Você está na página 1de 30

This article was downloaded by: [FU Berlin]

On: 02 May 2015, At: 14:58


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

The Engineering Economist: A Journal Devoted to the


Problems of Capital Investment
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/utee20

PRODUCT AND PROCESS COST ESTIMATION WITH FUZZY


MULTI-ATTRIBUTE UTILITY THEORY
a a a a
P.-K. DEAN TING , CHUCK ZHANG , BEN WANG , ABHIJIT DESHMUKH & BARBARA
b
DUBROSKY
a
Florida State University
b
Pratt & Whitney
Published online: 31 May 2007.

To cite this article: P.-K. DEAN TING , CHUCK ZHANG , BEN WANG , ABHIJIT DESHMUKH & BARBARA DUBROSKY (1999)
PRODUCT AND PROCESS COST ESTIMATION WITH FUZZY MULTI-ATTRIBUTE UTILITY THEORY, The Engineering Economist: A
Journal Devoted to the Problems of Capital Investment, 44:4, 303-331, DOI: 10.1080/00137919908967526

To link to this article: http://dx.doi.org/10.1080/00137919908967526

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
PRODUCT
AND PROCESSCOSTESTIMATION
WITHFUZZYMULTI-ATTRIBUTE
UTILITYTHEORY

P.-K. DEANTING,CHUCKZHANG,
BENWANG,
ANDABHIJITDESHMUKH
~ o h d State
a University

BARBARADUBROSKY
Downloaded by [FU Berlin] at 14:58 02 May 2015

Pratt & Whitney

The need to provide cost-effective manufacturing technoiogies is critical in today's globally


competitive environment. The ability to make sound decisions about the cost effectiveness
of new technologies is the key to achieving the affordability goals of future products and
systems. Effective assessments of manufacturing technologies for future investment and
applications must reflect the total financial impact in trade-offs between cost and
performance.

Most of the traditional tools for cost modeling, reasoning, and computing are crisp,
deterministic, and precise in character. By crisp we mean dichotomous, that is, yes-or-no
types rather than more-or-less types. In the real manufacturing environment, many
parameters have an uncertain nature. The structures and parameters of the model are not
definitely known in early design stages of, in particular, products that do not have a
manufacturing base. The goal of this study is to develop a set of efficient procedures for
providing the manufacturing industry with the ability to assess and evaluate the cost of new
technologies and affordabilityat early productengineeringdesign stages.This model, termed
"fuzzy muld-attribute utility (FMAU)," can be applied to evaluate the cost index for each
product. In addition, an optimization method is introduced for fine-tuning the cost model
with historical data. The relationship between real cost and cost index can be determined by
the empirical function. Also, this method can reduce the estimating error from subjective
judgments by the tuning process. The proposed method is illustrated with an example of cost
estimation for an ingot process in propulsion part manufacturing.
A systematic accounting of incomplete or conflicting information, constraints, and
consequence has not been adequately addressed in the manufacturing industry
during engineering design and costing. In the case of incomplete information,
commonly used parametric techniques provide limited assessment capabilities. The
use of probabilistic evaluations has improved the handling of certain types of
uncertainty and has recently been applied in industry for the handling of likelihood.
However, its use is limited to cases where an adequate historical base exists upon
which to apply probabilistic studies.
A new methodology is needed to provide an effective assessment of advanced
Downloaded by [FU Berlin] at 14:58 02 May 2015

technologies, as well as the uncertainty. The use of fuzzy attributes for cost models
and affordability applications addresses the problems of ( I ) the limited data in the
situation of materials with limited characterization data, (2) processes with limited
empirical data, and (3) manufacturing processes with little or no manufacturing
base. This paper presents a method of integrating a fuzzy costing method and
expert opinions to develop a cost model, namely "fuzzy multi-attribute utility
(FMAU) model," for incomplete, uncertaindata. Existing cost modeling techniques
are reviewed, an overview of the multi-attributeutility theory (MAUT) is provided,
details for the development of acost model based on MAUTand fuzzy sets for cost
estimation are given, and an example and concIusions are presented.

