This article was downloaded by: [Virginia Tech Libraries]

On: 18 October 2014, At: 12:38

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

IIE Transactions
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/uiie20

One-piece flow manufacturing on U-shaped production

lines: a tutorial
a
JOHN MILTENBURG
a
School of Business , McMaster University , 1280 Main Street West, Hamilton, Ontario, L8S
4M4, Canada E-mail:
Published online: 27 Apr 2007.

To cite this article: JOHN MILTENBURG (2001) One-piece flow manufacturing on U-shaped production lines: a tutorial, IIE
Transactions, 33:4, 303-321, DOI: 10.1080/07408170108936831

To link to this article: http://dx.doi.org/10.1080/07408170108936831

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the
publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations
or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any
opinions and views expressed in this publication are the opinions and views of the authors, and are not the
views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be
independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses,
actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever
caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 liE Transactions (2001) 33, 303-321

One-piece flow manufacturing on U-shaped

production lines: a tutorial

JOHN MILTENBURG
School of Business, Me Master University, 1280 Main Street West, Hamilton, Ontario US 4M4, Canada
E-mail: [email protected]

Received February 2000 and accepted June 2000

Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

Now-a-days shorter product life cycles and increased demands for customization make it difficult to produce some products on
traditional production lines. Often the best that can be done is to produce them in batch flow systems that have been improved
through the incorporation of line flow principles. This is one-piece flow manufacturing. Traditional cells with irregular material
flows are replaced by U-shaped production lines within which flow is regular and paced by a cycle time and between which flow is
controlled by pull signals. This tutorial examines the research literature on one-piece flow manufacturing. It begins with the
decisions rules that determine when one-piece flow is appropriate. Next the unique elements of one-piece flow (takt time, standard
work, flow manufacturing on U-shaped lines, pull production, and jidoka) are reviewed. Then the mathematical models that are
used to design one-piece flow systems are examined. Finally areas where more research is needed are discussed.

I. Introduction tions are completed at a regular rate called the cycle time.
The production system provides high levels of the cost
Production systems can be classified according to the type and quality competitive outputs.
of products produced (e.g., discrete versus continuous), Different production systems are used at different
the materials used (e.g., steel, plastic), the facility layout stages in a product's life cycle. In the birth stage products
(e.g., functional, cells, lines), the basis of competition (i.e., are produced in a job shop. During the growth stage
cost, quality, delivery, performance, flexibility, innova- demand for the product increases rapidly and production
tiveness), the timing of production (i.e., engineer-to- switches to batch flow and then to line flow when the
order, make-to-order, make-to-stock), and so on. One of product design stabilizes and the volume is enough to
the most useful classifications is the product-process utilize a dedicated production line. In the past the batch
matrix (Fig. I), which suggests that: flow production system was a temporary production
system, a stop on the way from job shop to line flow.
• there are a small number of production systems (i.e.,
Now-a-days shorter product life cycles and increased
project, job shop, batch flow, line flow, continuous
demands for customization for some products make it
flow);
difficult to move production to a line flow and batch flow
• production systems use different layouts and are
becomes a permanent production system. When this
appropriate for different mixes and volumes of
happens customers may become dissatisfied with the
products;
levels at which the cost and quality competitive outputs
• production systems are limited in what competitive
are provided. Attempting to improve the levels of these
outputs they can provide.
outputs by incorporating line flow principles into the
Compare, for example, the batch flow and line flow batch flow production system leads to one-piece flow
production systems in Fig. I. Batch flow produces me- (Fig. I). This new production system,
dium volumes of many different products on equipment
that is organized in traditional cells. Material flow inside • produces many products in medium volumes;
the cells is usually varied and flow between cells usually • on equipment arranged in cells in which material
follows patterns. The production system provides high flow is regular and paced by a cycle time; and
(i.e., desirable) levels of the flexibility and innovativeness • provides high levels of the flexibility and innova-
competitive outputs. The line flow production system tiveness outputs (like the batch flow system) and high
produces high volumes of several different products on levels of the cost and quality outputs (like the line
equipment that is organized in a traditional line. Opera- flow system).

0740-817X © 2001 "liE"

 304 Miltenburg

Traditional production systems

COMPETITIVE OUTPUTS
.__-------A.....-----____...'\
I"
One Delivery Cost Quality Perform- Flexibility Innova-
~~ ance ~~
very high
volume
Project layout;
flow extremely

- -
varied

Functional layout;
D c=::J CJ c=::J

• --
flow varied
Cellular layout;
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

flow varied 0 D D

-- -- • - -
with patterns
--------. ---------
U·lines paced
bycycle time,
pull production •
~._------._-----

Line layout;
tlow regular, paced
bycycle time
• ---------

_ _ c=::J
---------
D CJ CJ

Line layout;
flow rigid, pace
iscontinuous

POOR GOOD
I

One-piece flow production system

Fig. I. Clussificntion of production systems.

In the nex t section we examine decisions rules that more, given a choice between these three production
determine when one-piece flow production is appropriate. systems, line flow is preferred to batch flow, which is
Section 3 discusses the unique elements of one-piece flow preferred to job shop. If n is the number of different
production. Section 4 reviews mathematical models that products and P is the number of units produced per hour,
arc used in the design of this production system. Con- then the decision rule is:
cluding remarks follow in Section 5. Note that it is not the If 1 ::; n ::; 5 and I ::; P ::; 1000 then usc the line' flow
purpose of this paper to give a careful comparison of production system.
production systems or to examine critically the advanta- Elseif 5 ::; n ::; 100 and I ::; P < 50 then use the batch
gcs and limita lions of particular systems (Zangwill, 1992; flow production system".
Murphy, 1994). It is simply to give a useful description of Elseif 100 ::; n ::; several thousand and P ::; I then use
what is currently known about a way of organizing the job shop production system.
manufacturing called one-piece flow manufacturing. * One-piece flow production can be used whenever batch
flow production is appropriate.
Sekine (1992, pp. 39-50) recommends drawing a Pareto
2. Determining when one-piece flow is appropriate chart of the (annual) production volume for each product
and looking at the shape of the chart to determine
Askin and Standridge (1993, p. II) suggest that choosing whether the line flow or one-piece flow production system
between the job shop, batch flow, and line flow produc- should be used. The decision rule is:
tion systems depends on two things: (i) the number of Draw a Pareto chart of production volume for each
products, and (ii) the volume to be produced. Further- product.
 One-piece flow manufacturing on Urshaped production lines 305

If the first 30% of the products accounts for 70% or D, = expected demand per year for product
more of the total volume then i;
Use the line flow production system. Ai = W/ D, = allowed production time for one unit of
Elseif the first 40% of the products accounts for about product i;
60% of the total volume then Ri] A, = ratio of required to allowed production
Use the one-piece flow production system. time for product i;
Else maxdoid/Ai = ratio of maximum operation time to
More analysis is required before a choice can be made. allowed production time for product i;
mean.] oid / Ai = ratio of mean operation time to al-
Example I lowed production time for product i;
Consider the Pareto chart in Fig. 2 where annual pro-
di = random variable denoting demand per
unit time (say units per week) for
duction volume for 40 products produced in a factory is
product i with mean )1, and standard
displayed. For example, the annual volume for the
deviation a.;
highest volume product is 7200 units per year, which is
(Ji/)1i = coefficient of variation of demand per
17% of the total volume for all products. Then the first
unit time for product i.
30% of the products (i.e., 12 products) accounts for 72%
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

of the total volume. So the factory should be organized Then the decision rule is:

For i = I to n products
Assign i to job shop if Assign products to job shop when:
Product design is unstable - Product design is changing
or a, /)1i is high - Variability in demand is high
or Ri/A i is low - Required production time is low
Assign i to continuous flow if Assign products to continuous flow when:
Ri/A, is high and meank{oid/Ai is high - Required production time and average time
to complete an operation are both high
Assign i to line flow if Assign products to line flow when:
R, / Ai is high - Required production time is high
If low <Ri/A i < high then
If i is similar to a product already assigned to a production system then
assign i to the same production system
Elseif max.] Did/ Ai is low then
assign i to job shop
Else
assign i to batch flow or one-piece flow
Endif

for high volume production on a line flow production Versions of this decision rule were used at Caterpillar Co.
system. One-piece flow production is not appropriate. (Venkatesan, 1990), SKF Bearings, and J.J.Case. The rule
Notice that the first 40% or 16 products accounts for recognizes that most manufacturers use more than
79% of total volume, which is not close to the 60% target one production system (perhaps a different one in each
for one-piece flow production. focused plant-within-a-plant).
More comprehensive decision rules should be developed.
The one that follows is a small step in this direction. It is
based on rules used at real companies to make choices 3. Unique elements of the one-piece flow
between continuous flow, line flow, batch flow (and hence production system
one-piece flow), and job shop production systems. First
define the following notation. There are five elements that together make one-piece flow
a distinct production system. They are takt time, flow
i = index for products; manufacturing on U-shaped production lines, standard
k =index for operations; work, pull production control, and jidoka.
0ik =time (in hours) to complete operation k
for product i;
3.1. Takt time
R, = I.koik = required production time to complete
all operations for one unit of product i; Customer demand is the starting point for design and
W = available production hours per year; operation of the one-piece flow production system. The
 306 Miltenburg

the narrow side of each machine is on the edge and the

7000 deeper sides of the machine run away from the edge of the
100
U. Machines with this shape are preferred. Operators
6000
work inside the U-line. One operator supervises both the
i 5000
80
entrance and the exit of the line. Product flow and hence

