This article was downloaded by: [University Aut Ciudad Juarez]

On: 02 April 2013, At: 04:08

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

Evaluating Flexible Manufacturing Systems

a

G. John Miltenburg & Itzhak Krinsky

Production and Management Science Faculty of Business, McMaster University Hamilton,

Ontario, Canada, LSS 4M4
b

Finance and Business Economics Faculty of Business, McMaster University Hamilton,

Ontario, Canada, L8S 4M4
Version of record first published: 09 Jul 2007.
To cite this article: G. John Miltenburg & Itzhak Krinsky (1987): Evaluating Flexible Manufacturing Systems, IIE
Transactions, 19:2, 222-233
To link to this article: http://dx.doi.org/10.1080/07408178708975390

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions
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.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should
be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims,
proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in
connection with or arising out of the use of this material.

Evaluating Flexible Manufacturing

Systems
G. JOHN MILTENBURG
Associate Professor of Production and Management Science
Faculty of Busin<ss,McMaster University
Hamilton, Ontario, Canada, L8S 4M4

ITZHAK KRINSKY

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

Associate Pmfersor of Finance and Business Economics

Faculty of Business, McMarter University
Hamilton, Ontario, Canada, L8S 4M4

Abstract: Many firms are considering Flexible Manufacturing Systems (FMS) as a means for increasing productivity,
quality and profitability. In this paper a methodology for properly comparing and evaluating FMS's is presented.
T h e appropriate financial criteria are presented. Mathematical models of different FMS's are presented. T h e important stochastic variables are determined. T h e principles of stochastic dominance, risk preference and the value
of information; and a decision analysis cycle are used to evaluate the FMS's.

A Flexible Manufacturing System (FMS) can be described

as a set of transformation stations on which a variety of parts
are automatically processed and between which parts are automatically transported. FMS's are considerably more complex than this definition suggests. An overview of the
technology, and descriptions of FMS's can be found in several
articles. (A summary of some of the theoretical work that is
currently being done on FMS's can be found in [17], and in
recent articles in the Journal of Manufacturing System 11982
on].)
Many firms have either introduced or are considering flexible manufacturing technology (Suri and Whitney 1211). These
firms see FMS as a means for increasing productivity, profitability and quality as well as a means of maintaining a competitive edge in the market place. Unfortunately, many
researchers have reported that traditional approaches to the
financial justification of FMS's tend to discourage their adoption. (See, for example, Kaplan 1121, Burstein and Talbi [5],
Gold [lo], and Suresh and Meredith [20].) Michael and Milen
1151 suggest that traditional financial evaluation models are
more suited to meet short-term profitability goals rather than
long-term strategic goals. This paper presents a methodology
for using traditional financial evaluation models to help evaluate FMS's. It is argued that when these models are properly
used, they do indeed encourage the adoption of FMS's.
The most important property of a FMS is its flexibility.
Flexibility has three components. (1) There is the flexibility
to produce a variety of products using the same machines as
Received February 1985:revised August 1985, F e b w 1986, and June 19.36.
Handled by the Manufacturing and Automated Production Department.

222

0470-817X~85M600-01821$2200i0
O 1987 " I I E

well as the flexibility to produce the same product on different

machines. (This allows firms to easily increase or decrease
production capacity.) (2) There is the flexibility to produce
new products on existing machines, and (3) there is the flexibility of the machines to accomodate changes in the design
of products. (See Buzacott [6] and Browne et al. [4] for a
more complete discussion of flexibility.) When evaluating
FMS's, these components of flexibility must be accurately
modelled.
In this paper we propose a methodology for selecting from
among a number of flexible manufacturing systems. Components of the methodology include:

1) using deterministic variables, stochastic variables, and

models to capture the nature of each manufacturing system;
2) using the correct financial criterion-discounted
flows;

cash

3) a mechanism for screening the more important stochastic

variables from the less important (so that attention can be
directed to the important variables);

4) a framework for doing sensitivity analysis and determining

the value of attempting to reduce the amount of uncertainty in the problem (value of information);
5) a mechanism for taking account of the risk attitudes of the
decision makers.

Although other researchers use some of these components

(Knott and Getto 1131, Suri and Hildebrant 1221); most researchers assume that the problem is deterministic (BoothVolume 19, Number 2, IIE Transactions, June 1987

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

royd [3]), use the wrong financial criterion (Dorf [$I), or

consider primarily the qualitative aspects of the problem (Arbe1 and Seidmann [I], Kulatilaka [14]).
Failure to accurately model and capture the nature of a
FMS, and using the wrong financial criterion, are, in our
opinion, the primary reasons why FMS's seem to be difficult
to justify.
There are two kinds of manufacturing systems that can be
considered-assembly systems and forming systems. Assembly systems assemble components into final products while
forming systems actually form components or final products.
This paper evaluates only assembly manufacturing systems
because assembly activities are similar in most manufacturing
firms. The methodology suggested below captures the nature
of each manufacturing system better than any other methodology we are aware of. It integrates all the key demand
conditions (volumes, number of products, growth), engineering variables (cycle time, capacity, quality, etc.) and financial variables (taxes, interest rate, depreciation, etc.).
The next section briefly describes and justifies the financial
evaluation model that will be used. Then mathematical
models for six flexible manufacturing systems are presented.
(The systems differ in the "amount" of flexibility that they
have.) This is followed by a review of the methodology, and
an illustration of the complete methodology with a worked
example. The final section summarizes the results in this paper and describes possible extensions.

