packaging);
PACKk,m,p,t —Packaging of skus on par- (C4) For each sku, balances inventory
ticular packaging lines, for example, the from the previous week plus current pack-
klbs of sku k ⳱ 00122 packed on packag- aging plus incoming shipments with out-
ing line m ⳱ LP09 at the p ⳱ Lancaster going shipments, plus consumption in as-
plant in week t. semblies if this is a constituent sku, plus
HOLDk,p,t —Inventories of skus, for ex- exogenous demand assigned to the plant;
ample, the klbs of sku k ⳱ 00525 held in (C5) Satisfies safety stock requirements
inventory at the p ⳱ Battle Creek plant at with inventory of each sku at each plant;
the end of week t. (C6) Coordinates processing lines and
SHIPk,p,p⬘,t —Shipments of skus to or packaging lines as needed during each
from other plants and DCs, for example, time period. For example, we may require
the klbs of sku k ⳱ 00525 shipped from that the time spent packaging an sku does
the p ⳱ Battle Creek plant, in week t, to not exceed the time spent processing the
November–December 2001 5
BROWN ET AL.
product from which that sku is derived. —Product and sku codes, product-to-sku
With respect to these constraints, a DC relationships, including recipes for assem-
is just a plant without production-related bled skus, and case weights;
constraints and a co-packer is a just a Kel- —The identities of the processing and
logg plant with no exogenous demand. packaging lines at each plant, the products
Constraints (C1), (C2), (C4), and (C5) are or skus that can be processed or packed
implemented as elastic goals that can be on those lines, nominal yields in klbs per
violated at a price: When an elastic goal
constraint is violated, a linear penalty per For its size, KPS is curiously
unit of violation is assessed. The occur- difficult to solve.
rence of such an elastic violation may indi-
cate that a little overtime is needed, or it shift, nominal processing or packaging
may signal a bottleneck that cannot be cost for each product or sku in dollars per
avoided. Either way, the model recom- klb;
mends a systemwide plan that is opti- —Inventory costs for each type of sku at
mally adjusted to deal with all such prob- each plant in dollars per case per week;
lems over all locations and time periods. —Shipping costs (dollars per case) by lane;
KPS does not model raw materials but and
does model some intermediate products. —Various per-unit penalties for unmet de-
An intermediate product is viewed as a mand, unmet safety stock, and line
constituent sku that can be shipped to overcapacitation.
other plants where it is further processed Data that vary by week are:
or combined and packed with other con- —Production and processing line avail-
stituent skus to create a finished, assem- abilities at each plant measured in shifts;
bled sku. The Variety Pak described ear- —Variations in nominal yields or costs to
lier is one instance of a constituents-to- account for time-of-year effects (for exam-
assembly recipe. Also, semiprocessed Rice ple, it may take longer to dry certain prod-
Krispies, called bumped rice, are produced ucts during humid summer months) or ef-
at one plant and shipped in bulk totes (la- fects associated with new lines, products,
beled and modeled as an sku) to a co- or skus where yields typically improve for
packer to be further processed and packed the first few weeks after commencing
into Rice Krispies Treats. KPS handles the production;
packaging of assembled skus by straight- —Estimated demands, in cases, for skus
forward modifications of the flow-balance at Kellogg plants and DCs based on fore-
and packaging-line constraints: The assem- casts made by the marketing department;
bly and packaging of an assembled sku and
consumes only packaging capacity and —Desired minimum inventory levels
draws constituent skus from inventory or (safety stocks) at demand locations.
concomitant packaging. The basic objective of KPS is to mini-
The basic, time-invariant data for KPS mize the total cost of meeting estimated
are: demands. The full objective function in-
INTERFACES 31:6 6
KELLOGG
cludes penalty terms for violating process- partment’s original forecast. Nonetheless,
ing and packaging capacities, for not the overarching goal of Kellogg, and thus
meeting demand, and for not meeting KPS, is to meet these customer demands.