..
1:
operator movement may be clockwise or counterclock-
f 4000 60 e
wise. As each finished product is pulled off the line, a new
& 3000
0..

set of material is input to the line. This happens once each

g 40
cycle time. No material enters the line unless a finished
~ 2000
20 product leaves. There is at most one unit of work-in-
1000
process inventory at each machine on the line. (See •
o 0 symbol in Fig. 3a).
1 3 5 7 9 11 1315171921232527293133353739
Product
This simple V-line is used alone or as a building block
for the more complex U-lines shown in Fig. 4(a--e).
Fig. 2. Pareto chart for Example I. Figure 4a shows production lines for three products A,
Band C organized in a single U-line manned by two
tuk t time Cj for each product i = 1,2, ... is calculated operators. This U-line is flexible enough that one op-
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

from an analysis of customer demand. Takt time is the erator can run all three lines when reduced output is
maximum interval between completion of consecutive required, and extra operators can be added when in-
units of a product and is calculated from: creased output is required. Figure 4b shows two U-Iines
manned by three operators. Again the arrangement is
(;, = available production time for product i flexible enough that any number of operators can run
customer demand for product i the lines to accommodate changes in required output.
The denominator includes all types of demand for the The embedded U-line arrangement in Fig. 4c has a
product-products for customers, samples, tests, spares, large U-line encircling a small U-line, all manned by
and so on. Once takt time is calculated, the objective for two operators. This is the least common complex U-
the production system is to do the necessary work within line, because it can require a lot of space and so require
this time. Traditional objectives such as high machine a lot of walking. A bigger disadvantage is that the same
utilization and high production volume are less impor- operator is usually not able to man the entrance and
tunt. At one company, for example, "high-speed is out, exit of a line. The complex U-line in Fig. 4d also has
tukt time is in" (Byrne, 1995, p. 10). this disadvantage, but it is less severe. Here three U-
Takttimes may change frequently. For example, once a lines are manned by two operators. Operator I runs line
month Toyota analyzes customer demands and calculates A and the first half of line B. Opera tor 2 runs the other
tukt times. At some companies changes in customer de- half of line B and all of line C. This arrangement is
mand of ± 20% are handled by adjusting available pro- called a figure-8-pattern because the paths taken by
duction time through overtime, vacations, etc. so that each operator have a figure-8-shape. The multi-U-line
takt times do not change. When changes are larger, takt facility (Fig. 4e) was first discussed by Monden (1983, p.
times are recalculated. 105-107). Another Japanese author calls it "the large
After takt time is calculated, cycle time is set. Cycle room effect" and describes it as follows. "Fractional
time, Ci, is the actual interval between completion of jobs will inevitably develop in any line. If operations are
consecutive units of product i. So C :::; Ci- A goal of the divided among operators each month on the basis of
one-piece flow production system is C, = Cj for each i. takt time, it will often happen that fractions of opera-
(More on this follows in Section 4.1). tors will be needed to clean up leftover operations.
These fractions of operations should be added up for all
lines, and given to one of the best operators in the
plant. ... Ultimately, the goal should be to tie all of
3.2. Flo", manufacturing on Li-shaped production lines
the lines together in one large room, and make one
Equipment layouts must be appropriate for the produc- large U-line out of them" (Nikkan Kogyo Shimbun,
tion system being used. In one-piece flow, production 1991a, pp. 56-57).
lines arc Ui-shaped, production is paced by a cycle time,
and a pIIII production control system is used. Note that 3.2.1. Work mode
U-shaped lines are sometimes used in other production U-lines work in one of three modes; chase, fixed stations,
systems (Irani, 1999). and overlapping stations. The U-line in Fig. 3a works in
On a U-shaped production line, machines are arranged chase mode because an operator takes the product
around the edge of a U-shape (see Fig. 3a) in the order in through the entire line. The U-lines in Fig. 4(a--e) work in
which operations are performed on the product. Ideally fixed station mode. Operations are grouped into stations
 One-piece flow manufacturing on V-shaped production lines 307

(3)

Standard Operation Chart

Quality check" 0
Wo~·in·prOCElSS producl:.
Number 01 work-in-
process products
Takllimu"
cycrenme

(b)
Required quantity per day (units): 750 Manual task
Available time per day (minutes): 465
Standard Operation Combination Chart Takt time (seconds): (465x60)/750 = 37'
Automatic task
Walking ..•••••
-....--.
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

Cycle time (seconds): 35'

Operation time, ok
Operation, Description Activities (manual, automatic, walking)
Manual, Automatic, Walking,
k 2' 4' 6' 6' 10' '2' 20' 22' 24' 26' 28' 3D' 32' 34' 36'
Pk a v '4' '6' '6'
k k

.
Take materials from 0'
1
pallet " " ~: .-
2 Gear cutting 3' 27'
" " .•
• • • • 1-. • • • • • •• • • • •• ~ • :
3

4
Drill

Chamfer
3'

3'
g'

g'
"
" "
'.
• •• • • .
. • • •• ••
". ... I
..
5

6
Ream

Attach fining
5'

3'
g'

'8'
"
" • • • ••
" . • •• • •
'. . • •• • • ••
. ". ...
:

7 Clean 2' 26'

" • •• •• •• ~. • • •• •• ~I :'1-- .... •• • • .1 10
•

: ~:
8 Balance 3' 7'
" • •
~- ~.
Place finished product 2' 0' 2'
9
on pallet

Totals: 25" 115' 10'

Fig. 3, (a) standard operation chart; and (b) standard operation combination chart.

and one operator is assigned to each station. In the sequence of operator-work, machine-work, and operator-
overlapping stations mode, some operations are assigned movement that is required to produce one unit of a
to adjacent stations and the operators in these stations product. Thus standard work defines cycle time. Standard
co-operate to complete these operations. work is constantly studied to see if improvements can be
Chase mode can also be used with more than one op- made. One principle in standard work is that operator-
erator. It is, however, more common in this case to use work is separated from machine-work so machines can
fixed or overlapping stations. For example, at Briggs and work unattended. Another principle is that operator-
Stratton "the cells were designed with equipment close work is more important than machine-work (Nikkan
together so they could be operated, if required, with one Kogyo Shim bun, 1991a, p. 62), which means that high
person in a chase mode - the single operator would chase a utilization of operators is more important than high
part through the entire process. If volume dictated two or utilization of machines.
three people per cell, they could continue to use the chase The standard operation chart and standard operation
method; or they could sub-divide the cell so that each combination chart are used to specify standard work. See,
person would operate a zone of stations" (Edwards, 1993). for example, Fig. 3(a and b). In the example shown there,
one operator works the U-line in chase mode. Nine op-
erations are performed. The first one loads new material
3,3, Standard work
onto the line. The next seven operations are performed at
Standard work outlines how operators and machines in- seven different machines. The last operation unloads
teract to produce a product. That is, it documents the a finished product from the line. Each operation at a
 308 M iltenburg

machine includes a manual task performed by the oper- takt time, it is necessary to improve operation k' by re-
ator and an automatic task performed by the machine. ducing Pk' or (lk',
The manual task usually consists of unloading a product
Consider Fig. 3(a and b) again. The values of Pko ai; Vk
from the machine, performing a quality check (see Co
are given in Fig. 3b. For example, P2 = 3, (12 = 27
symbol in Fig. 3a), loading a new product into the ma-
and V2 = I. The sum of the walking time and manual
chine, starling the machine, and watching the machine for
task times is Lk(Vk + Pk) = 35 seconds. Then YI =
a short time to make sure everything starts properly. The
I + 0 - 35 = -34, Y2 = 3 + 27 - 35 = -5, and so on. The
automatic task is usually machining, drilling, joining, etc.
The time to complete an operation is the sum of the
largest Yk is .Y2, so k: = 2 and Yk' = -5. Since Yk' 0, the :s
operator does not wait at any operation and the cycle
manual task time and the automatic task time for that
time is Lk(Vk + Pk) = 35 seconds as expected. Now sup-
operation. The longest operation time is a lower bound
pose that an additional 10 seconds are required to com-
on the cycle time, {;2. In this example, {;2 = 30 seconds
plete each automatic task. That is, (II = 0 + 10 = 10,
(from operation 2). The takt time C is calculated at the (12 = 27 + 10, and so on. Then YI = I + 10 - 35 = -24,
top of the standard operation combination chart .Y2 = 3 + 37 -35 = 5, and so on. The largest Yk is )'2, and
(Fig. 3b). Seven hundred and fifty parts are required
during a production shift of 465 minutes; so the takt time
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

is 465(750 = 0.62 minutes or C = 37.2 seconds. Then (a)

cycle time, C, satisfies Product C
alIT ,,
aisle'