Traditional Financial Evaluation Models

Many financial models can be used to help evaluate projects. For the problem in this paper (the Iinn must choose the
best FMS from among a number of mutually exclusive alternatives) the Net Present Value (NPV) is the appropriate criterion. See, for example, Schall et al. [19]. The Net Present
Value (NPV) is the difference in the present value of the
after-tax cash inflows and outflows. If the NPV is positive
(that is, inflows exceed outflows), the project should be accepted. Since it would be incorrect to compare the NPVs of
projects having different lives, each project's NPV should be
converted to an Annuity Equivalent (AE), and the AE's of
the projects should be compared. The larger the A E the more
attractive the project. By so doing, we assume a like-for-like
replacement of projects system because of the difficulty of
determining the costs of acquiring and operating future generations of manufacturing systems. The alternative to AE is
the NPV common life multiple technique which uses an evaluation internal equal to the lowest common multiple of the
lives of the projects under consideration. (In the example,
which follows, the lowest common multiple is 840 years.)
Hence, A E simplifies the computations. Based on a survey
of U.S. companies, Gitman and Forrester [9] reveal that some
firms use other criteria. Namely, 53.6 percent of the companies use the Internal Rate of Return (ZRR) and 8.9 percent
use the Payback (PB) as their primary technique for evaluating projects. The IRR of a project is that discount rate which
equates the present value of the after-tax cash inflows with
June 1987, IIE Transactions

the present value of the after-tax cash outflows. O n the surface ZRR and NPV look the same. However, as pointed out
by Copeland and Weston 17, p. 551, only "the NPVmle avoids
all the problems which the IRR is heir to. It obeys the value
additivity principal, it correctly discounts at the opportunity
cost of funds, and, most importantly, it is precisely the same
thing as maximizing the shareholders' wealth." The payback
(PB) period of a project is found by counting the number of
periods it takes before the cumulative forecasted cashflows
equal the initial investment. (Some companies discount the
cashflows before they compute PB.) In any case, the PB and
the discounted P B rules depend on an arbitrary cutoff date,
beyond which the project will be rejected, and ignore all
cashflows after that date. We will concentrate on NPV and
A E although, for interest sake, we will also look at IRR and
PB.
The details of the specific financial model that is used are
described in Appendix 1.

Mathematical Models for Flexible Assembly Manufacturing

Systems
Six assembly manufacturing systems will be c o n s i d e r e 6
namely, manual assembly (MA), manual assembly with mechanical assistance (MAM), rotary or in-line indexing machines with special purpose workheads and free transfer lines
with special purpose workheads (SPW), free transfer lines
with robot workheads (RW), and robot assembly cells (RAC).
(Figure 1 shows these manufacturing systems.) Consider the
production of a family of assemblies where each assembly
consists of a number of parts. The assembly is produced on
one of the above assembly systems. Assume that a number
of different assemblies are to be produced in year 1. The
number of assemblies (NP,), the number of parts (NAj) in
each assembly i ( i = 1, 2, . . . , NP,), and the production
requirements for each assembly (VS,,,), may or may not all
be known. The manufacturing system consists of a number
of stations and at each station NS parts are assembled. Each
station has a "workhead"-either
a human operator (WA),
a special purpose workhead (CW), or a rohot (CR). The
work-in-process is mounted on a work carrier or pallet (CC),
and is transferred from one station to the next by a transfer
device (CB o r CT). The production schedule is as follows.
Set up and produce the entire year's requirements for assembly 1. Then change over for assembly 2 and produce the entire
year's requirements for assembly 2. Continue for the entire
family of assemblies. The more flexible assembly systems can
be changed over to manufacture all assemblies. (The transfer
devices, the robots and the assembly operators can be reused.
However, the work carriers (CC), the mechanical feeding
devices (CF), the robot part magazines (CM) and the robot
grippers (CG) must he changed.) If non-programmable automation (SPW) is used, then a different assembly system is
required for each assembly. U p to three s h i s (SH = 3) can
he used to produce the total requirements for all assemblies.
During the life of an assembly, a number of parts may he
redesigned (for marketing considerations). If this happens

a) MA

Manual Assembly

model for the Robot Assembly Cell and an illustrative example are given here. (In what follows the subscript 1 (year)
will be suppressed for clarity.)

Work Carriers (CC,

b ) MAM

rn

Manual Assembly with Mechanical assistance

Mechanical Feeding Oevicer (CF)

. indexing and buflered lines wiln Special Purpose Workheads

c,d) SPW

Robot Assembly Cell, RAC (Figure 1-f)

The robot assembly cell uses two robots. Each of these
sophisticated robots has six degrees of freedom. While one
robot is assembling a part, the other robot can change grippers
and pick up the next part. If the assembly cell is car?fully
designed, the average time to complete a good assembiy i, is
TPi = NAi * (TRE + P Q * TD); where TR is the average
time (in seconds) for the robot to assemble one part, P Q is
the part quality (the fraction of defective pans to good parts),
and TD is the average downtime at a station due to a defective
part. S H is selected so that the required production capacity

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

does not exceed the available capacity. That is

i"s

VSi * TP;I

i=l

e) RW

. iransfer

3600) s HR * PE * SH; where HR is the number of promachine with Robot Workheadr

duction hours per shift, and PE is the plant efficiency. (Of

course if there is insufficient capacity over 3 shifts, a second
assembly cell would be added.) The required initial investment per cell is I1

Assembly Robot lCRl

f l RAC

2 * CR

NP

1 (CC + NTj * (CM +

i=l

Rabol Assembly Cell

CG)) where CR is the cost of a robot. If a part is redesigned

then the pan feeder (CM) and robot gripper (CG) will have
to be changed. If one supervisor is assigned to monitor each

Asasmbly Rob3 (CRI

Roo04 Pad Magszine lCMl
Work Carrier iCCl

cell, the annual labour cost is WT

WS *

NP

(VS,

* TPjI

i=l

Figure 1. 6 Assembly Manufacturing Systems

3600)l(HR * PE), where WS is the annual cost of a supervisor.

some of the assembly equipment must be changed--specif,cally the carriers (CC), partfeeders (CF), robot grippers
(CG), robot part magazines (CM), and the special purpose
workheads (CW). In addition, if some of the assemblies have
common parts (NT), the automatic feeding devices ( C F ) ,
robot grippers (CG) and robot part magazines (CM) can be
reused.
For the sake of analytical tractability we will use these
simple models. (The extension to more complex models is
left for future research.) For example, producing the entire
year's requirements in a single run, overlooks inventoq lot
sizing considerations, production scheduling, etc. Conceptually, it is a straightfonvard task to add more variables and
make other assumptions so that the models are more appropriate for a particular firm. This is suggested in the final
section.
The details of the models for all the assembly systems (except the Robot Assembly Cell) are given in Appendix 2. The