safety stocks. Perhaps KPS should be a multistage
A fundamental assumption behind a lin- stochastic-programming model that di-
ear program is that each decision variable rectly handles uncertainty in demand, and
may take on any value in a continuous possibly uncertainty in manufacturing
range, but production and packaging deci- yields and line availabilities. But such a
sions at Kellogg (and other manufacturers) model would require an unwieldy amount
are not that flexible. For instance, KPS of data and would be too difficult to solve:
might suggest that about one third of a As a deterministic model, KPS is large
shift of a low-demand sku be produced at and can take several hours to run. A
some plant in each week of the planning stochastic-programming version would re-
horizon, but the plant manager requires a quire orders of magnitude longer to solve.
one-shift minimum for that sku because of So KPS simply uses planned safety stocks,
setup overhead (production time lost be- that is, minimum inventory levels, as a
cause of required equipment adjustments). buffer for uncertain demand. A huge body
Theoretically, it is possible to add binary of literature addresses safety stocks in
variables to KPS to handle such situations, production-inventory models (for instance,
but we have not yet done this because the Silver, Pyke, and Peterson [1998, Chapter
model has been hard enough to solve as a 7] and the 103 references they list), but
simple linear program. (Technology is im- these models typically require strong
proving, however, and a mixed-integer probabilistic assumptions and do not ex-
version of KPS to handle production and tend to multistage, capacitated, produc-
packaging setups is on the drawing tion, inventory, and distribution models
board.) Therefore, managers review KPS- like KPS. KPS uses simple rules for setting
suggested production plans to modify safety stocks that have been tuned manu-
plans that don’t quite fit the realities of the ally over time: Experience is a good
plant floor. teacher in this case.
Meeting Uncertain Demand—Forecasts In KPS, safety stocks for an sku are set
and Safety Stocks only at locations that see demand for that
Much uncertainty is associated with the sku. Nominally, the safety stock for sku s
data for a long-term production-inventory at location p in week t is the sum of de-
model, and for KPS, the greatest uncer- mands there in weeks t and t Ⳮ 1, or
tainty is in actual demands for skus. In the some other function of future demands.
first few weeks of a time horizon, demand However, if an sku is to be promoted in a
numbers may be fairly accurate because special advertising campaign starting in
they are largely based on firm orders from week t, the safety stock in week t is set as
customers. But even at week three or four the sum of estimated demands in weeks t
of the horizon, actual demands may de- through t Ⳮ 4, or some other horizon that
part substantially from the marketing de- is longer than that for an sku not being
November–December 2001 7
BROWN ET AL.
promoted. This extra buffer is kept be- enforcing solution persistence [Brown,
cause the actual demand for a promoted Dell, and Wood 1997]. That is, we require
sku is higher and more variable than for that the solution of KPS covering, say,
one that is not. weeks t through t Ⳮ 29, persist to some
Safety stocks also attenuate undesirable degree with respect to last week’s solution
end effects in KPS. In particular, a cost- covering weeks t ⳮ 1 through t Ⳮ 28. If
minimizing, finite-horizon, production- the time lag in ordering certain materials
inventory model will always try to drive is longer than a week, variables may also
inventories to zero at the end of the plan- be fixed in weeks beyond the first week of
ning horizon. Even in a model with safety the horizon.
stocks, we do not trust a finite-horizon Fixing variables to given values is a
model’s prognostications in the last few strong form of persistence; KPS also uses
time periods; without safety stocks, the less coercive techniques, such as requiring
number of periods of untrustworthy re- a variable to lie within a specific range or
sults would be even greater. penalizing deviations of the variable from
The Rolling Horizon and Solution a target value. For instance, it is common
Persistence for a plant manager, with guidance from
Kellogg uses KPS in setting a rolling ho- KPS, to decide that his plant will pack a
rizon [Schrage 1999, pp. 187–188]: certain sku in week 3, say, of the planning
Multi-period models are usually used in a roll- horizon. However, he will let the model
ing or sliding format. In this format, the model decide (for now) exactly how much of this
is solved at the beginning of each period. The
recommendations of the solution for the first sku to produce above a specified mini-
period are implemented. As one period elapses mum level. We may also view safety-stock
and better data and forecasts become available,
the model is slid forward one period. The pe- levels and penalties for not achieving
riod that had been number 2 becomes number them as a form of solution persistence. In
1, etc., and the whole process is repeated. general, KPS exploits persistence to (1)
KPS has one difference, however: Pro- handle lead times of raw materials, (2) re-
duction and packaging decisions in the duce volatility in suggested production
first week are fixed, and it is largely the and distribution plans as the model hori-
second week’s decisions that are set in zon rolls ahead, and (3) incorporate mana-
motion at the beginning of week 1. The gerial knowledge into the production plan
main reason for this is that it takes time to that is too complicated to model more
get raw materials and packaging materials explicitly.