Deliveringreliable, high-quality products and processes at low cost has become the
key to survival in today's global economy. Driven by the need to compete on cost
and performance, many quality conscious organizations are increasingly focusing
on the optimization of product design. This reflects the realization that quality
cannot be achieved economically through inspection. Designing for cost and
quality is more effective than trying to inspect and re-engineer it after a product hits
the production floor or worse, after it gets to the customer. Thus, the early design
phase of a product or process has the greatest impact on life-cycle cost and quality
[6, 11, 16, 23, 261. Therefore, cost estimating in the early design phase is
indispensable to achieving significant cost savings and improvements in quality.
In industrial practice, activity-based costing (ABC), parametric cost estimating
(PCE) and function cost estimating (FCE) are the three technologies widely used.
ACTIVTY-BASED COSTING .
Conventional approaches to product and service costing have been subject to
sustained and varied criticism in recent years [13]. Key limitations pinpointed by
critics have included:

Lack of visibility given to increasingly important areas of resource


consumption [22];
Systematic distortion of product costs caused by the perpetuation of
convenient but inappropriate volume-based' methods of production
overhead absorption [4];
Neglect apparent in the development of suitable means of unitizing non-
Downloaded by [FU Berlin] at 14:58 02 May 2015

production overheads [ 5 ] ;
Absence of reliable long-term variable cost information for strategic
decision-making [15];
Lack of costing data availability at the preproduction stages in the product
life cycle [I].

Against this background, ABC has emerged during the last decade as one
means of addressing these shortcomings. ABC is a cost planning system for
portraying the organization. Activities within an organization are identified and an
average cost is associated with each activity. The total cost of a product is the sum
of the costs of activities required to bring forth, sustain, and retire the product. The
cost of an activity for aproduct is the average cost of the activitymultiplied by the
number of times the activity is required for that product. ABC systems were
originally developed to obtain improved product cost information.Recently, other
applications have been described in the accounting literature, including cost control .
and process improvement.

PARAMETRIC COSTE S T I M A ~ G
PCE is the mathematical procedure or process in which product or service
descriptors directly yield consistent cost information. A parametric cost model is
an estimating system comprised of cost-estimating relationships (CER) and other
parametric estimating functions, e.g. cost quantity relationships, inflation factors,
staff skills, schedules, etc. Parametric cost models yield product or service costs at
designated cost or work breakdown structure (CBSNVBS)levels and may provide
departmentalized breakdowns of generic cost elements. PCEprovides a logical and
repeatable relationship between input variables and the resulting costs.
PCE has been developed since the late 1940s. The state-of-the-art in
parametric estimating has been steadily improving by an explosive growth in the
number of practitioners, important methodological improvements, and greatly
expandeddatabases. It is widely used in many different fields. Melin [20] designed
a parametric estimation method to structure the detailed line items of completed
projects into smaller groupings, or assemblies, with a relationship to known
variables, or parameters. Through these parameters, such as gross square footage
or length of pipe, a new detailed estimate can be generated from an old estimate.
Son [25] used various approaches to collect pkarnetric values of the cost elements
in the advanced manufacturing systems. Yan et al. [28] described parametric
approaches used in the conceptual design of mechatronic products.

COSTESTIMATING
FUNCTION
Downloaded by [FU Berlin] at 14:58 02 May 2015

FCE,as defined in value engineering, is an accounting allocation of cost and


importance to product function. Value engineers use it to determine thecost of each
product function. In the United States, the function costs are typically determined
by a bottom-up analysis of the cost of all items, components, labor, etc., necessary
for the product to do each function. In Japan, cost tables are often used [7]. FCE
focuses on costing the components of service (known as functions) that products
or services (including internal services) offer to customers. It involves a team
approach allowing the organization to draw on the skills of a variety of business
disciplines(includingsupplier andcustomer intelligence) to help meet the strictures
of tightly set target costs.
The practice of FCE applied to products is described by Yoshikawa etal. [29].
It involves expressing each function as a verb and a noun. The functions, which
become the focus of costing activity, thus provide an abstract view of what the
product offers to the customer. French and Widden [lo] quoted rolling-element
bearings and squirrel-cage induction motors as an application of FCE. In
applications involving products with a high proportion of purchased components,
functional costing seems likely to be a very useful technique for early cost
estimation.
In the last sections, three commonly used methods for cost estimation have
been briefly reviewed. Although each of these methods has greatly enhanced the
industry's ability to provide cost estimations, each one has several certain
weaknesses [13, 141. The major shortcoming is their inability to handle cost
estimation with uncertain and incomplete data. To overcome this drawback, a new
method for product cost estimation has been developed in this study.
In practice, cost estimation often involves handling of data that may be uncertain
in nature and the information associated with each cost factor may be qualitative.
For example, the cost of a product may be related to the availability of tools. This
information is fuzzy in nature. To accurately estimate costs, designers need to
develop techniques that reduce the inherent inaccuracies and subjectivity of the
data. Utility theory and related assessment techniques can overcome these problems
in the cost estimation of products. Moreover, the mathematics behind these tech-
niques is based on amore general object, the binary relation that makes the method
capable of comparing different scenarios.
Downloaded by [FU Berlin] at 14:58 02 May 2015