Example

Suppose that two assemblies are to be produced in a robot

assembly cell for the next seven years. Assembly 1 has
NA, = 6 parts and assembly 2 has NA2 = 10 parts (of which
NT2 = 8 pans are different from assembly 1).A new assembly
will be introduced in year 4. This assembly will have NA3 =
8 parts and NT3 = 6. All assemblies will have one part redesigned each year. Suppose that TR = 6 seconds, TD = 30
seconds, PQ = 0.05, HR = 2000 hourdshiftlyear and P E =
0.95. Costs are as follows-cost of robot CR = $110000,
robot pan magazine CM = $5000, robot gripper CG =
$1700, carrier CC = $1500 and the annual w s t of a supervisor
WS = $55000. The tax rate is TX = 0.40, the robot assembly
cell is depreciated over D = 10 years and all assemblies are
sold for $0.65 each. The average assembly times are TP, =
27 seconds, TP2 = 45 and TP3 = 36. The annual production
requirements are:

Year, t

VSI,~

2
140250
168300

127500
153000

154275
185130

2869
2
1

3156
2
1

3471
2
1

169702
203643
119000
5008
3
1

186672
224007
130900
5509
3
1

205340
246408
143990
6060
2
2

225874
271049
158389
6666
2
2

Js
'29

vs3,t

X VS;,, * TPi13600
No. shifts, S H
No, cells
22A

1IE Transactions, June 1987

(Recall that the capacity of a shift is HR*PE = 1900 hours.)

Notice that in years 6 and 7 there is insufficient capacity with
one assembly cell, and so a second assembly cell is needed.
The required initial investment is II = $267800, with additional investments of $6400 in years 2, 3 and 4 because of
part design changes in assemblies 1and 2. In year 4 assembly
3 is introduced, necessitating an investment of NT3 * (CM +
CG) + CC = $20700. An additional investment of $9600 is
required in year 5 because of pan design changes in the three
assemblies. In year 6 a new robot assembly cell is required
at cost Z6 = 2'CR

then determined. The next (informational) phase uses the

results of the first two phases to determine the economicvalue
of eliminating uncertainty in each of the important variables.
If there are profitable sources of information, then the decision should be to gather the new information. Hence the
design and execution of the information-gathering program
follow. Often the new information requires revisions to the
original analysis and so the first three phases are repeated.
(Usually the modifications are slight.) This iterative procedure is repeated until there are no longer any profitable
sources of new information. At this point the principles of
stochastic dominance and risk preference are used to determine the "best" decision from the probability distributions
of the outcome variable for each alternative. (For additional
information see, for example, Howard and Matbeson [ l l ] o r
Raiffa [IS].)

(CC + NTi(CM + CG)) = $288500.

i=l

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

As well, pan design changes require an investment of $9600

for the original cell. Finally in year 7 an additional investment
of 2*$9600 is required because of part changes. Using equations A1 to A7 (from Appendix l ) with this model gives:
Year

Revenue
REV,
$182325
200557.5
220613.25
320024.58
352027.03
387229.74
425952.71

At a discount rate of 20%;

NPV = $181455.89,
IRR = 0.556,

Labour Cost
LAB,

$60400
66442
73074
105432
115979
127579
140337

Investment
1,
$267800
6400
6400
27100
9600
298100
19200

Depreciation
DEP,
53560
44128
36582.4
34685.92
29668.74
83354.99
70523.99

634600

352504.04

Selecting the Assembly Manufacturing System

A E = $251701
PB = 1.8 years.

We will now use the methodology and the models of the

previous sections, to select the "best" assembly manufacturing system.

Clearly the robot assembly cell is an excellent investment.

i) Deterministic Phase
For the financial model and the assembly system models
discussed previously, the decision maker's specialists are
asked to provide estimates for the most likely values of all
the variables. For the sake of illustration, suppose they gave
the most likely values listed in Appendix 3 (Deterministic
Variables). Substituting these estimates into the models of
the previous sections, gives the results shown in Tables 1 and
2. Since the appropriate criterion is the Annuity Equivalent
(AE), a Free Transfer Line with Robot Workheads (RW)
would be the best manufacturing system. The second best

Methodology
The methodology takes the form of an iterative procedure
called the Decision Analysis Cycle (see Figure 2). The procedure is divided into three phases. In the first (deterministic)
phase, the variables affecting the decision are defined and a
model showing the relationships between the variables is developed. The impoltance of each variable is also determined.
In the next (probabilistic) phase uncertainties (that is, probabilities) are encoded on the more important variables. The
associated probability distribution of the outcome variable is
DETERMINISTIC
PRIOR
INFORMATION I
)
PHASE