:s
(i) {;2:S C C; and
,,,
(ii) the sum of the manual task times and the walking
time for each operator is :sc. ,,
,,
In this example, the sum of the manual task times and the ,
walking time is 25 + 10 = 35. So any cycle time that Product B Product A
:s :s
satisfies 30 35 C:S 37.2 is okay, say C = 35 seconds.
A change in takt time may require a change in cycle time, (b)
which would require a change in standard work. Because a Product B
alIT aisle' IN
typical factory has many U-lines, this is completed most
easily when operators, rather than industrial engineers, are
charged with developing standard work. However this re-
quires experienced operators (Hall, 1998).
Ohno and Nakade (1997) carefully examine how cycle
time is calculated from the standard operation combina-
Product A
tion chart (for a U-line working in chase mode) and give a
useful result. Define the following additional notation.
(c)
time to complete manual task at operation k;
time to complete automatic task at operation
k;
time to walk from operation k to the next
operation;
Yk Pk + Ok - Lk(Vk + Pk);
k' arg maxdyd - operation with largest value
of Yk (if there is a tie select the smallest k);
Yk(k'-I) nUtXk<k'_1 {yd;
Yk(O) O·
2Yk- - Yk(k'-I)
/I
Yk' - Yk(k--I)
Result
When Yk' :s
0, the operator does not wait at any opera-
tion (for a machine to finish). The cycle time is aisle
Lk(Vk + Pk), and if this exceeds the takt time it is necessary
to add a second operator. When Yk' 2': 0, after cycle n the Fig, 4, Complex U-lines: (a) three lines in a single U; (b)
operator waits Yk' only at operation k' for the machine to double-dependent U-lines; (c) embedded U-lines; (d) figure-
finish and the cycle time is Pk' + ai-, If this exceeds the S-pattern; and (e) multi-U-line facility.
 One-piece flow manufacturing on V-shaped production lines 309

(d)

IN
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

(e)
IN

Product B

Product A Product C

Fig. 4. (Continued).

so k" = 2, Yk' = 5 > 0, Yk(k'-Ij = Yk(l) = maxkSI {yd = cards, kanban squares, reusable containers, electronic
Yl = -24, and n = (2 x 5 -(-24»/(5 - (-24) = 1.2. kanban, and so on. Kanban squares, for example, are
Therefore the cycle time is Pk' + ai- = 3 + 37 = 40 sec- limited storage areas for output, which are placed be-
onds and beginning at the second cycle the operator waits tween a producer and a user. The user takes output from
5 seconds at operation 2 only. Since this cycle time the kanban squares, and the producer produces output as
exceeds the takt time, improvements are required at long as there are empty kanban squares in which to place
operation 2. it. A type of electronic kanban called "Faxban" was used
in the early 1990s at Zytec Co., a manufacturer of power
supplies used in high-tech products, to fax pull signals to
3.4, Pull production control
its suppliers.
Production at a "producer" V-line begins when a "user" Thus users (beginning with the external customer) pull
(usually a downstream V-line) signals that it has a re- products through the production system. When the cycle
quirement for products. Signals take the form of kanban time is approximately equal to the takt time, pull signals
 310 Miltenburg

must occur uniformly over time; otherwise, from time-to- The U-shaped line helps when problems occur.
time, requirements from users will exceed what producers "(Wlorkers are physically closer and work together as a
can produce causing shortages and disruptions. This is team. They check each other's work, and discuss prob-
achieved by levelling the production schedule at the "last" lems as they arise. ... Items needing rework are routed
production line, which is usually the final assembly line. immediately to the station causing the problem, leading
(More on this follows in Section 4.3.) The well-known to quick diagnosis and solution" (Anon, 1988).
rules for the usc of kanban are:
I. The user goes to the producer to get products.
2. The producer produces only the quantity removed 4. Designing the U-shaped production line
by the user.
3. Defective products are never given to the user. Designing a V-line requires: (i) selecting the work mode
4. Kunban always accompany products to prevent (Section 3.2); (ii) assigning operations to locations around
over-production. the U-shaped line and, if the mode is fixed or overlapping
5. Production schedules are levelled to avoid fluctua- stations, organizing operations into stations; and (iii) if
tions in pull signals. the U-line produces more than one product, selecting the
6. Reducing the number of kanban reveals opportu- sequence in which products will be produced.
nities for improvements.
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

4.1. Assigning operations to locations and organizing

3.5. Jidoka operations into stations
Quality practices must support the flow manufacturing There are two restrictions on the assignment of opera-
on U-lines and pull production control elements. Because tions to locations on the Uvshaped line. The first, is that
products are produced one unit at a time and only the the assignment must satisfy whatever technological
number of products required by a user are produced, even requirements exist for producing the product. These are
a single defective product will disrupt production. In this specified as precedence constraints on the order in which
environment, the only effective quality control method is operations may be completed. The second restriction
automatic, thorough inspection of each product as it applies when the U'-Iine works in fixed or overlapping
moves through aU-line. stations modes. In these cases the assignment must also
Jidoka is the name given to this quality control meth- permit operations to be grouped into a minimum number
od. It means endowing a line with the ability to recognize of sta tions.
that something has gone wrong and to stop of its own A U'-Iine may be described as shown in Fig. Sa. Denote
accord so that the problem can be corrected. Jidoka by a the point where material enters the line, and a' the
mechanisms enable machines to stop when: (i) a quality point where finished products leave. Let each operation
problem occurs; (ii) a machine problem occurs; or (iii) k = 1,2, ... req uire a distance h 2': 0 on the line. Define
processing is completed. Jidoka is also called auto no- the middle of the line to be a point e located a distance
mation, which means "automation with a human touch" greater than or equal to the integer part of ('f.kh)/2 from
and implies that machines have a human-like ability to a. Define the front of the line to be the locations from a to
sense when something goes wrong. Jidoka is even more e, and the back of the line to be the locations from a' to e.
important on U-lines that produce more than one prod- Let F = Iia - ell be the length of the front of the line and
uct (see Section 4.2 below) because operators make more B = lid - ell be the length of the back of the line, where
mistakes when they produce more than one product. F + B 2': 'f.kh.

(a) (b) (e)

Back
2 2
8=5 4 102 9 8
2
42 10
2
92 8
2

(F
e
B'

e
B'

e
6
4
B'

11= B B
54

7
3
B

Fig. 5. Example 2: (a) the U-line; (b) the ILP solution; and (c) the solution with the shortest walking distance.
 One-piece flow manufacturing on Uvshaped production lines 311

Example 2 distance is available for the operations assigned to the

Ten operations, each requiring one unit of distance, are to front and back, respectively, of the line. The next con-
be placed around the U-shaped production line shown in straint ensures that the precedence constraints are satis-
Fig. 5a. The line begins at point a and ends 10 locations fied for each operation assigned to the front of the line.
later at point a'. The middle of the line is 10/2 = 5 units The last constraint does the same for the operations as-
of distance from a. So five operations should be placed on signed to the back of the line.
the front of the line and five operations should be placed Example 2 (continued)
on the back. Suppose C = 10; the manual task times are two time
Integer linear programming (Urban, 1998; Scholl and units for operations 3, 4, 8, 9, 10, three time units for
Klein, 1999) and dynamic programming (Miltenburg and operations I, 7, and four time units for operations 2, 5, 6;
Wijngaard, 1994; Miltenburg, 1998) are two modelling and the precedence constraints are (3, I), (5, 10), (6, 9).
approaches that have been used to assign operations to Figure 5b shows an optimal solution obtained by solving
locations and organize operations into stations. (ILP). Figure 5c shows an alternative optimal solution.
Notice that the total walking distance is less in the solu-
4.1.1. Integer linear programming model tion in Fig. 5c than in Fig. 5b, and so Fig. 5c is actually a
better solution.
Define the following notation.
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