Let R be a general binary relation, and X a set with general elements x and y.
If R is negative transitive and weakly connected, and the set is not uncountably
large, then a real-value function exists, such thatxRy if and only if U(x)> UCy) [9].
We may equally define "R" to represent "is moreexpensive than" and the term "U"
as a'cost function. Therefore, utility theory generalizes easily to such diverse
problems as quality indexing,organization performance measurement, and personal
evaluation with suitable substitution of the appropriate professional judgment for
the binary relation R.

IDEASOF MULTI-AITRIBUTE UTILITYTHEORY


What is MAUT, and how does it relate to other assessment approaches? Edwards
et al. [8] discussed that question in 1975. MAUT depends on a few key ideas:

1. When possible, evaluations should be comparative;


2. Programs normally serve multiple constituencies;
3 Programs normally have multiple goals, not a11 equally important;
4. Judgments are inevitably a part of any evaluation;
5. Judgments of magnitude are best when made numerically; and
6. Evaluations typically are, or at least should be, relevant to decisions.

Some of those six points are lessinnocent than they seem. If programs serve
multiple constituencies, evaluations of them should normally be addressed to the
interests of those constituencies; different constituencies can be expected to have
different interests.
The willingness to accept subjectivity into evaluation, combined with the
insistence that judgments be numerical, serves several useful purposes. First, it
partly closes the gap between intuitive evaluation and the more quantitative kind.
Indeed, it'rnakes coexistence of judgment and objective measurement within the
same evaluation easy and natural. Second, it opens the door to an easy combination
of complex concentrations of values. For instance, evaluation researches often
distinguish between process evaluations and outcome evaluations. Process and
outcome are different, but if a program has goals of both kinds, its evaluation can
and should assess its performance on both. Third, use of subjective inputs can, if
need be, greatly shorten the time required for an evaluation to be completed. A
MAUT cost evaluation can usually be completed in a shorter period of time than
conventional cost assessment methods. The inputs to such an abbreviated evalu-
ation activity will obviously be almost entirely subjective. But the MAUT
technique at least produces an audit trail such that the skeptic can substitute other
Downloaded by [FU Berlin] at 14:58 02 May 2015

judgments for those that seem doubtful, and can then examine what the conse-
quences for the evaluation are. When time constraints on the decision made are
necessary, the MAUT can quickly provide some orderly basis for decision making.

CURRENT OF MAUT
APPLICATIONS
A number of researchers have applied MAUT to various fields. Keeney et al. [17]
used MAUT to construct a cost-equivalent function and evaluated the benefits of
four alternatives to improve electrical system reliability. Metri [21] presented a
semantic network which was merged with a decision-theoretic MAUT to provide
an indexing technique for databases that store information in the form ofjudgrnents
or evaluations. Reed et al. [24] applied MAUT to public transit system designs.
Locascio and Thurston [19] employed MAUT to construct the mathematical basis
of modeling competing design tools simultaneously and to make rational decisions
to affect overall design improvement. A11 these studies have shown the effective-
ness of the MAUT in various applications.