PROBABILISTIC
PHASE

INFORMATIONAL
PHASE

INFORMATION
GATHERING

Figure 2. The Decision Analysis Cycle p. 9, Howard and Matheson [I9831

June 1987, IIE Transactions

After Tax Cash

Flow, A TCF,
- $173221
91720.5
96756.5
115529.9
143896.3
- 108967.6
180379.0

DECISION

I
)ACT

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

Table 1. Results of Deterministic Phase--Resuits Under Most Likely Conditions

Assembly System
AE
IRR
1. MA-Manual Assembly
$251670
44%
2. MAM-Manual Assembly with Mechanical Assistance
285819
22
3. SPW-Assembly Machines with Special Purpose Workheads
-678363
4
4. RW-Assembly Machines with Robot Workheads
845477
23
5. RACRobot Assembly Cell
352684
18
system is the Robot Assembly Cell (RAC). Notice that if the
wrong criterion is used-namely IRR or PB-then the Manual Assembly (MA) would he selected. This is so because the
initial outlay is smallest for Manual Assembly. (The second
choice would be RW.) For the rest of this paper we will limit
our attention to the Annuity Equivalent (AE), since it is the
appropriate criterion.
While estimating the most likely values it becomes evident
that some variables are deterministic (the value of a deterministic variable is known with certainty) while other variables are stochastic (their values cannot be estimated exactly).
To illustrate; suppose the decision-maker's experts gave the
most likely values and ranges shown in Appendix 3 (Stochastic Variables). That is, eighteen of the variables are deterministic while sixteen are stochastic. (Of course, the mix
of deterministic and stochastic variables will vary from firm
to lirm depending upon the nature of the firm and its experience with different FMS's.)
Although there are many stochastic variables, some are
more important than others. The sensitivity of the output
variable A E to each of the stochastic variables is then measured so that the important stochastic variables can be identified. (If the output variable is sensitive to changes in a
stochastic variable, then the stochastic variable is called an
important stochastic variable.) Table 3 shows the sensitivities,
for the estimates in Appendix 3. These sensitivities were calculated as follows. A stochastic variable was set to its low
value and all other variables were set to their most likely
values. Using the models of the previous sections, AE's were
calculated for all five assembly systems. The average AE,
E ;the mean absolute deviation of AE, MAD; and the ratio
MAD/= were then calculated. (This ratio is a measure of
risk per unit of return.) The stochastic variable was then set
to its high value and the procedure was repeated. Then the
average ( M A D I E ) , based on all the high, low and most
likely values, is calculated. Those stochastic variables which
produce a large average (MAD/=) are called important
stochastic variables because they have an important effect on
the variabilities of the AE's for the five assembly systems.
The stochastic variables which produced a small average
( M A D I E ) do not have an important effect on the variability
Table 2. Ranking FMS's From Table 1
Criterion
Ranklng
IRR
AE
I-Best
RW
MA
RW
2
RAC
3

4
sworst
226

MAM

MAM

MA
SPW

RAC
SPW

PB
MA
RW
MAM
RA C
SPW

PB
3.4 years
>5
>6

5.06
6,16

of the AE's and so can be treated as deterministic variables.

An alternative procedure would be to treat each FMS separately and pick the important stochastic variables for each
FMS, rather than for all FMS's. There are other more sophisticated procedures for determining the relative importance of the stochastic variables. The above procedure has
the advantage that it is accurate, it takes both risk and return
into account and it is easy for a decision-maker to understand.
Table 3 shows that the most important stochastic variable
is the annual demand for the assemblies, MS. Other stochastic
variables such as market share, MSHARE, and market
growth, MGROW, which affect the annual demand, are also
very important. (These three variables emphasize the great
importance of demand.) Next in importance are the project
lives and the risk-free interest rate, IR; followed by the number of new assemblies introduced each year, NAP, and the
number of annual part design changes, ND.Finally there are
the other production variables such as assembly times, part
quality, etc. Notice that the important stochastic variables are
either variables which measure the "required flexibility" of
the assembly system, o r are financial variables. (Recall that
flexibility has three components-flexibility to meet changing
demands (MS, MSHARE, MGROW): flexibility to accommodate new products (NAP); and flexibility to modify products (ND).)
To summarize; in the deterministic phase a model was developed, a most likely case was analyzed, and the important
stochastic variables were identified.
ii) The Probabilistic Phase
Since some of the variables are stochastic, the output variable, AE, for each assembly system will also be stochastic.

Table 3. Deterministic Phase Results-Sensltivities of AE's to

Stochastic Variables
'100
Rank
Variable
Average (MADIE)
NP
11.78
10
NAP
25.10
4
1
MS
69.15
MGROW
6.27
It
MSHARE
18.27
6
Project Life
32.22
2
TX
12.76
9
k
31.25
3
TA
30.66
TM
29.22
7
TW
2.98
TR
17.45
TD
4.11
PE
6.27
12
PO
13.59
8
ND
23.88
5

IIE Transactions, June 1987

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

To encode uncertainty in all of the important stochastic variables, cumulative density functions (cdf s) are estimated for
each of them, by asking appropriate questions of the experts
within the firm. By systematically substituting all possible
combinations of variable values into the models, the cdf for
AE for each of the five asembly systems can be calculated.
This procedure is depicted in Figure 3. Each path represents
one particular setting of all the variables. (Notice. that the
tree grows very rapidly when there are many important stochastic variables. If possible one should treat all stochastic
variables as important stochastic variables. Unfortunately
Limited time and computer resources usually force us to identify and focus our attention on the important stochastic variables.) Each important stochastic variable's cdf is then
approximated by a number of discrete values. As many values
as possible can be used. For illustrative purposes, three
branches are nsed for the annual demand MS, and two
branches are used for NP, NAP, MGROW, Project Lives,
k, TA, TM, TW, TR, PQ and ND. (The number of branches
depends upon the importance of the stochastic variable, the
nature of its distribution, and the available computer re-

sources.) The variables PE, TD and TX will be treated as

deterministic variables. Although they are stochastic variables, they are not important stochastic variables (Table 3)
and the ranges of possible values for these variables are relatively small (Appendix 3).
Using the cdf s from Appendix 3, our computer program
calculated an A E cdf for each of the five manufacturing systems. (See Figure 4.) Looking, for example, at the cdf for
the robot assembly machine we see that there is a 50:50
chance that A E will be above or below $750000, and there
is a 20-percent chance that A E will exceed $1.25 million. To
order the manufactnhg systems from best to worst, the following two step procedure is used. First; use first-order, second-order and thid-order stochastic dominance (SD) to
order the systems. (See, for example, Bawa [2] for a review
of stochastic dominance..) Second; if SD doesn't exist between
some systems, estimate the risk preferences of the decisionmaker and use this to order the remaining systems. From
Figure 4, we see that the assembly machine with special purpose workheads, SPW, is the worst manufacturing system
(using the principle of first-order SD). It appears that RW,