(ILP) cannot take walking distance into account and

j index on stations;
so may not find the best U-line design. Also it cannot
k index on operations;
generate alternative optimal solutions. The following
5 upper bound on the number of stations (e.g., 5 is
dynamic programming model, on the other hand,
equal to the number of operations);
can include walking distance and can easily generate
Pk time to complete manual task at operation k;
alternative optimal solutions.
P {(k, I): operation k must be completed before op-
eration I can begin} - precedence constraints; 4.1.2. Dynamic programming model
if operation k is located on Let r be a sequence specifying the assignment of opera-
front of line and assigned to station i. tions to locations on a U-shaped line. r is of the form
otherwise. r=kLkLkL ... where k u is the operation in the uth po-
sition in the sequence. Some operations may have a
if operation k is located on negative sign in r. If an operation appears without a sign,
back of line and assigned to station i. then it is placed in "the next available location" on the
otherwise. front of the line. If it appears with a negative sign, it is
placed in "the next available location" on the back. By
if station j is used, "the next available location" we mean start at point a or
otherwise. a' (if the operation has a negative sign) and move away
from these points until a location is available.
Then select Xjk, Yjb Zj to,
Let r f c r denote the subsequence of operations in r
(ILP) Minimize L~~l Zj that are located on the front of the line and let r b c r
denote the subsequence of operations in r that are located
Subject to: on the back of the line (i.e., the operations with negative
for each operation k, signs). Then r is afeasible sequence if four conditions are
Lj (Xjk + Yjkl = I
satisfied.
LkPk(Xjk + Yjk) ::::; CZj for each station j,
Lk t, LjXjk ::::; F, I. If I E / and (k,1) E P then k must precede I in r.
Lk t, LjYjk ::::; B, 2. If -k E r b and (k,1) E P then -I must precede -k in
Lj (5 - j + I)(xjk - xjl) ~ 0 for all (k, I) E P, r.
Lj (5 - j + I)(yjl- Yjk) ~ 0 for all (k, I) E P, 3. LkErfh::::; F.
for all i, k. 4. LkErbh::::; B.
Xjk, Yjk, Zj E {O, I}
The first two conditions ensure that precedence con-
The objective function minimizes the number of stations straints are satisfied, and the last two conditions ensure
that are used. The first constraint ensures that every op- there is enough room on the front and back of the line for
eration is located on the front or back of the line and is the operations assigned there.
assigned to one station. The second constraint ensures
that the time required to complete the manual tasks for Example 2 (continued)
the operations in each station does not exceed the cycle Suppose that r=3_-8_-9_2_-10_-4_-6_1_7_5. This
time, C. The next two constraints ensure that enough means operation 3 is assigned to the first location on the
 312 Miltenburg

trout or the line; namely the location starting at point a. 5 - k - denotes 5 - {k}, where k is an operation;
Then operation 8, which has a negative sign in T, is as- 5 + k - denotes 5 U {k};
signed to the first location on the back or the line; namely F(5) - cost of the subset sequence T(5) having the
thc location starting at point a'. Then operation 9 is as- minimum cost. Denote this T*(5).
signed to the next available location on the back of the
The cost of T(5) is (n - I)C + Wla51 where n is the number
line. This is to the immediate left of operation 8. Next
of stations and Wla5! is the work in the last station in the
operation 2 is assigned to the next location on the front of
induced assignment of operations to stations for T(5).
thc line, which is to the immediate left of operation 3.
Figure 5c shows the result for this sequence. The se- T(5 - k)-.k - sequence with operation k appended at
quence is feasible because the precedence constraints are the end of sequence T(5 - k);
satisfied, and five operations are on the front and five g( T(5 - k)-.k) - incremental cost when operation k is
opera lions are on the back. appended at the end of sequence
The organization of operations into stations is obtained T(5-k).
from T as follows. Starting at the beginning of T assign
The objective is to find F(K) and its minimum cost se-
as many consecutive operations as possible to the first
quence T*(K). Dynamic programming does this by com-
station. Thcn starting from the beginning of the remain-
puting F(5) recursively over the feasible subsets 5 ~ K
ing sequence assign as many consecutive operations as
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

until it reaches F(K). That is,

possible to the next station, and so on. Operations are
assigned to a station as long as the sum of the manual (DP) F(5) = minkEs{F(S - k) + g(T*(5 - k)-.k)},
task times and the walking time is less than or equal to F(¢) = O.
the cycle time. We write this T = [kl_k2_ ... -kul-
[kll+ l_kll+2_ _k,,]_[k"+I_kv+2_'" to indicate that opera- This recursion works best when the feasible subsets are
tions [kl , ki, , kll] are assigned to the first station, generated in order of size.
operations [kll+ l_kll+ 2_ ... _k,,] are assigned to the second
Example 3
station, and so on. This is called the induced assignment of
Four operations, each requiring one unit of distance, are
operations to stations for T. Then the work assigned to
to be placed around the U-shaped line shown in Fig. 6a.
station s is
Suppose C = 8; the manual task times are five time units
for operations 1,2, and two time units for operations 3,4;
HI, = LPk, +cr:(L hi +r.,. +cs) :::; C. and the precedence constraints are (1,3), (2,3), (3,4). Let
k,I1I.\ k; InS the front and back of the line be F= 3 and B = 2 distance
units, and let cr: = 0.2 time units to walk one distance unit.
The first term is the sum of the manual task times. The
Then the caleulations for (DP) are given in Table I.
second tcrm calculates the total walking distance in the
station and converts this to a time. cr: is the time to walk First all feasible subsets of one operation are generated -
one unit or distance. r, and c, ani the return distanee {I}, {2} and {4}. The corresponding sequences are [1_,
and crossover distance in station s. For example, in the [2_, and [-4_. In the first sequence operation 1 is assigned
first station in Fig. 5c rand c each equal two distance to the first location on the front of the line. In the last
units. sequence operation 4 is assigned to the first location on
the back of the line. The cost for sequence [1_ is five time
Example 2( continued) units to complete operation I plus the time to walk one
Suppose C = 12 time units and cr: = 0.2 time units/unit distance unit to complete the operation and one distance
distance. Then the induced assignment of operations to unit to return to the beginning - 0.2 x (I + I). So
stations is, T=[3_-8_-9_2U- 1O_-4_-6UI_7_5]. This F( {I}) = 5.4 time units.
is the solution shown in Fig. 5c. Tasks 3, 8, 9 and 2 are
in station I and W1 = (2 + 2 + 2 + 4) + 0.2 x ( I + I+
I + I + 2 + 2) = 11.6 time units. The sequence and in- (3) (b)
duced assignment for the solution in Fig. 5b is
[3_-8_-9_-10_-4U2_ -6UU_5]. e _ _L -_ _- - ' a' e
I-~

I------L-----l a'

Now define the following additional notation.

2
K set of all operations; 3

5 ~ K feasible subset of K;
T(5) feasible subset sequence-a sequence of a I---''----,r--''---,a
1'=3 ~ 5
operations in S. 2

A subset S is feasible if a sequence T(5) can be con- Fig. 6. DP solution in Example 3: (a) the Ll-line; and (b) op-
structed that satisfies conditions 1 to 4 above. timal solution.
 Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