STEPSIN A MAUT EVALUATION


MAUT has been researched for many years. The main steps involved in applying
MAUT to product cost estimation are as follows:
Step 1:Identify the objects of evaluation and the functions that the evaluation
is intended to perform. In this case, the objects and the functions are propulsion
products and manufacturing processes, respectively.
Step 2: Identify a set of attributes that contributes to the overall product cost.
Proper selection of attributes will increase the accuracy of cost prediction. In this
paper, we assume that all attributes are preferential and utility independent. To
illustrate the above conditions, suppose that four attributes for a given product cost
estimation are profitability, flexibility, quality and throughput. Preferential inde-
pendence means that if we judge quality, for example, to be more important than
flexibility, this relation is true regardless of whether the level of profitability is
high, low, or some value in between and also regardless of whether throughput is
good, bad, or somewhere in between. Utility independence means if we are
deciding on the preference order for probabilistic trade-offs between, for example,
quality and flexibility, that can be done regardless of whether the level of
profitability is high, low, or some value in between and also regardless of whether
throughput is good, bad, or some amount in between [3]. In practice, if some
attributes are not preferential and utility independent, then the final result can be
discounted to reflect these assumptions. Keeney and Raiffa [I81 discuss how an
analyst, through discussion with the decision maker, could determine whether the
independence between attributes holds in a particular problem.
Step 3: Construct a utility function for each level of attributes. Many
Downloaded by [FU Berlin] at 14:58 02 May 2015

approaches have been proposed for constructing a utility function. The first
approach used to estimate the utility function was pioneered by von Neumann and
~ o r ~ e n s t e [27].
r n The basic type of utility functions is

where U ( x : ) is the utility function of attribute i at A level, x f is the highest


level giving rise to highest cost of attribute i and xy is the lowest level leading to
lowest cost of attribute i . The questioner elicits the probability p such that xi
is the certainty equivalent of ( ( x f , X ~ ) ; (-~p, )I) .
Step 4: Design two types of preference-query questions. These questions are
used for evaluating weights of each attribute. Keeney and Raiffa [18] designed two
types of questions for assessing the weight ki for each attribute i . Suggested
good practice in assessing the k 's would be first to rank all attributes, then to use
Question A to evaluate the largest k i , and then to use Question B successively to
evaluate the magnitude of the other k i 's relative to the largest ki , or relative to
any other k i that has been determined.
Question A: Decide the weight of the most significant factor. For the
probability p are you indifferent between:
*
1. The lottery with a probability of winning, p , for x * = ( x , ,x2,.. . , x i )
9

and probability of winning ( I - p ) for x0 = ( x p , x;, ..., x,0 ) , and


* ,. ..,x,0 ) .
2. The consequence xl0 ,... , x i0-,, xi+,
The result of such an assessment is that k , = p .
Question B: Decide the weight of all other factors except the most significant
factor. Select alevel of x i (e.g., x[ ) and a level of x (e.g., x J ) such that you are
indifferent between:
I. An outcome yielding x f and x 9 together; and
6
2. An outcome yielding XIand xi together
Thus one can use the relation kiU ( x l ) = k ,U ( x J ) to solve for either k i or k
depending on that is unknown.
Step 5: Decide and compare the utility values for every alternative by using
the multiplicative utility model [18].

The model assumes the order of the overall scaling factor K of all attributes
is proportional to the order of the interaction term. Because utility value ranges
Downloaded by [FU Berlin] at 14:58 02 May 2015

from 0 to 1, the overall scaling factor K can be calculated by the following


equation from the individual scaling factor ki [18].

This will result in a cost index within the range of 0 to 1.


Step 6: Map the cost index on to the cost.
In the conventional utility theory, the decision maker's preference is taken as
utility values. The utility values can be obtained through the decision maker's
answers to the preference-query questions. This utility calculation is often done
based on the probability theory. However, the fuzzy set method may also be used
to acquire utility values, particularly when the information is uncertain [ 2 ] .

OF COSTMODELWITHFMAU
CONSTRUCTION

In the traditional MAUT approach, the decision variables are deterministic and the
utility values are crisp. Therefore, the general MAUT method is unable to handle
problems with incomplete and uncertain data. In industrial practice, practitioners
and experts often describe objects or events with uncertain information. For
instance, engineers may use a "fuzzy" term such as "very difficult" to describe the
degree of machining difficulty of a certain material. In other cases, they may give
a range of a certain parameter for describing an object. Therefore, in applying the
utility theory to costing, the experts may provide a range of utility values for a
specific cost driver level. Therefore, it is desirable to use the fuzzy set theory to
obtain a sensible result. Based on this understanding, a new method, called the
FMAU method, which combines MAUT and fuzzy set theories, is developed to
deal with the cost-estimation issue in evaluating new product and process design.
FIGURE1 shows the procedure for constructing an FMAU model.