IMPORTANT STOCHASTIC VARIABLES

ASSEMBLY
SYSTEM

MS

Figure 3. Decision Tree Structure (5 x 3 x

June 1987, IIE Transactions

NP

= 3840

NAP

PROJECT
MGROWl

ASSEMBLY
I TIMES

pQ

ND

branches)
227

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

AE (Annulty Equlvalenl), Smillion

Figure 4. Cumulative Density Functions of Annuity Equivalents for
Five Assembly Manufacturing Systems

the assembly machine with robot workbeads, is the best system. Unfortunately, we cannot use SD to prove this. Since
SD does not exist for RW, M A , MAM and SPW we need to
estimate the risk preferences of the decision-maker. (For a
review of risk preference see, for example, Howard and
Matheson [Ill.) Suppose, for the sake of illustration, that
the decision-maker has the preference curve shown in Figure
5. Table 4 shows the certainty equivalents for the five assembly systems. The best assembly system is the assembly machine with robot workheads, followed by the robot assembly
cell, manual assembly, manual assembly with mechanical assistance and the assembly machine with special purpose workheads.
iii) Informational Phase
In this phase the uppermost question is; "Should more
information be gathered before the final decision is made?".
A special case is the case of perfect information-that is, the
value of an important stochastic variable will be known exactly after perfect information is gathered. The value of perfect information provides an upper limit on the amount of
resources that can be expended, in further studies and analysis, to reduce the uncertainty in a variable. The value of
information is the difference in the output variable between
the best decision without information and the best decision
with new information. The information has no value if it
results in the same decision as in the without information
case. (See Raiffa [IS].)
In the illustrative example the value of perfect information
is zero for all the important stochastic variables (considered
alone). Eliminating the uncertainty in any one stochastic variable does not change the decision. The best assembly system

Assembly
System
MA
MAM
SPW
RW
RAC

228

AE(Annu1ty Equivalent), Ernlllion

Figure 5. A Typical Preference Curve for - 2 s AE s 3 ($million)
is still RW. However with perfect information about MS and
M G R O W (that is-the total demand) the best FMS might be
RAC. The value of perfect information about MS and
M G R O W is
A E (perfect information about MS and MGROW) -AE
(no information) = $908450 - $891049 = $17401
where the first term is obtained by revising the original decision tree and the second term is the expected A E of the
best decision from the original tree (Table 4).
iv) Decision
If all uncertainty could be eliminated about future demands, at a cost of less than $17401 then it would be profitable
to do the necessary studies. A study which would eliminate
part of the uncertainty would be worth less. It is unlikely that
such studies could be done for less than $17401. If this is the
case, the final decision should be made. In this example the
"best" decision is to use a robot assembly machine (RW) to
manufacture all assemblies.

Summary and Extensions

This paper set out to show the proper methodology for
using traditional financial evaluation models to evaluate flexible manufacturing systems. Specifically, we set out to select
the "best" flexible manufacturing system for a particular situation. There was considerable uncertainty about the amount
of flexibility that was required (demands, new products, and

Table 4. Expected Preferences For Five Assembly Systems

Annuity Equivalent
Expected
Mean
Variance
Preference
$365093
363141
- 659597
891049
407069

+ 10
10
+ 11
+ 11

3.650E
8.357E +
2.808E
3.696E
9.770E

10

0.5716
0.5705
0.3454
0.6673
0.5790

Certainty
Equivalent

Rank

361,982
356,459
-681,874
862,123
399,259

3
4
5
1
2

IIE Transactions, June 1987

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

product changes), about the financial variables (discount rate,

tax rate, etc.) and about the production variables (assembly
times, quality, etc.). Simple mathematical models of five potential FMS's were developed. The important stochastic variables were identified, and estimates for all variables were
obtained from experts. Cumulative density functions of the
annuity equivalent for each potential FMS were evaluated
and the "best" FMS was selected. An example was worked
and it showed that the best FMS was a robotic assembly line,
while the next best FMS was a robotic assembly cell.
This methodology also gives valuable insight into the
amount of flexibility that is required, the important decision
variables, and the value of doing studies and gathering additional information. In another project we are investigating
the general conditions under which certain types of FMS's
are "best". We believe that this paper and our other work
show that traditional financial evaluation models, if properly
used, do encourage the adoption of robotic FMSs.
Possible extensions to this work would be:
i) Add lot-sizing and scheduling considerations to the mathematical models for the five FMS's. (It should be noted that
these considerations would make the robotic assembly line,
and the robotic assembly cell even more attractive. Setup
times are very small for robotic assembly lines and cells and
so small-lot production is profitable.)
ii) Take account of the dependencies (correlation) between
the stochastic variables; especially between the demands for
various assemblies, and between the demands and the financia1 variables.
iii) Consider mixed manufacturing systems. For example, use
a robotic assembly cell for assemblies 1, 2 and manual assembly for assembly 3, etc.
iv) This paper has considered only the financial aspects.
There are other factors which will also influence the firm's
choice of flexible manufacturing system. These factors include
strategic considerations (such as the direction in which the
firm's industry is going, domestic and foreign competition and
markets, and the nature of the firm's labour relations); quality
considerations; the desire to develop high-tech expertise
within the firm; and so on. Some of these factors can also be
modelled and included in the analysis.
ACKNOWLEDGEMENTS
This research was supported, in part, by grant A5474 from
the Natural Sciences and Engineering Research Council of
Canada.
REFERENCES
[l] Arbel, A. and A. Seidman, "Performance E~aluationof Flexible Manufacturing Systems". IEEE Trruuacrionr (SMCJ, 14. 606617 (1984).
[2] Bawa, V. S., "Stochastic Dominance: A Research Bibliography", Management Seienee, 698-719 (1982).
131 Bwthroyd, G., "Emnomia of Assembly Systems", l o u r ~ of
l Monufanuring System, 1, 111-127 (1982).