Table 1. Calculations for Example 3 a

:::
Size Step Subset Sequence Cost Calculations -.:,""
m II sm-l r'(S;:'-I) F(S;:'-l) i {k} S;:' r(S;:',) Cost -.
h
= S'i/~l + {k} = r(S;:''-I)Jc =(n-I)C+W1ast ""'"
""
~
0 S;:' = 0 r(S;:') = - Cost = 0 c
-e
I I 0 0 I {I} 0+ {I} [1_ [(I - 1)8] + [5 + 0.2(1 + I))
= {I} = [0] + [5.4] = 5.4
s
l:l
2 {2} 0+ {2} [2_ [(1-1)8]+[5+0.2(1 + I)]
:::
§.,
l:l
= {2} = [0] + [5.4] = 5.4
3 {4} 0+ {-4} [-4_ [(I - 1)8) + [2 + 0.2(1 + 1)] ::::
= {-4} = [0] + [2.4] = 2.4 ...,
-'"
~.
(merge) results for m = I: h I 2 3
S;:' {I} {2} {4} c
:::
r' (5;")
h., [1_ [2_ [-4-
F(S;:',;l 5.4 5.4 2.4 c:::,
2 I {I} [1_ 5.4 I {2} {I} + {2} [IU2_ [(2 - 1)8] + [5 + 0.2(1 + I)] :::-
'"
= {l,2} = [8] + [5.4] = 13.4 {;
2 {4} {1,4} [1_-4_ [(1-1)8] + [5+2+0.2(1 + I + I + I)] l:l..
""
= [0] + [7.8J = 7.8 ~...,
2 {2} [2_ 5.4 I {I} {2,I} [2UI_ [(2 - 1)8] + [5 +0.2(1 + I)] = 13.4 c
2 {4} {2,4} [2_-4_ [(1 - 1)8] + [5+ 2 + 0.2( I + I + I + I)] = 7.8 ~
3 {4} [-4_ 2.4 I {I} {4,I} [-4_1_ [(1-1)8] + [5+2+0.2(1 + I + I + I)) = 7.8 c
-'"-.
2 {2} {4,2} [-4_2_ [(1-1)8]+ [5+2+0.2(1 + I + I + I)] = 7.8 :::
merge results for m = 2: h I 2 3
.....
s·
Sri {l,2} {1,4} {2,4} ""'"
r" (S~i) [IU2_ [1_-4_ [2_-4_
F(S~,) 13.4 7.8 7.8
3 I {1,2} [IU2_ 13.4 I {3} {1,2,3} [IU2_3_ [(2 - 1)8] + [5 + 2 + 0.2(1 + I + 1.4)] = 15.68
2 {4} {1,2,4} [IU2_-4_ [(2 - 1)8) + [5 + 2 + 0.2( I + I + 1.4 + 1.4)] = 15.96
2 {I A} [\_-4_ 7.8 I {2} {1,4,2} [1_-4U2_ [(2 - 1)8] + [5 + 0.2(1 + I)] = 13.4
2 {3} {IA,3} [1_-4U-3_ [(2 - 1)8] + [2 + 0.2(1 + I)] = 10.4
3 {2,4} [2_-4_ 7.8 I {I} {2A, I} [2_-4UI_ [(2 -1)8] + [5 +0.2(1 + I)J = 13.4
2 {3} {2,4,3} [2_-4U-3_ [(2 - 1)8] + [2+ 0.2( I + I)] = 10.4
merge results for m = 3: h 1 2 3 4
S;:' {1,2,3} {1,2,4} {1,3A} {2,3,4}
r'(5;")
h,I [IU2_3_ [I_-4U2_ [1_-4U-3_ [2_-4U-3_
F(S;:'J 15.68 13.4 10.4 10.4
4 I {l,2,3} [IU2_3_ 15.68 I {4} {1,2,3A} [IUUU-4] [(3 -1)8] + [2+0.2(1 + I)J = 18.4
2 {1,2,4} [1_-4U2_ 13.4 I {3} {1,2,4,3} [1_-4U2_3] [(2 - 1)8] + [5 + 2 + 0.2(1 + I + 1.4)] = 15.68
2 [1_-4U2_-3] [(2 - 1)8] + [5 + 2 + 0.2(1 + I + I + I)] = 15.8
3 {1,3,4} [1_-4U-3_ 10.4 1 {2} {1,3,4,2} [1_-4U-3_2] [(2 - 1)8] + [2 + 5 + 0.2( I + I + I + I)] = 15.8
4 {2,3,4} [2_-4U-3_ 10.4 I {I} {2,3,4,I} [2_-4U-3_1] [(2 - 1)8) + [2 + 5 + 0.2( I + I + I + I)) = 15.8
merge results for m = 4: h I
S;:'i {I ,2,3A}
r" (5;:',) ti,-4U2_3]
F(St,) 15.68 w
w-
 314 Miltenburg

Next all feasible subsets of two operations are gener- "attach casters when possible to machines and worktables
a led by adding one operation to lhe feasible subsets from to make them easily mobile.... Machines bolted to the
the previous step. Operations 2 and 4 may be added to floor are permanently in place. Changes in the market-
subset {I}. When operation 2 is added, the feasible subset place and plant improvements, however, may require
is {1,2} and the sequence is 1_2. When operation 4 is changing machine layouts by product type. Floor bolts
added, the feasible subset is {I ,4} and the sequence is are our enemies! Machines must be movable. Casters
1_-4. Note that the sequence 1_4 is not possible because will help" (Hirano and Black, 1988, pp. 69 and 119).
it violates feasibility condition I. The induced assignment
of operations to stations for 1_-4 is [1_-4_ because I and
4.2. Mixed-model production
-4 can be placed in the same station. The walking dis-
c,
tance is I, + + 14 + 1', = I + I + I + I = 4 distance The ability to produce different products (or models) on
units. So the cosl is 5 + 2 = 7 time units to complete the the same Ll-line depends on the setup time.
manual operations plus 0.2 x 4 = 0.8 time units for
• When setup times are negligible, V-lines can operate
walking, for a total of 7.8. Notice that the subset {l,4}
as mixed-model production lines where each station
can also be formed by adding operation I to the feasible
can produce any model in any cycle. The U-Iine is
subset {4} from the previous step. The sequence is -4_1,
said to be a flow shop with a permutation schedule
and the induced assignment is [-4_1_. This has the same
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

and batch size of one. (Askin and Iyer, 1993)

cost as [1_-4.. If the cost was different, then the merging
• When setup times are small (i.e., less than the cycle
operation, which corresponds to the minimization in
time), a "cycle setup change" is used in which a
(DP), selects for {1,4} the sequence with the lowest cost.
dummy workpiece is sent through the line when it is
The calculations continue in this way. The optimal
time to switch production from one product to an-
solution is F(K) = 15.68. The corresponding sequence is
other. The arrival of the dummy workpiece at a
r'(K) = [1_-4L[2_3] as shown in Fig. 6b.
machine tells the operator to complete the required
This model can easily incorporate other conditions on
setup. (Nikkan Kogyo Shim bun, 1991a, p. 68)
locations where operations can be placed (e.g., some
• When setup times are large, there are two possibili-
operations may need to be close together, others may
ties.
need 10 be separated, some machines may be in fixed
• The Ll-line produces each product in a small
locations, and so on) by adding conditions to feasibility
batch, sayan amount equal to the product's daily
conditions I to 4. This is important when U-lines are
requirements. When a batch is completed, the V-
rebalanced.
line stops and undergoes a setup to switch to the
4.1.3. Rebalancing V-lines next product. As large setup times are reduced,
batch sizes decrease towards the goal of one unit
U-Iincs should be designed with rebalancing in mind.
of each product. For example, at one Pratt &
"The really important thing, however, is to decide how
Whitney plant "32-piece lot sizes decreased to six"
equipment can be added to or removed from the line in
and the plant is still "driving towards one-piece
the future to cope with increases or decreases in pro-
flow." (Tonkin, 1997).
duction demands. The same is true with your operators.
• Complex Ll-lines like those in Fig. 4(a--e) (espe-
It might be difficult to think in terms of doubling or tri-
cially d and e) are formed.
pling the number of operators you have, but changes of
20 percent or 50 percent may need to be implemented on The apparel industry is one where manufacturers have
a day-to-day basis" (Nikkan Kogyo Shimbun, 1991b, experimented with one-piece flow on mixed-model U-
p. 23). lines. Knapton (1989), for example, discusses this at
Customer demand docs not stay the same forever. Jantzen Inc., a swim and sportswear garment manufac-
When it changes, a sequence of changes is triggered. Takt turer. Manufacturers in the apparel industry must re-
time may change, which may change cycle time, which spond quickly to style- and season-changes. In the
may change standard work, which may require V-lines to industry's traditional batch flow system, products are tied
be rebalanced. Rebalancing once a month is common into bundles of 10 to 50 pieces, which move from machine
(Monden, 1983; Hall, 1988). One US manufacturer re- to machine for processing. An operator takes a bundle
balances its lines every week (Conti, 1996). Rebalancing from his/her machine's work-in-process buffer, unties it,
may involve: (i) adding operators to increase output or completes the required operations on each piece, reties
removing operators to decrease output; (ii) moving ma- the completed pieces into a bundle, and sends the bundle
chines; and (iii) changing the standard operations charts to a buffer at another machine. Operators are paid a piece
(Section 3.3). Item (i) is easier when operators are multi- rate for each completed piece. In one-piece flow on
skilled: "flexibility also makes the addition or subtraction mixed-model V-lines, operators follow procedures known
of opera tors from the cell easier as demand changes." as the Toyota Sewn Products System (Bischak, 1996).
(Anon, 1988) Item (ii) is easier when the following is done: Because many different products are produced and
 One-piece flow manufacturing on V-shaped production lines 315

products are added and dropped quickly as styles change, every 5 + 5 + 2.5 + 2.5 = 15 minutes. So the cycle time
the operation times vary from product to product and is 15/2 = 7.5 minutes. Machines I and 2 are the slowest
also from operator to operator. The easiest way to model machines and so give an upper bound on the expected
this is to let operation times be stochastic (Nakade and production that can be achieved by the line. For example
Ohno, 1997). Then some small lines can be modelled as in a production shift of 465 minutes = 27 900 seconds the
Markov chains, and simulation can be used to analyze expected production at machine I is 465/5 = 93 products
other lines. if one operator is assigned full-time to the machine. Each
To illustrate the use of Markov chains consider a V- operator spends 1(3 of their time at machine I, so the
line working in chase mode. Each operation is performed machine is working 2(3 of the time. Then the upper bound
at a machine. Machines cannot work unattended and so becomes 2/3 x 93 = 62 products per shift.
automatic task time is zero and operation time equals
The transition probability diagram for the corresponding
manual task time. Define a Markov chain with states
Markov chain is also shown in Fig. 7a. Let the time unit
{lIlk : k = 1,2, ...} where IIlk = 0, 1,2, ... is the number of
be I second. Then },I = 1/(5 x 60) = 0.0033, }'2 = 0.0033,
operators at machine k. When m, = 0 machine k is idle
A3 = 0.0017 and },4 = 0.0017, and one transition occurs
because no operator is working there, m, = I means one
each second. The transition probability matrix P, the
operator is working at machine k, and m, = 2 means one
vector of steady-state probabilities n, and covariance
operator is working at machine k while a second operator
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

matrix C are calculated easily and are given in the figure.