FORMULATION OF THE FMAU MODEL


A classical set is normally defined as a collection of elements or objects x E X
that can be finite, countable, or overcountable. Here the utility value for each
element in a classical set is constructed by EQUATION I . If U ( x ) is a collection of
objects denoted generally by the'utility function, u ( x ) , then a fuzzy set A"
in
U ( x ) is a set of ordered pairs:
Downloaded by [FU Berlin] at 14:58 02 May 2015

where pi ( u ( x )) is called the membership function or grade of membership (also


degree of compatibility or degree of truth) of u ( x ) in A that maps utility value
to the membership grade. In fact, there exist mappings from crisp value to utility
value, then to membership grade for each attribute. Similarly, the procedurecan be
applied to the determination of the membership function of scaling factor for each
attribute l , p l ( k l ) . k l is the weight of the attribute I . The FMAU model is
expressed in EQUATIONS 5 and 6 ,

where p (U ( x ) ) is the membership function of cost index in EQUATION 5 . Because


the input variables, U ;(xi ) and k i of EQUATION 5, are in types of membership
function, the cost index can be calculated by the basic fuzzy arithmetic. The scaling
factor K calculating from EQUATION 6 can range the cost index from 0 to 1, so
kimax should be set as the maximum value of the membership function for each
attribute in EQUATION 6.

CONSTRUCTION
OF MEMBERSHIP
FUNCTION
Before a mathematical model is constructed to estimate the product cost, the cost
drivers that are to be estimated in the multi-attribute utility model should be first
determined. Pratt & Whitney summarizes their six major manufacturingprocesses
( Preliminary I Detailed Product Design I

( Rank Product Cost Drivers I


' Set Lower and Upper Bound for
Each Level of Every Cost Driver
Downloaded by [FU Berlin] at 14:58 02 May 2015

Decide Membership Function for


Each Level of Every Cost Driver
I

) the Weight of Each Cost Driver I


I
1 Fuzzy Multiattribute Utility Model I

FIGURE1. Procedures for Cost Estimation Using FMAU.


for certain forged products as ingot, extrusion, forging, heat treatment, machining
and miscellaneous. Different cost drivers are designed in each process based on
different considerations for material, processes and tools. Also, a comprehensive
table ("repertory grid"), which relates the major processing parameters to the cost,
is designed to elicit the membership function for each cost driver and rank cost
drivers in each process. Respondents or decision makers are responsible for
eliciting the utility range of each level and deciding the membership function for
each level and the importance of each cost driver based on their experience. The
original range of classification is the same for all respondents, but different
respondents who use EQUATION 1 and preference-query questions will produce the
different utility values of attribute in the same level and scaling factor of each
Downloaded by [FU Berlin] at 14:58 02 May 2015

attribute, respectively. The detailed repertory grid for each manufacturing process
is appended in TABLES 4-9.
After the repertory grids are collected, the membership functions for each
attribute and its scaling factor can be established. The method of integrating
respondents' opinions is based on the mean value. For example, for triangular
membership function, the minimum and maximum utility values of each level
among all responses are chosen as two bottom points and the mean value with the
membership grades of 1 among all responses as the top point. For trapezoidal
membership functions, the same procedure for the triangular membership function
is applied to decide the two bottom points, and the mean values with the
membership grades of 1 among all responses are set as the two top points.

DEFUZZIFICATION METHODS
When some historical data are available, regression and optimization methods can
be applied to predict the cost of the new product. First, a suitable defuzzification
method needs to be selected to convert each membership function that is expressed
in terms of a fuzzy set (EQUATION 3) into a cost index. Different defuzzification
strategies may produce different cost indices. An optimization method may be
needed to find the most proper strategy for product cost estimation. A number of
defuzzification methods leading to distinct results were proposed in the literature.
Each method is based on a certain rationale. The following three defuzzification
methods have been predominant in the application of fuzzy logic [12].