June 1987, IIE Transactions

[4] Browne, J . , Dubois. D., Rathmill, K., Sethi, S., and K. E. Stecke,
"Classification of Flexible Manufacturine
T k FMS Moxarino,
- Systems",
.
114-117 (1984).
[S] Burstein, M. C, and M. Talbi, "Emnamic lustlflcation for the Introduction o f Flexible Manufarmring Technology: Traditional Procedures
Venus a DynamicrBased Approach", Proceedings of tkr First ORSAI
TIMS Conference on FMS, 1W-106 (1984).
161 Buzaeott, I. A,, "The Fundamental Principles of Flexibility in Manufa:l~rtng Sys~ernr",Proceed.ngr of rhr t 8 r r 6 fncemononal Conicrenrr on
F.V.+Orrohrr 1082 honh Holland Yuol,rhlng Co., Kru York
[7] Copeland, T. E. and 1. F. Weston, Fimncwl Theory and Corpomle
Policy, Addison-Wesley Publishing Co.. Reading, Masachusens,
(1984).
[8] Dod, R. C., Robots and Automared Mnnufmturvlg, Restan Pubbrhing
Co., Reston, Virginia (1983).
[9] Gitman, L. 1, and 1. R. Forrester, "A Survey of Capital Budgeting
Techniques Used by Major U.S. Arms", Fimnciol Mamgcmenr, 66-71
(1977).
[lo] Gold, B., "CAM Sets New Rules for Production", Howard Business
Review, 88-94 (1982).
[ l l ] Howard, R. A, and J. E. Matheson. Rcodings on the Principles and
Applications of Deckion Annfysis. Strategic Dceisionr Group, Menlo
Park, California (1983).
[I21 Kaplan, R., "Measuring ManufacturingPerformance: A new Challenge
for Managerial Accounting Rerearch", The Accounting Review, 686-705
(1983).
1
Knott. K. and R. D. Getto, "A Model For Evaluating Robot Syrtems
Under Uncertainty", Internotional lourno1 of Production Ruearch, 20,
155-165 (1982).
[14] Kulatilaka, N., "Financial, Economic and Strategic Issues Concerning
The Decision To Invest In Advanced Automation", I n r c r m t i o d l o u w l
of Produetion Research, 22, 949-968 (1984).
[IS] Michael, 0. I. and R. A. Millen, "Economic Justi6cation of Modem
Computer-Based Factory Automation Equipment: A Status Repon",
Proceedings of the Fin1 ORSAITIMS Conference on FMS, 30.35 (1984).
1161 Operations Research Saciety o f America, and The Institute of Management Sciences, Proceedings of the First ORSAITIMS Confernnee on
FMS-Operdonr Resenreh Models and Applicotionr-August 1984.
[I71 Proceedings of the Firstlntermtionol Confererne on FMS-October IW.
Nonh Holland Publishing Co., New York.
[IS] Raiffa, H., Deckion A d y s k , Addison-Wesley, Reading: Mass. (1968).
[I91 Schall, L. D., G. L. Sundemand W. R. Geusbeek, "Survey and Analysis
of Capital Budgeting Methods", The lourml of F i m n , 33, 281-287
(1978).
[20] Surerh, N. C. and 1. R. Meredith. "A Generic Approach to Jurtifying
Flexible Manufacturing Systems", Proceedings of the First ORSITIMS
Conference on FMS, 36-42 (1984).
[21] Suri, R. and C. K. Whitney, "Decision Suppon Requirements in Flexible
Manufacturing", Iournal of Mvwfocruring Systems, 3.61-69 (1984).
[22] Suri, R. and R. R. Hildebrant, "Modelling Flexible Manufacturing Syr
temr Using Mean-Value Analysis", lournnl of Monufacruring System,
3, 27-38 (1984).

Appendix 1-Financial

Model

The inputs to the financial model are the demand variables,

engineering variables and financial variables, while the outputs are the NPV and AE, IRR, and PB for each FMS.
Exogenous Variables

NP,-number of different assemblies to be manufactured

in year r
MSj,l--total market size for assembly i in year 1 (in units
demanded per year), i = 1, 2, 3, . . . , NPI
M S H A R E t s h a r e of the assembly i market for the fmn
229

MGR0W;-annual
1,-investment

market growth rate for assembly i

in year t, t

1, 2, 3,

...

N--useful life of the project (for NPV calculations)

D-life

of the project for depreciation purposes

WT,-total

wage costs in year t

SPj,--selling price for assembly i in year t

TX-tax
k-risk

rate

vestment costs. (That is overhead costs are ignored.) Depreciation is calculated using the double declining balance
method (equation A2). Finally, notice that the last term in
the NPV and IRR equations is the discounted book value
(BV) of the project at the end of its useful life. This simple
model can be reduced or enriched (in terms of the number
of variables and relationships) to meet the needs of a particular firm.
Appendix 2-Mathematical

Models for FMS's

free interest rate

Define the following variables.

Endogeneous Variables
MSi,,-total market size for assembly i in year t

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

VSi,,-annual sales of assembly i in year t by the firm

REV;-total

revenues in year t

DEP,-total

depreciation in year t

BVt-book

TM - time to assemble one part manually with mechanical

assistance

TW - time to assemble one part with specialpurpose work-

value of the project in year t

head (nonprogrammable automation)

ATCF,--after tax cash Row in year t

TR

TD - downtime at a station due to a defective part

VSr,, = MS,,, * MSHARE,

NPt

2 VS,, SP,,,

('41)

,=I

DEP,
BVo
ATCF,

' I-'

2 I, -

i=o

,-I

j;o

DEP,)

+ I,

+ DEP,TX - 1,

ATCF,

('4-2)

(ii) EQUIPMENT COSTS (dollars)

CB - cost per station of transfer device on manual assembly system
CT - cost per station of transfer device on automated assembly system
CC - cost of carrier (pallet)

0, BV, = BV,-1 - DEP,

(REV, - WT,)(1 - TX)

- time to assemble one pan with robot workhead (programmable automation)