is waiting to use the machine. Assume that the first ma-
The productive states are states 4, 7, 8 and 10. So
chine is never starved for work and the last machine is
A =0.2609 and V=41.847. That is, the line is in a pro-
never blocked from delivering its completed work. Mea-
ductive state 0.2609 x 100 = 26.1% of the time. But
sure operation time in small units of time, and let that
0last = 04 = 1/ }'4 = 150 (productive) seconds are needed,
time be geometrically distributed with rate h for machine
on average, at the last machine to complete a product. So
k so that the expected time to complete the operation at
in a production shift of 465 minutes = 27 900 seconds, the
machine k is Ok = 1/ Ak. Define the following additional
95% interval estimate for the actual production is
notation.
27900 x (0.2609/150) ± 1.96 x 27900J(41.847/150 2 ) =
i = index for states; 48.5 ± 14.1 = (34.4,62.6) products.
P transition probability matrix;
=
Figure 7b depicts the V-line working in fixed and
I identity matrix;
=
overlapping stations modes. The expected work in each
1! = nP = vector of steady-state probabilities station is 7.5 minutes in both modes. Both modes can be
with elements nj; analyzed using Markov chains.
<D = square matrix with each row = 1!; This type of analysis was used in Bischak (1996) and
Z = (1- P + $)-1 = fundamental matriz for Markov Zavadlav et al. (1996) to analyze lines such as these used
chain with elements zij; at companies like Oshkosh, Wrangler, and Blue Bell.
C = covariance matrix with elements Among their results are the following.
_ { 1!jzij + njzji - 1!jnj i =F j; I. When the variability in operation times is high the
c..I) - nj(2zij - I - nj) i = j; chase mode gives the best expected production.
Overlapping stations mode is second best, and fixed
SI {states with IIllast 2: I} = set of pro-
=
stations mode is the worst. When there is no vari-
ductive states: those where opera-
tor works at last machine; ability in operation times all modes give the same
production.
S2 = {states with IIllast = O} = set of un-
2. When variability in operation times is high, ex-
productive states: those where no
pected production can be increased by locating
operator works at last machine.
buffer inventories between machines when chase
Then the number of units produced by the V-line during mode is used, and between sta tions when fixed
a period of T cycles (where T is large) is normally dis- stations and overlapping stations modes are used.
tributed with mean A = T(~jESlnj/Olast) and variance However buffer inventories increase work-in-pro-
V = T(-~jESICjj/o?ast). See Kemeny and Snell, (1976, cess inventory and throughput time.
jES2
p. 89) for a detailed discussion. 3. Chase mode has the highest requirements for
training. All operators must be trained to perform
all operations for all products. Fixed stations mode
Example 4 has the lowest training requirements.
Consider the U-line shown in Fig. 7a. There are four ma- 4. Additional rules can be added to the chase mode.
chines, two operators, and the line works in chase mode. For example, the Toyota Sewn Products System
The expected operation times are 5, 5, 2.5 and 2.5 minutes includes some rules. Among other things rules will
at machines I, 2, 3 and 4. Two products will be completed reduce the training requirements somewhat. An
 316 Miltenburg

(a)
~ Expected completion lime
(2.5) (2.5)

Markov chain

(5) (5)
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

e.g.: aII a represents the state where there is

one operator at machine 2. one at machine
3, and none at machines I and 4.
Results 2 4 6 7 8 9 10
I 0.9967 0.0033 0 0 0 0 0 0 0 0
2 0 0.9933 0.0033 0 0.0033 0 0 0 0 0
3 0 0 0.9900 0.0067 0 0.0033 0 0 0 0
4 0.0067 0 0 0.9900 0 0 0.0033 0 0 0
5 0 0 0 0 0.9967 0.0033 0 0 0 0
p=
6 0 0 0 0 0 0.9900 0.0067 0 0.0033 0
7 0 0.0067 0 0 0 0 0.9900 0.0033 0 0
8 0 0 0.0067 0 0 0 0 0.9867 0 0.0067
9 0 0 0 0 0 0 0 0.0067 09933 0
10 0 0 0 0.0067 0 0 0 0 0 0.9933

2 4 6 7 8 9 10
7t= 0.1739 0.0870 0.1739 0.0870 0.0435 0.0435 0.0435
0.1739 0.0870 0.0870

2 4 5 6 7 8 9 10

1 60.72 -8.15 -1.00 0.62 -23.69 -8.87 -11.28 -2.68 -3.23 -2.44
2 -8.15 26.97 0.44 -1.10 -8.15 -1.10 0.44 -2.75 -3.30 -3.30
3 -1.00 0.44 13.73 1.58 -11.28 -0.79 0.38 0.09 -0.58 -2.56
4 0.62 -1.10 1.58 11.39 -8.87 -2.75 -0.79 0.29 -0.77 0.41
5 -23.69 -8.15 -11.28 -8.87 60.72 0.62 -1.00 -2.68 -2.44 -3.23
c= 6 -8.87 -1.10 -0.79 -2.75 0.62 11.39 1.58 0.29 0.41 -0.77
7 -11.28 0.44 0.38 -0.79 -1.00 1.58 13.73 0.09 -2.56 -0.58
8 -2.68 -2.75 0.09 0.29 -2.68 0.29 0.09 4.89 1.24 1.24
9 -3.23 -3.30 -0.58 -0.77 -2.44 0.41 -2.56 1.24 10.54 0.70
10 -2.44 -3.30 -2.56 0.41 -3.23 -0.77 -0.58 1.24 0.70 10.54

(bl
(2.5) (2.5) (2.5) (2.5)

,,

(5) (5) (5) (5)

Fig. 7. Example 4: (a) chase mode; and (b) fixed stations mode and overlapping stations mode.
 One-piece flow manufacturing on Usshaped production lines 317

example of rules is the following. s c ... a worker index for products produced at
moves forward with a piece, processing it at each last process;
machine in sequence. At some point the worker will index for time periods in units of
no longer be able to go forward for one of two cycle time at last process (i.e.,
reasons: either the worker is blocked by another one product is completed each
worker or the worker has reached the last machine. time period at the last process);
The worker will either drop off the piece and move number of different products
backward to find another piece to work on, or the produced by process j;
worker will wait in a 'worker queue' behind the weight for variation in produc-
busy worker. To reduce the number of machines tion of products at process j;
that each worker visits regularly and thus must be w, weight for variation in work at
trained to use, an operational rule that we refer to station s in last process;
as the 'preemption' rule is implemented in the sys- number of units of product u at
tem. The preemption rule can be stated as follows. process j 2: 2 required to pro-
Suppose worker i begins moving backward without duce one unit of product i at last
a piece in hand. If worker i encounters worker i - I process;
before finding another piece to work on, worker i (given) demand for product i at
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

takes the piece that worker i - I is working on and last process;

continues processing that piece at that machine. demand for product u at process
After handing off the piece, worker i - I moves j 2: 2;
backward and repeats this search for work. This D j -- ",'Ii
~u=l
d
u] demand for all products at pro-
process continues until all workers have a piece in cess j;
hand. In a system without buffers the initial back- ratio of demand for product u at
ward move causes a chain reaction of hand-offs process j to demand for all
which concludes with the start of a new piece by the products at that process;
first worker" (Bischak, 1996, p. 725). Xilt number of units of product i
produced during periods I to tat
4.3. Levelling production last process;
number of units of product u
In any production environment where pull production
produced during periods I to t at
control is used, the focus of attention for scheduling is
process j 2: 2;
the last production process (which is usually a final as- x)t -- ","J
Lo u= I XII)I total number of units produced
sembly line), because production at all other processes is
during periods I to t at process j
pulled, directly or indirectly, from this process. In ad-
(XII = t because one product is
dition to the usual goals for the production schedule at
completed in each period at the
the last process - e.g., meet customer due dates, avoid
last process).
overloading stations, minimize finished goods inventory
- the schedule must ensure that pull signals occur uni- Then the level production problem selects Xii, to,
formly between all processes. Otherwise, from time to D[ "j
time, requirements at user processes will exceed what
can be produced at producer processes (Section 3.4)
Minimize L L I>Vj(xujl - rUjXjl)2, (I)
1=1 j?2 u=1
causing shortages and disruptions. The mechanism for
getting uniform pull signals is levelling the production Subject to
schedule at the last process. For example at one man- "I
ufacturer of electronic products, "models with different LXill=t t=I,2, ... ,D 1 , (2)
specifications move along the same line.... Rather than i=l
producing large quantities of one model, the mixed
XilDl =dil i= 1,2, ... ,nl, (3)
model line levels production by producing small quan-
tities of various models. Parts subassembly lines ... are O:SXill-XiI,I-I:S I i= 1,2, ... ,111, t=2,3, ... ,D I ,
synchronized with the main line." (Hirano and Black, (4)
1988, p. 133) A small literature exists on the problem of
levelling production schedules. XiiI IS a non-negative integer
Define some additional notation. i= 1,2, ... ,111, t= 1,2, ... ,D I . (5)
j = index for production process; The objective function seeks the schedule at the last
j =I is last process, j = 2 IS process XiiI that minimizes the squared difference between
second-last process, etc.; the actual and target production of each product in each
 318 Miltenburg