Center of Area (COA) Method


In this method, which is referred to as the center of gravity method or centroid
method, the cost index, C I ( A ) , is calculated as follows:
Center of Maximum (COM)Method
In this method, the cost index value, C I ( A ) , is defined as the average of the
smallest utility value and the largest utility value for which A ( z ) is the height,
) A .Formally,
h ( ~ of
Downloaded by [FU Berlin] at 14:58 02 May 2015

i n f M +sup M
dcm(C)=

z )h ( c ) ]
where M = [ z E [ o , ~ ] ~ A ( =

Bisector of Area (BOA) Method


In this method, the cost index is defined as C I ( A ) , the value for which the area
under the graph of the membership function A is divided into two equal subareas.
This value is calculated as follows:

N G FMAU WITH H~STORICAL


F ~ E - T U N I OFTHE DATA
Based on different considerations of manufacturing technologies, different
attributes are chosen to construct the cost models. MAUT is useful for adjusting
scaling factors and attributes to generate a standard cost index under different
manufacturing environments. Using fuzzy logic will reduce the bias from the
respondents. Accurate product cost evaluation requires some historical information
to detennine the CER between cost index and real cost. Real cost is a dependent
variable in CER;cost index is an independent variable. In practice, most cost data
can be fit empirically using one of the functions shown in TABLE1. All these
functions represent either straight lines or well-behaved curves. It is rare in cost
estimating to use CER that have inflection points. Use of such a model would
suggest that multiple values for the independent variables correspond to the same
manufacturingcost. Data indicatingsuch relationships are therefore suspicious and
need further investigation.
Downloaded by [FU Berlin] at 14:58 02 May 2015

1: Commonly Used CER Functions.


TABLE

Equation Type Linear Power Exponential Logarithmic

Equation F O ~ y =a+bx y =ax 6 y = a e bx y =a+bInx

LinearEquationForm y=a+bx Iny=Ina-I-blnx Iny=Ina+bx y=a+bInx


Coefficients a,b a,b a,b a ,b
Usually, the output value of the multi-attribute utility model is dependent on
the scaling factor and utility value attached to each attribute by decision makers.
Also, it will be influenced by the choice of the defuzzification method and the CER
function. Different strategies often providedifferent results for cost estimation. The
following two cases may arise while tuning the cost model:
Case I : If there are no historical data, the comparison analysis can be applied
so that the decision makers just compare the designed product with existing
products using similar material and manufacturing processes to estimate cost. Then
the decision makers evaluate the range where the product cost will be, and whether
it is in the acceptable range. However, another similar manufacturing process
should be chosen to evaluate the designed product cost if there is disagreement
Downloaded by [FU Berlin] at 14:58 02 May 2015

among decision makers.


Case 2: If there are some historical data, the optimization concept can be
applied to tune the CER. Solutions to the following optimization model yield the
best defuzzification method for the cost model, when the error, E, is minimized.

Min E
St.

where
THE ENGINEERING ECONOMIST* 1999 VOLUME 44 NUMBER4

n : number of points in the historical data set


m : number of cost drivers
d : index of the defuzzification method
P: number of defuzzification methods
a,b : parameters of the CER equation
DFd : defuzzification function
E: sum of squared errors for all data points based on the different
defuzzification methods
p,( k , ) : membership function of the individual scaling factor for the
attribute 1;
K . scaling factor
Downloaded by [FU Berlin] at 14:58 02 May 2015

p, (U, (G)) : membership function of utility value for attribute 1

In this model, the CER equation is assumed to be y = b x U . For each


defuzzification method, EQUATION 10 determines the errors by the regression
method. However, different defuzzification methods result in a different cost index
and parameters in CER.An optimumdefuzzification method can be determined by
minimizing the objective function.

For clarifying the methodology of cost estimation, an example is introduced below


based on the ingot process in the propulsion industry. TABLE 2 shows an example
of the evaluation of the scaling factor values and utility values for each cost driver
by the repertory grid. The utility value range for each level is determined by the
experts. All levels of each cost driver are classified by different normal triangular
membership functions. A normal triangular membership function can be defined
as a triple (u,, u, u,) with the height of membership function equal to I where u, is
the lower bound; u, is the upper bound; and u, is the point with the membership
value equal to 1. After the three points have been determined. the membership
function can be constructed by interpolation. The formulae in the interpola(ion are
shown as the following:

where u is the utility value and p is the membership value.


TABLE
2: Respondents give opinions about the utility value and scaling factor of each cost driver by
repertory grids. These different opinions are transformed to the triangular membership function listed in
the table. Lower limit and upper limit for each cost driver means the utility value is equal to 0 and 1,
Downloaded by [FU Berlin] at 14:58 02 May 2015

respectively.