in place for production, but KPS does not The Tactical Linear Program
model such materials. Thus, at the begin- Even though we originally envisaged
ning of week 1, the production and pack- KPS as only an operational model, we
aging plan is locked in place, having been have also developed a tactical version of
made the week before or earlier along KPS for long-range planning, on the order
with orders for any materials that may not of 12 to 24 months. Kellogg uses long-
have been on hand. range planning to develop plant budgets,
Fixing model variables is one method of investigate capacity-expansion issues, test
INTERFACES 31:6 8
KELLOGG
November–December 2001 9
BROWN ET AL.
—If KPS shows a processing line being flexibility to change the schedule, they re-
heavily utilized for a particular product, view only large deviations from the KPS-
planners will enforce a regular schedule suggested packaging plan for potential
with production every week on that line modification.
using a whole number of shifts; Kellogg runs KPS each Sunday morning
—If KPS shows consistent, low levels of (the end of week 0) to guide production
production on a line, they will aggregate decisions in week 2 and beyond. Most
production into a sequence of larger pro- model variables for week 1 are fixed: By
duction runs, in a whole number of shifts, Sunday, it is too late to make changes to
once every few weeks; and production plans for week 1 because raw
—If KPS shows unmet demand for a par- materials and packaging materials for
ticular sku, planners may schedule week- week 1 are already at the plant or on the
end overtime for production and packag- way.
ing of that sku. Data for the weekly run of KPS come
The scheduled production at each plant, from a variety of sources. Demand data
by product and totaled over the quarter, comprise a combination of forecasts from
usually conforms closely to that suggested the marketing department and firm or-
by KPS. However, KPS might source a ders. Structural data on line capacities,
particular product at Battle Creek rather yields, and the like are averages compiled
over time, with data on new lines taken
When Kellogg completes this from engineering estimates. New or over-
project, it estimates the hauled lines will have start-up curves as-
savings will be between $35 sociated with them. That is, yields will im-
prove for several weeks as operators gain
and $40 million annually.
experience with the lines and product
than Omaha because of a very small cost changeovers become smoother. Additional
difference, a difference that planners real- data come from conferences with plant
ize is negligible. In this case, planners managers: For example, a packaging line,
might schedule Omaha for that product to BP25, at Battle Creek unexpectedly may be
create a more flexible or balanced schedule scheduled for maintenance in weeks 3
for the plants or to make the schedule through 5, or a run of at least 200 klbs of
more closely follow budgetary guidelines product 123 must be made in week 4 be-
established months before using tactical cause raw materials are reaching age lim-
KPS. its and must be processed, or the yield on
As time passes, planners compare the processing line OL01 at Omaha must be
schedule against a weekly run of KPS in reduced in weeks 2 through 4 because of
order to adjust process quantities and tim- projected humid weather.
ing and to identify any approaching risks A typical model has roughly 100,000
of unmet demand; planners may look sev- constraints, 700,000 variables, and 4 mil-
eral months ahead. In the shorter horizon lion nonzero coefficients. It is solved with
of four to six weeks, when they have little the X-System [Brown and Olson 1994] in
INTERFACES 31:6 10
KELLOGG
two to four hours on a DEC Alpha com- differentials. After planners became com-
puter with 512 megabytes of RAM or in fortable with KPS and verified data, they
less than 20 minutes on a 500 MHz Pen- introduced actual costs into the model. At
tium III laptop. For its size, KPS is curi- that time, 1994, it was estimated that sav-
ously difficult to solve, and we have done ings of $4.5 million per year accrued from
research to find out why. Kellogg’s system following the model’s production, inven-
has scant slack capacity; small changes in tory, and distribution recommendations
plans affect many facilities and time pe- [Scott 1999].
riods; and about 70 percent of the model’s Tactical KPS in Action
constraints are taut at optimality. These KPS is just as important for tactical
are tough linear programs. planning as it is for operational planning.