period at all other processes. xu)! is the actual production number of subassemblies, fabricated parts, and purchased
of product u at process j ;::: 2 during periods I to I. ru)X)t parts, and product 3 requires the largest number. So a
is the corresponding target production if product u is "balanced" or level schedule seems to be one where
produced at a constant rate. So the objective is to select a product I separates products 2 and 3; for example 2_1_3,
schedule for the last process that keeps all other processes or the reverse sequence 3_1_2. The optimal schedules,
producing at a constant rate. (While differences are usu- determined by solving (I) to (5), are 100 repetitions of
ally squared, there may be cases where absolute differ- 2_1_3 or 100 repetitions of 3_1_2.
ences, minimizing the maximum difference, and so on are
appropriate.) The first constraint ensures that exactly I Kubiak (1993) showed that this problem is NP-hard.
products arc produced in t periods at the last process. The Algorithms for solving it are examined in Miltenburg
second constraint ensures that the demand for each and Sinnamon (1989) and Cheng and Ding (1996).
product is satisfied. The last constraint ensures that in Four special cases of the problem are studied in the
each period, either one unit of product i is produced or literature.
product i is not produced at all.
Case 1. Pull signals at some processes are more important
Example 5 The weights, 11') in (I), are used to vary the importance of
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

Consider a final assembly line that must produce 100 different processes. When pull signals are equally im-
units of each of three products during a 465 minute shift. portant for all products at all processes, set 11') = I for
Then the takt time is 465/(3 x 100) or one product every I each j. Sometimes it is the processes that are equally im-
minute and 33 seconds, and the cycle time is something portant. But some some processes produce more products
less than or equal to this. Define a time period to be one than others. In this case, set 11') = I/n) for each j. When
cycle time. So I = 1,2, ... ,300 and we need to select the one process is more important than other processes (e.g.,
products that will be produced by the final assembly line subassembly at Toyota in the next case), set 11'), > wf!')'
in each of these periods. The bills of material for products where j' is the important process.
arc shown in Table 2. Final assembly is the last process
U = I). The other processes are subassembly U = 2), Example 5 (continued) Table 4 shows the optimal
fabrication U = 3), and purchased parts U = 4). Each schedules for various values of the weights. When 11') = I
process may have one or more U-lines. U-lines in pro- for each j the optimal schedules are the ones mentioned
ccsses j ;::: 2 only produce parts when they receive a pull earlier. When 11') = I/n) - 11'2 = 11'4 = 1/3, 11'3 = 1/4, the
signal from a user U-line. It is not obvious from looking same schedules are optimal. When only pull signals be-
at the bills of material what the best final assembly tween final assembly U = I) and subassembly U = 2) are
schedule is. This changes when hiJ ? 2,1 is calculated important, 11'2 = I and 11'3 = 11'4 = 0 and the optimal
(Table 3). Notice that product 2 requires the fewest schedules change.

Table 2. Hills of material

Filial assembly Subassembly proccss Part fabrication process Purchased parts process
process
Subassemblies Subassemblies Fabricated parts Fabricated parts Purchased parts

Products 2 3 I 2 3 4 I 2 3
I I 0 I I I 0 0 1 I o o
2 0 0 2 I I I I 2 o I o
3 I I 3 0 0 I I 3 o I o
4 o o I

Table 3. ltiJ ? 2,u values

j=1 j=2 j=3 j=4

Filial assembly Subassembly Part fabrication Purchased parts

II = 2 3 u = I 2 3 4 u = I 2 3

product i = I I 0 2 2 I I 2 3 1
2 0 0 I I I I I 2 1
3 I I 2 2 2 2 2 4 2
 One-piece flow manufacturing on V-shaped production lines 319

Table 4. Objective function values for different values of the weights

Sequence Objective junction: (I) (6) (7)
Weights \\'2, \\'J. \\'4: 1,1,1 1/3,1/4,1/3 1,0,0

100 repetitions of
123 169 51 72 78 133
1_3_2 135 41 61 78 133
2_1_3 126 40 89 III ill
2_3_1 135 41 61 78 ill
3_1_2 126 40 89 III 133
3_2_1 169 51 72 78 133
Note: Minimum objective function values are underlined.

Case 2. The goal chasing method minimize the variation at the last process (i.e., final as-
The most interesting case is from the 1980s at the Toyota sembly). That is,
Motor Company. It is called the goal chasing method and D, 111
differs from (I) to (5) in two ways. Minimize L L (XiiI - rilXII)z, (7)
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

\. It selects the schedule for the last process) = I (final 1=1 ;=!
assembly line) that minimizes only the variation at Subject to (2), (3), (4), (5).
the second-last production process) = 2 (subas-
sembly process). That is, it tries to smooth pull (7) is said to be an approximation for (I), and certainly it
signals between final assembly and subassembly. simplifies the problem. However for those situations
The reason is computational. "It is difficult to apply when an approximation is needed, minimizing variation
the goal chasing method since the number of dif- at the second process (i.e., subassembly) is much better.
ferent parts used in an automobile is about 20 000. At least this ensures that the pull signals between final
Therefore, the parts are represented only by their assembly and subassembly are uniform. Recall that this is
respective subassembly ... The main subasssembly what the goal chasing method does. (7) gives no assur-
names are as follows: I. body types 2. engines 3. ance that pull signals will be uniform at any process, and
Transmissions ... " (Monden, 1993, p. 264). it's hard to imagine a real company where it would be
2. The target production for product u at process) = 2 used.
is duzt / D 1• In (I) the target production is ruzXZt • But Example 5 (continued)
ruzXzI = duzXzI/Dz. These targets are about the Table 4 shows that any sequence is optimal when (7) is
same because t/D 1 ~XzI/Dz. That is, the fraction used.
of all products produced during periods I to t,
t/ D 1, should be about the same as the fraction of all Kubiak and Sethi (1989) show that this case can be re-
subassemblies produced during the same periods, duced to an assignment problem. (Good algorithms are
Xz,jD z. Notice that calculating t/D 1 is less com- available for solving large assignment problems.) The
putational work and that may be the reason it is reduction is done by making each unit of each product a
used. See Miltenburg and Sinnamon (1989) for a distinct product, determining the due date for each dis-
more thorough analysis. tinct product, calculating the cost of producing each
product in a period different from its due date, and then
So the goal chasing method selects XiiI to: solving an assignment problem that assigns products to
D n2
time periods to minimize costs. They show that the due
Minimize L 2)xuzI - duzt/D1)z, (6) date for the Ith copy of product i is 21 = 1(21 - I )/(2r;)1
1=1 u=1
and the cost of producing this product in time period tis,
Subject to (2), (3), (4), (5).
if t < ZI,
Example 5 (continued)
if t z,
=
The optimal schedules for goal chasing method is are 100
repetitions of 1_2_3 or 1_3_2 (or 100 repetitions of the if t > ZI.
reverse sequences 3_2_1 or 2_3_1). See Table 4. Notice
how similar this solution is to the solution to (I) to (5) Example 6
when the weights are 1,0,0.
Consider three products with demands 2, 2 and I. Table 5
Case 3. Variation in the last process is important shows that the due dates for product I are period 2 for
A small academic literature has developed on a level r
the first copy (1(2 x I - I)/(2 x 2/5)1 = 1.251 = 2) and
production problem where a schedule is selected to period 4 for the second copy. The copies of product 2
 320 Miltenburg

Tablc 5. Solving problem with objective function (7)