Rank Cost Driver Scaling Factors Level 1 Utility Lower


1 %of totalcost (0.15.0.2.0.36) 1 1 (0.55,0.67,0.8) 15%
2 Raw material costs (0.13,0.2,0.3)
3 High variability in raw material costs (0.14,0.18,0.25)
4 Process control of raw material (0.11.O. 15.0.23) C#> 1.3 c,>o
5 Ingot size (0.1.0.15.0.2) Standard Nonstandard
6 Ingot weight (0.09,0.13,0.18) 600 1b 2500 lb
7 Raw material lead time (0.08.0.12.0.15) 2 months 12 months
8 Material I alloy system (0.05.0. l ,O. 15) Common New
9 Variation in material properties (0.06.0.09.0.13) 0% 50%
10 Number of defects (0.05.0.08.0.l2) 0 loo0
11 Yield
12 Cncking probability
THE ENGINEERING
ECONOMIST 1999 VOLUME44 NUMBER4 319

FIGURES 2a-2c illustrate utility functions of major cost drivers. The functions
can map the crisp value to the utility value of each cost driver. The utility function
for each attribute can be decided by preference-query questions. This procedure
makes the outcome of each attribute convert into a standard index ranging from 0
to 1.

Utility Function of % Total Cost


Downloaded by [FU Berlin] at 14:58 02 May 2015

- 0.4

,p
40
*,
\'\o * ,, ,, #,,
,%9
*\o ,$\0 0\$, , &\0

4.4.
8\0
$s
G\O

,9
O\O

% of Hlgh Cost

FIGURE 2a. Utility Function of Percent of Total Costs.

Utility Function of Raw Material Costs

; ; ;
E 0.4
3 0.2
0
Q
\ s 4- \6 %
\ q? p q? 50 5% 4' 96 9% 9
Raw Material Cost

FIGURE
2b. Utility Function of Raw Material Costs.
L

Utility Function of Variability


in Raw Material Costs

1 =-
Q

~ ~ - ~ a t - m m ~ ~ o ~ m - t m a t - m m q
999440"'T'""o
Downloaded by [FU Berlin] at 14:58 02 May 2015

0 0 0 0 0 0 0 0 0 0 0 - 0 0 0 .
Variability

FIGURE2c. Utility Function of Variability in Raw Material Costs.

FIGURES3a-3c show the membership function for all levels of cost drivers
corresponding to FIGURES2a-2c. The repertory grid categorizes seven levels for
each cost driver. Each level can be expressed by a triangular membership function.
The input of each membership function is the utility value.

w
-
E
C3
1.2
1
Membership Functionaf %Total Gost
-- Lwd -3
L d -2
.-
C
a 0.8 a. - Lwd -1
g 0.6 -X- Lwd 0
-*-
-
&I
f 0.4 Lwd 1
2 0.2
0 Lwd 3
0 0.2 0.4 0.6 0.8 1
Utility Value

FIGURE 3a. Membership Function of Percent Total Cost.


THE ENGINEERING
ECONOMIST
1999 VOLUME
44 NUMBER 4 32 1

6p 0.8
.-
.c
1.2
1
Menbership Fiarctionaf Raw Material C m b

--- L w ~-3
L w -2
~
Lwd -1
2 0.6
Q) -X- L d0

-
P
5 0.4 -X-Lwd I
0.2
0 Lwd 3
Downloaded by [FU Berlin] at 14:58 02 May 2015

0 0.2 0.4 0.6 0.8 1


Utility Value

hGum 3b. Membership Function of Raw Material Costs.

B
E
u
1.2
1
Membership Fumtion of kriability
in Raw Material Qsts
-- Levd -
Lwd -
.-
n 0.8
C
2 0.6
Q)
a
E 0.4.
2 0.2
0
0 0.2 0.4 0.6 0.8 1
Utility Value

FIGURE
3c. Membership Function of Variability in Raw Material Costs.
The output of the FMAU model is shown in FIGURE 4. Obviously, the function
is not in types of triangles due to the multiplication operation of the fuzzy
arithmetic. There are 20 historical data sets applied to tune the model and three
defuzzification methods, COA, COM and BOA, which will be considered. TABLE
3 lists the results that show COA is the best method in the ingot process. FIGURE
5 shows the CER curve under the consideration of COA, which is used to predict
the product cost. In this example, the cost of the new product in the ingot process
is $14,457.
Downloaded by [FU Berlin] at 14:58 02 May 2015

Output of FMA U Mode1


l.m-
8 am--
?!
0
.-Cn 0.600- -
E
: a4m--
E
a m --
0.0001
am am am am am 1.m
Cost Index

FIGURE4. The Membership Function of the Cost Index Can Be Evaluated


By Using EQUATIONS 5 and 6.