One key to solving KPS efficiently is the Some representative examples demon-
X-System’s generalized-network factoriza- strate its value in the tactical arena.
tion [Brown and Olson 1994; McBride Prior to the start of each fiscal year,
1985]. In particular, a selection of up to 95 planners populate the KPS database with
percent of constraints C1, C3, and C4 will estimated plant-cost and throughput data
have at most two nonzero elements per and forecasted demands for the fiscal year
column and thereby form a large plus six months. We run the model to de-
generalized-network submodel. Such a termine the optimal sourcing of produc-
submodel is easily identified, and substan- tion to satisfy the forecasted demand for
tial computation savings accrue because the fiscal year; the extra six months of data
an explicit basis inverse (or other explicit mediate undesirable end effects in the so-
basis factorization) need not be main- lution. The firm uses the information on
tained for that submodel.
Once KPS is solved, results are loaded KPS has saved the Kellogg
into a database and checked for consis- Company millions of dollars
tency. If data problems are revealed, they since the mid-1990s.
can be corrected and the model rerun be-
fore central planners receive the results on production volumes to then establish fi-
Monday or Tuesday. nancial budgets within the plants, inven-
It is difficult to quantify the savings Kel- tory space requirements within the DC
logg obtains by using KPS rather than ear- network, and equipment projections for
lier manual methods. However, when we each transportation lane.
first introduced KPS for weekly planning, KPS plays an integral role in evaluating
management decided that production production capacity. By investigating the
costs should be equalized across plants. utilization of various processing lines,
That is, average production costs would planners can identify opportunities for im-
be used at all plants so that no plant provement. If they see that the utilization
would be likely to have a severe increase of the lines that make certain products is
or decrease in suggested production be- low, they may consolidate to reduce costs.
cause of not-yet-verified production-cost Conversely, if a set of such lines is fully
November–December 2001 11
BROWN ET AL.
utilized, they may seek additional capac- these alternatives off for more detailed fi-
ity. Also, if a product is produced on mul- nancial and engineering analyses. Those
tiple lines at different locations with some analyses, combined with an accounting of
lines operating at less than capacity, but fixed and variable cost implications for the
with the low-cost location operating at ca- scenarios, led to a decision about which
pacity, a capacity increase at that location lines to close. Then managers had to de-
may be justified. Managers must evaluate cide how to increase capacity.
the potential savings in variable costs us- By using KPS, we determined the re-
ing KPS and compare them with the cost quired overall capacity increase and gave
of the capital improvements. that figure to Engineering. Engineering
A recent consolidation project exempli- generated a list of implementable options
fies the use of KPS for capacity planning. that could deliver this increase and their
A combination of declining sales and in- costs. We then incorporated capacity infor-
creasing yields of certain products on cer- mation so that we could measure variable-
tain lines was leading to underutilization cost impacts of the various scenarios.
of other processing lines, and managers Combining this with the capital required
conjectured that they could make savings for each option enabled managers to make
by closing down some of the under- a financial comparison and select the best
utilized lines. option.
Multiple plants produced the relevant Having selected the consolidation plan
products, and no single line was com- to follow, we created a transition plan for
pletely idle. Initial runs of KPS also re- implementing it. KPS helped us answer
vealed that simply removing one of the these questions:
lines with low utilization would be un- —When could Kellogg take the lines tar-
wise, because the remaining lines would geted for capacity increases out of service
have too little capacity to fully support the and install new equipment? (Engineering
business. So, we used KPS to explore sets provided information regarding the time
of alternatives for shutting down a subset required and start-up curve estimates
of the lines and increasing capacity on oth- upon completion.)
ers. We undertook the study in two stages, —When should production cease on the
first determining which lines to shut down lines being eliminated?
and then deciding where to increase ca- —How much inventory should Kellogg
pacity. Ideally, we would look at both sets build to support the business during the
of decisions simultaneously, but the large transition?
number of combined options necessitated When Kellogg completes this project, it
this ad hoc decomposition. We identified estimates the savings will be between $35
reasonable alternatives for shutting down and $40 million annually.
lines by running KPS with data covering Kellogg also uses KPS to determine
18 months, with different combinations of where to produce new products, to assist
lines removed and additional capacity regular capacity reviews, and to justify or
spread across the open lines. We sent avoid the manufacturing and distribution
INTERFACES 31:6 12
KELLOGG
cost impacts of various projects. For its production and packaging, and co-packers
North American cereal business, we run have no exogenous demands.