Product Demand ratio Copy Due dale Cost, C;,

r., Z'I Period I = I 2 3 4 5

i = I 2/5 I 2 0.2 0 0.6 1.6 2.6

2 4 2.6 1.6 0.6 0 0.2
2 2/5 I 2 0.2 0 0.6 1.6 2.6
2 4 2.6 1.6 0.6 0 0.2
3 1/5 I 3 0.8 0.2 0 0.2 0.8

have the same due dates, and the due date for product 3 is It IS appropriate (Section 2) impressive results are ob-
period 3. The cost of assigning the first copy of product I tained (Miltenburg, 2000). Exactly how an implemention
to the 4th period is CI4 = L~=2111 - u x 2/51- should be done is a subject that would benefit from
c;t
II - I - II x 2/511 = 1.6. Then the values of in Table 5 careful research. A useful paper in this regard is Brown
form an assignment matrix. The optimal solution is ob- and Mitchell (1991).
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

tained by inspection, 1_2_3_1_2. Ideas for research projects on one-piece flow follow
easily from the topics discussed in this paper. Some ex-
Case 4. Workload ill stations all the final assembly line amples are: (Section 2) better decision rules for selecting
is also import ant the appropriate production system, (Section 3.3) models
In many final assembly lines products are mounted on a for calculating cycle time from the standard operation
conveyor and operators move with the conveyor while combination ehart for mixed-model U-lines working in
working on the product. An operator can work on a fixed or overlapping stations modes, (Section 3.4) the
product only when it is within a station defined by up- effect of kanban size and number on pull signals between
stream and downstream limits. If products with relatively mixed-model V-lines, (Section 4.1) assigning operations
longer operation times are seheduled consecutively an to locations and organizing operations into stations for
operator may be unable to complete them within the complex (Fig. 4(a-e» mixed-model U-lines working in
station and line stoppages or incomplete products will overlapping stations mode, (Section 4.1) easy procedures
result. In this case the objective function (I) should in- for rebalancing U-lines, (Section 4.2) the effect of setup
clude a term to smooth the workload at stations in the times on mixed-model production, and so on. Then there
final assembly line. For example: are the usual problems that arise when operation times
are stochastic (rather than deterministic) and perhaps
non-stationary because of the effects of learning.
However the one-piece flow production system is real
and practical, not abstract and theoretical. So especially
+ ~ L.= w, (~ o,.,XtI,) _ kO') 2 (8)
useful would be research that is anchored in real imple-
mentations. That is, research on problems like the ones
described above would be most interesting and useful if it
The second term measures differences between actual and is motivated by or compared to experiences at real com-
target production lime at each station in the last process in panies. While this requires a little more work by re-
each time period. OJ,. is the operation time for product i in searchers, the results should be well worth it.
station .1', and Os is the weighted (by demand) average A final line of interesting research projects is one that
operation time for all products at station s. Then follows real one-piece flow production systems over time
:E;'';'\ O;sXilt is the total production time at station s in pe- to learn what problems are most difficult at different
riods I to I. ko, is the corresponding target production points in time, how these problems are solved, how
time whcn work proceeds at a constant or average rate. product design changes (i.e., engineering change orders)
This problem is reviewed by Korkmazel and Meral (2000). are handled, how operators are trained and rotated
The general problem of smoothing workload on assembly through stations, and so on.
lines has a rich literature. See, for example, Yano and
Raehamadugu (199\), Bard et al. (1992) and Tsai (1995).
Acknowledgements
5. Concluding remarks
This research was supported by Research Grant A5474
One-piece flow is an interesting and useful production from the Natural Sciences and Engineering Research
system. When implemented carefully in a situation where Council of Canada. I would like to thank Professor
 One-piece flow manufacturing on V-shaped production lines 321

Michael Caramanis, the Focused Issue Editor for Design Miltenburg, J. and Sinnarnon, G. (1989) Scheduling mixed-model
and Manufacturing, for his help with this paper. multi-level just-in-time production systems. International Journal
of Production Research, 27(9), 1487-1509.
Miltenburg, J. and Wijngaard, J. (1994) The U-line balancing problem.
Management Science, 40(10), 1378-1388.
References Monden, Y. (1983) Toyota Production System, Industrial Engineering
Press, Institute of Industrial Engineers. Norcross, GA.
Anon (1988) Making life easier in work processing areas, Modern Murphy, F.H. (1994) Comments on 'The limits of Japanese production
Materials Handling, 43(3), 49-52. theory' by William I. Zangwill, Interfaces, 24(5), 77-94.
Askin, R.G. and Iycr, A. (1993) A comparison of scheduling philoso- Nakade, K. and Ohno, K. (1997) Stochastic analysis of aU-shaped
phies for manufacturing cells, European Journal of Operational production line with multiple workers. Computers and Industrial
Research. 69, 438-449. Engineering, 33(3), 809-812.
Askin, R. and Standridge, C. (1993) Modeling and Analysis of Manu- Nikkan Kogyo Shimbun (l99Ia) The Factory Management Notebook
facturing Systems, Wiley, New York, NY. Series: Autonomation] Automation, Productivity Press, Portland,
Bard, J.F., Dar-EI, E. and Shtub, A. (1992) An analytic framework for OR.
sequencing mixed-model assembly lines. International Journal of Nikkan Kogyo Shimbun (l99Ib) The Factory Management Notebook
Production Research, 30(1), 35-48. Series: Mixed-Model Production, Productivity Press, Portland, OR.
Bischak, D.P. (1996) Performance of a manufacturing module with Ohno, K. and Nakade, K. (1997) Analysis and optimization of a U-
moving workers, II E Transactions, 28, 723-733. shaped production line. Journal oj the Operations Research Soci-
Brown, K.A. and Mitchell, T.R. (1991) A comparison of just-in-time ety of Japan, 40(1), 90-104.
Downloaded by [Virginia Tech Libraries] at 12:38 18 October 2014

and batch manufacturing: the roles of performance obstacles. Sekine, K. (1990) One-Piece FloII', Productivity Press, Portland, OR.
Academy of Management Journal, 34(4), 906-917. Tonkin, L. (1997) Making improvements on many fronts at Pratt &
Byrne, A. (1995) How Wiremold reinvented itself with kaizen. Target, Whitney's rigid tube and manifold production operations. Target:
11(1),8-14. The Periodical of the Association for Manufacturing Excellence,
Cheng, L. and Ding, F. Y. (1996) Modifying mixed-model assembly line 13(1),43-44.
sequencing methods to consider weighted variations for just-in- Scholl, A. and Klein, R. (1999) ULlNO: optimally balancing U-shaped
time production systems. IlE Transactions, 28, 919-927. JIT assembly lines. International Journal of Production Research;
Conti, R.F. (1996) Variable manning JIT: an innovative answer to 37(4), 721-736.
tcam absenteeism. Production and Inventory Management, first Tsai, L.H. (1995) Mixed-model sequencing to minimize utility work
quarter, 24-27. and the risk of conveyor stoppage. Management Science, 41(3),
Edwards, J.N. (1993) Inside a smoother running engine: an American 485-495.
giant transforms itself. Target: The Periodical of the Association Urban, T.L. (1998) Optimal balancing of U-shaped assembly lines.
for Manufacturing Excellence, 9(3), 5-11. Management Science, 44(5), 738-741.
Hall, R.W. (1988) Cyclic scheduling for improvement. International Venkatesan, R. (1990) Cummins Engine flexes its factory. Harvard
Journal of Production Research, 26(3), 457-472. Business Review, March-April, 120-127.
Hall, R.W. (1998) Lean manufacturing, continuous improvement at Yano, C.A. and Racharnadugu, R. (1991) Sequencing to minimize
TABC, Long Beach, CA. Target, 14(2),5\-53. overload in assembly lines with product options. Management
Hirano, H. and Black, J.T. (1988) JIT Factory Revolution: A Pictorial Science, 37(5), 572-586.
Guide to Factory Design of the Future, Productivity Press, Port- Zangwill, W.1. (1992) The limits of Japanese production theory. In-
land, OR. terfaces, 22(5), 14-25.
Irani, S.A. (1999) Handbook of Cellular Manufacturing Systems, John Zavadlav, E., McClain, ).0. and Thomas, L.J. (1996) Self-buffering,
Wiley & Sons, New York. self-balancing, self-flushing production lines. Management Sci-
Kemeny, J.G. and Snell, J.L. (1976) Finite Markov Chains, Springer- ence. 42(8),1151-1164.
Verlag, New York.
Knapton, J. (1989) Exploding the traditions of an old-fashioned in-
dustry. Target, 5(2), 24-28.
Biography
Korkmazel, T. and Meral, S. (2000) Bicriteria sequencing methods for
the mixed-model assembly line in just-in-time production systems. John Miltenburg is a Professor of Production and Management Sci-
European Journal of Operational Research, (in press). ence in the Michael DeGroote School of Business at McMaster Uni-
Kubiak, W. (1993) Minimizing variation of production rates in just-in- versity in Canada. He received his Ph.D. from the University of
time systems: a survey, European Journal of Operational Research, Waterloo. His research interests are in production planning and con-
66. 259-271. trol, facility layout, and just-in-time production systems, and he has
Kubiak, W. and Sethi, S. (1991) A note on level schedules for mixed- published articles on these subjects in journals such as llE Transac-
model assembly lines in just-in-time production systems. Man- tions, Management Science, Naval Research Logistics. and European
agement Science, 37, 121-122. Journal of Operational Research. He worked for a time in Manufac-
Miltenburg, J. (1998) Balancing U-lines in a multiple U-Iine facility. turing Engineering at General Motors.
European Journal of Operational Research, 109, 1-23.
Miltenburg, J. (2000) U-shaped production lines: a review of practice. Contributed by the Decentralized Control of Manufacturing Systems
International Journal of Production Economics, in press. Department.