Relationships Between Variables

MS;,, = MS;,r-l (1 + MGROW;)

REV, =

Exogenous Variables
(i) TIMES (seconds)
TA -time to assemble one part manually

BVN

NPV
AE =
1
I-(1 + k)N

CF-

cost of mechanical feeding device for manual and

non-programmable assembly systems

(A3)

CM - cost of part magazine (feeding device) for robot systems

(A41

CW - cost of special purpose workhead

C1 -basic

(A5)

cost of robot

C2 - additional cost of robot per degree of freedom

CG - cost of robot gripper

IRR is calculated tom

(iii) PLANT VARIABLES

P E - plant efficiency (fraction of available time worked)
PQ - part quality (fraction of defective parts to acceptable
parts)
NR - number of assembly workers per station on automated assembly systems

PB satisfies
PB

2 [(REV, - WT,)(l - TX) - It] = 0

1=

(A7)

NOS - number of supervisors on an assembly system

HR - number of production hours per shift

For simplicity it is assumed that the risk free rate k (used to

discount the cash Rows) is constant over the life of the project.
The total cost per year is made up of labour costs and in230

WA - annual cost of one assembly worker

WS - annual cost of one supervisor
IIE Transactions, June 1987

(iv) PRODUCTION VARIABLES

NS; - the number of parts assembled at each station for
assembly i
NAj - number of parts in assembly i, i

1, 2, . . . , NP.

NT; - number of parts in assembly i that are differentfrom

assemblies 1, 2, . . . , i - 1
ND;, - number of part design changes for assembly i in
year t
Endogenous Variables
TPi - the average production time for a non-defective assembly i
NSR - number of stations required in the assembly system

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

S H - number of production shifts for assembly system

Mathematical models for the first five assembly systems of
Figure 1 will now be described. They are based on models
developed by Boothroyd 131. In what follows the subscript r
(year) will be suppressed for clarity.
MA - MODEL FOR MANUAL ASSEMBLY (Figure 1-a)
Select the required number of shifts SH, and the number
of parts to be assembled in each station NS, for each assembly
i (i = 1, 2, . . . , NP), so that there is sufficient capacity to
NP

meet the production requirements. That is:

1VS; * TP, s

SPW-MODEL FOR ROTARY AND IN-LINE INDEXING ASSEMBLY MACHINES (Figure 1-c)
These assembly machines are non-programmable. Therefore each assembly requires a dedicated machine with diferent special purpose workbeads, different work carriers and
different part feeders. Each station in the assembly machine
assembles one part only. (Obviously, for such "hard automation" to be profitable, the volumes must be very large.)
For each assembly i, (i = 1, 2, . . . , NP) select SHi, so
that there is sufficient capacity to meet the demand requirements. That is VS;'TP;I3600 s HRSPE*SHwhere TPj =
TW iNAj*PQ*TD. Each station in the assembly machime
consists of work carriers (CC), a feeding device (CF), and
a special purpose workhead (CW) (all of which must be
changed if a part is redesigned) and a transfer device (CT).
The initial investment required for assembly machine i is
Ii = NAja(CC + CF
CW CT). The labour wsts consist
of operator and supervisor costs. One operator runs 1INR
stations on the assembly machine and one supenisor monitors
llNOS assembly machines. The annual labour wst for machine i, is:

WT;

NP

WA

The initial investment in all NP indexing assembly ma-

NSR*(NP*Z*CC + CB)

+ CF'

NP

1 NE.

,=I

June 1987, IIE Transactions

NP

chines is I

NP

WT =

I; while the total annual labour cost is

;=I

WT;.

i=l

SPW-MODEL FOR IN-LINE FREE TRANSFER (BUFFERED) ASSEMBLY MACHINES WITH SPECIAL PURPOSE WORKHEADS (Figure 1-d)
If the number of parts in assembly i (NA;) is large, downtime due to defective parts, can be excessive when rotary and
in-line indexing assembly machines are used. In this case an
in-line assembly machine with buffer inventories between the
stations will be used. Boothroyd [3] recommends the in-line
free transfer assembly machine be used whenever NAi 3 10
and that the size of the buffer inventories (BI) be approximately [O.S*TDITW]where [XIis the smallest integer greater
than or equal to X.He estimates that, on average, a good
assembly is produced every TW PQ'TD seconds. For each
assembly i, where NAi 3 10 (i = 1, 2, . . . ,NP), select the
required number of shifts SH; such that VSj*TPjB600 s
HR*PE*SH;, where TPi = TW + PQ'TD. The initial investment for assembly machine i is I; = NA;*(CC + CF +
CW + CT t BFCC), and the annual labour wst is WT; =
(NRXNA;*WA+ NOS*WS)*(VS;*TPjB600)/(HR*PE).
(If a
part is redesigned then a feeding device (CF) and a special
purpose workhead (CW) will have to he changed.)

1 N R VS, TP;

MAM-MODEL FOR MANUAL ASSEMBLY WITH MECHANICAL ASSISTANCE (Figure 1-b)

The difference between this model and the previous model
is that mechanical feeding devices are available at each station
to speed up the assembly process. (A different feeding device
is required for each different part. When a part is redesigned,
a new feeding device is required.) The equations are as above,
except that TPj = NSjbTM*(l + PQ), and
I1 =

(NR*NAiYWA
t NOS*WS) *(VSjaTPjI%OO)/(HR'PE).