Indices
KPS about 30 times a month to answer
f —food (product).
these types of what-if questions. k —stock-keeping unit (sku).
Conclusions p —plant (or distribution center).
After more than 10 years, KPS is still in t —time period.
development: Business never stops chang- l —processing line (for foods).
ing. Global operations will require some m —packing line (for skus).
h —production process.
refinements and more flexible inputs. We
f(h)—the food produced by process h.
will introduce binary variables to more ac- H(l, p)—processes h that use line l at
curately model the realities of line sched- plant p.
uling. We will more accurately model pro- K(m, p)—skus k that are packed on line
duction and packaging operations that are m at plant p.
tightly coupled; this should result in im- H( f, p)—processes h that produce food f
at plant p.
proved solution quality and possibly
K( f, p)—skus k that are packed from
speed. We may also model some critical food f at plant p.
raw materials with long lead times. K⬘(k, p)—skus k⬘ that are assembled
In both its tactical and operational roles, from (constituent) sku k at plant p.
KPS has saved the Kellogg Company mil- M(k, p)—packing lines m that pack sku
lions of dollars since the mid-1990s. Kel- k at plant p.
Data and [Units]
logg is introducing KPS into Latin Amer-
␣hlpt —fractional shifts used on process-
ica to improve operations there, and it is ing line l to produce one klb of food f(h) at
studying a global model. The advent of plant p during time period t [shifts/klb].
the European Union has simplified cross- bkmpt —fractional shifts used on packing
border operations in Europe, and the east line m to pack one klb of sku k at plant p
coast of the United States and the coasts of during time period t [shifts/klb].
ckk⬘p—fractional klbs of sku k used to
Europe are getting closer all the time.
make one klb of sku k⬘ [klbs/klb].
Acknowledgments dkpt —demand for sku k at plant p dur-
The authors are grateful for the support ing time period t [klbs].
of Don Scott of the Kellogg Company, and holdkpt —safety stock for sku k at plant p
thank Richard Powers of INSIGHT, Inc., during time period t [klbs].
who has shepherded this application since ulpt —capacity of processing line l, plant
p, time period t [shifts].
its inception. They also thank two anony- u⬘mpt —capacity of packing line m, plant
mous referees for a careful review of the p, time period t [shifts].
paper and the resulting improvements. holdkp0 —initial inventory of sku k, plant
APPENDIX p [klbs].
The following linear program is a didac- Decision Variables
tic specimen of the Kellogg Planning Sys- MAKEhpt —klbs of product f(h) pro-
tem. The constraints and variables corre- duced using process h at plant p during
spond to a generic location. Kellogg plants time period t.
exhibit all features described, distribution PACKkmpt —klbs of sku k packed on line
centers have no entities associated with m at plant p during time period t.
November–December 2001 13
BROWN ET AL.
INTERFACES 31:6 14
KELLOGG
Scott, D. 1999, “Custom optimization models have continued to enhance KPS’s capabil-
provide competitive advantage in the supply
ity and expand its use across all business
chain planning process,” Paper presented at
the Council of Logistics Management Annual channels. KPS’s planning capabilities have
Conference, Toronto, Ontario, Canada, Octo- played a vital role in Kellogg’s ability to
ber 17–20. continually reduce costs and inventory
Silver, E. A.; Pyke, D. F.; and Peterson, R. 1998,
Inventory Management and Production Planning levels while improving service and capac-
and Scheduling, John Wiley and Sons, New ity utilization throughout our total supply
York. chain.”
Wagner, H. M. and Whitin, T. M. 1958, “Dy-
namic version of the economic lot size
model,” Management Science, Vol. 5, No. 1,
pp. 89–96.
Zangwill, W. I. 1969, “A backlogging model
and multi-echelon model of a dynamic lot
size production system—A network ap-
proach,” Management Science, Vol. 15, No. 9,
pp. 506–527.
November–December 2001 15