;=I

HR * PE * SH where TP; = NSj*TA(l + PQ). (Every TP

good assemblv will be ~roducedbv
seconds. on average,
- . a the assembly machine.) The number of stations required
(NSR) in this manual assembly system is: NSR = max [NA,I
NS,] (i = 1, 1, . . . ,NP). Assume that there is one operator
at each station and that each operator has access to two work
carriers (CC) on the conveyor (to allow for minor delays)
and that each station has a transfer device (CB). Finally,
notice that only the work carriers have to be changed when
the assembly system changes over to a different assembly.
The initial investment for a manual assembly system is, therefore:
I, = NSR(NPS2*CC + CB). The total annual labour costs
are:

RW-MODEL FOR IN-LINE FREE TRANSFER (BUFFERED) ASSEMBLY MACHINE WITH ROBOT WORKHEADS (Figure I-e)
This assembly machine is the same as the in-line free transfer machine with special purpose workheads except that the

231

workheads are now programmable. That is, the robot workheads can be reprogrammed for different tasks. (In that sense
this assembly system is similar to manual assembly with mechanical assistance-model 2, because the human operators
can also do different tasks.) The cost of each robot is CR
where CR = C1
D F C 2 . C1 is the basic cost of a robot
and C2 is the additional cost per degree of freedom of the
robot. It is reasonable to assume that four degrees of freedom
are required for assembly Line robots.
Select the required number of shifts SH, and the number
of parts to be assembled at each station NS,, for each assembly
i, i = 1, 2, . . . , N P such that there is sufficient capacity

Appendix 3 Most Likely Values and Ranges for Variables

Defined in Traditional Financial Valuation Models
Deterministic Variables (Most Likely Values)

CB = $7000
C T = 16000
C C = 1500
C F = 7000
CM = 5000
CW = 10000
C1 = 50000
C2 = 10000
CG = 1700
WA = 40000
WS = 55000
HR = 2000
NR = 0.333
NOS = 0.25
D = 10

NP

to meet the production requirements. That is,

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

VSi*TP;I

i-1

3600 S HR*PEaSHwhere TPi = NSiX(TR PQ*TD). The

number of stations required in the assembly machine is,
NSR = max[NA;lNS;], i = 1, 2, . . . , NP. Again we have
buffers of size BI = [0.50*TDITR] located between all stations. The required initial investment is
NP

I,

NSR*(CR

+ CT) + C M * Z NT;
i=l

,VP

+ (CG + CC(1 + 81))' 2 (NA,INS;).

i=l

If design changes are made then the robot part feeders (CM)
and the robot grippers (CG) need to be changed. Labour
costs consist of operator and supervisor costs. (Notice that if
assembly i has NA; parts and NS, parts are assembled at each
station, then NAjINSi stations are required to complete the
assembly.) Therefore the annual labour costs are;

6 parts

6 parts

6
7
8
9
SP,, = $0.65 i = 1 , 2 , . . . , 9 r = 1 , 2 , . . . (For simplicity,
the selling price for all assemblies is the same.)

IIE Transactions, June 1987

Stochastic Variables
Variable

NPI

Most Likelv Value

number of different assemblies to be manufactured in year 1

NAP, = number of new assemblies to be manufactured
in year t
=

t=2...I
t = 3 ...0
1 = 4 ...I
t = s . . .0

i = 1
i =2
i =3
i =4
i =5
i =6

Downloaded by [University Aut Ciudad Juarez] at 04:08 02 April 2013

MS<,I= total annual demand for item i in year 1

750000
900000
700000
550000

300000
500000
300000
800000
750000

i = 7

i = 9

MGROWi
MSHARE;

0.10
0.17

Project Life

TX
k

Range

i = l , 2, . . . , 9
i = l , 2, . . . , 9

MA
MAM
SPW
RW
RAC

Tax Rate
= Risk free interest rate

6
7

8
0.40
0.10

TA
TM
7w

10 seconds
8
5

TR

TD

30

PE
PQ
NDi,

4 years
5

Annual Part Design Changes

1'

0.95
0.05
i=1,2, ...,9

*Each assembly will have 1 pan design change every year, for 5 years, beginning in the next year after it is introduced.
Probability Distributionsi For Important Stochastic
Variables
Variable

(Value; Probability),

NPt
NAP,
MS,,I

(5; 0.6)
( 4 ; 0.4)
(1, 0, 1, 0; 0.6)
(2, 1, 1, 0;0.4)
(Most Likely Value from above; 0.5)
(Low Value; 0.3)
(High Value; 0.2)
(0.10; 0.50)
(0.07; 0.50)
(4, 5,6, 7 , 8; 0.5)
(7, 8, 9, 10, 12; 0.5)
(0.10; 0.60)
(0.12; 0.40)
(10, 8, 5,6; 0.6)
(9,7, 4,5; 0.4)
(0.05; 0.60)
(0.035; 0.4)
(1, 1 , . . . ; 0 . a )
(2, 2, . . ;0.40)

MGROWi
Project Life
k
TA, TM, 7W, TR
PQ
ND

(Value; Probability), . . .

Dr. G. John Miltenburg is Associate Profernor of Production and Managemen1 Science, in the Faculty of Business, at McMa~terUniversity, Hamilton,

June 1987, IIE Transactions

Ontario, Canada. Prior t o this, he worked in Manufacluring Engineering at

General Motors. John holds a Ph.D. horn the University of Waterlw. His
research interests indude the development of optimal and heuristic approaches
for analyzing mmplex inventory and production problems; including c w r dinated control of inventories, transfer lines, flexible manufacturing systems
and just-in-time production systems. Dr. Milfenburg has published anider in
l
of Production Research, INFOR, Engineering Costs
the I n t e r ~ ~ i o n oJournal
and Production Economies, IIE T r a m l i o n s and Novol Rcswrck Lopistier
Quanerly.

Dr. Itrhak Krinsky is an A~nociateProfessor of Finance and Business b

nomia, in the Faeully of Business,McMaster University, Hamilton, Ontario,
Canada. Dr. Krinsky has conducted research in insurance, portfolio analysis,
financial markets, and econometrics. His current research interens include the
development of prediction models for mergers and acquisition, calculating the
mst of equity capital for insurers and the Statistical pmpenies of elasticities.
Dr. Krinsky has published articles in the J o u m l o f Fimneiolond Quonrilotive
Analysis, Review of Economic$ and S!mislics, Journal of Risk and Insurmce,
Applied Economics, Journol of Inrrr~lionalMoney and FiMnce and athern.

233