TABLE 3. Different defuzzification methods result in different square


errors. The optimal cost index is calculated by the defuzzification
method with the minimum square errors.
Results \ Methods COA COM BOA
Cost Index 0.61 0.621 0.583
Square Errors 47032 48562 50123
-
0
d

Q G

.z 4
;;i

.g
5
:
0
o e
5 %
S t;
'S 8

+:3 3
Downloaded by [FU Berlin] at 14:58 02 May 2015

ss 4!
2 5
8 s
4-8
<
-25c,
0

U
22
-
.e D
65
.%
U

5 ,=
0 Fi
cr m
0 T
i
s5
.-
ti S
= 9
c?
3
'it u
E r
w .o
$
vi:

2
2
324 THEENGINEERING
ECONOMIST
1999 VOLUME44 NUMBER4

In this study, an FMAU model for evaluating product cost in the manufacturing
industry is developed. FMAUTcost estimation can be used if the information about
an object is incomplete or uncertain. In addition, this method is more efficient than
other traditional cost models because it is not necessary to collect a great amount
of historical data and can handle attributes with uncertainty and incompleteness in
nature. In FMAUT, cost estimation is done with a systematic procedure with the
experts' experiences and opinions. Because of the fuzzy operations of opinions
from a number of experts, the subjectivity is reduced in assessing productlprocess
costs. This paper also presents an optimization method for tuning the cost model.
Downloaded by [FU Berlin] at 14:58 02 May 2015

The costing method is suitable for design and implementation of new products,
materials and processes as the cost information is usually incomplete or uncertain.

TABLE
4: Repertory grid used for estimating the cost of the ingot process. The
ingot cost is affected by the ingot yield.

in mw material costs
Poor process control
of rnw materials of raw materials

Standard size of ingot

6 Large ingot weight Smnll ingot weight


Long raw material

New material / Common / standard


material / alloy system
Small variation in
material properties

I1 254byieId
THE ENGINEERING
ECONOMIST1999 VOLUME
44 NUMBER4

TABLE5. Repertory grid used for estimating the cost of the extrusion process. T h e cost
of extrusion process is the total cost of the extrusion process allocated to the number of
mults produced by the extrusion process.

2 Low cost I Ib
3 Small variability in cost Luge variability in cost
4 Small size for access Luge size for access
Downloaded by [FU Berlin] at 14:58 02 May 2015

Capable analytical
modeling
Wide processing N m w processing
windows
7 Low temperature
8 Good lubricity
Good control of heating, Poor con~rolof heating.
temperature, ntes temperature. rates

1I Good process control


12 Availnble tooling
13 Long die life

15 Low nose 1 mil loss High nose I tail loss


16 Low cracking probability High cracking probability
Good straightness
of extrusion
8multsperstandard 6 mults per standard
extrusion
THE ENGINEERING
ECONOMIST1999 VOLUME44 NUMBER
4

TABLE
6: Repertory grid used for estimating the cost of the forging process. T h e cost of
forging process equal the cost of forging the past plus the cost associated with t h e prior
costs of the parts scrapped during forging.

3 Low variation in cost High variation in cost


4 High commonality of parts Low commonality of parts
Downloaded by [FU Berlin] at 14:58 02 May 2015

Availability I cost
of furnaces
6 Low p u t complexity High part complexity
Common I standud New material / alloy
material I alloy system
Capable analytical Limited analytical
modeling capability modeling capability
9 Easy forgeability Difficult forgeability
10 Easy lubricity
II Large processing window Small processing window
12 Low temperature
13 Short forging cycle Long forging cycle
Acceptable press Unacceptable press
capability I load limits capability I load limits
15 Air furnace atmosphere Vacuum furnace atmosphere
16 Good process control Limited process control
17 Available tooling
18 Inexpensive tooling material Expensive tooling material
19 Long die life
20 Unlikely reforge 1
~nterrnedinteheat treat
Easy micmstructure Difficult microstructure
control
Small variation in
ECONOMIST1999 VOLUME44 NUMBER
THEENGINEERING 4 327

Table 7: Repertory grid used for estimating thecost of the heat treatment. T h e cost o f the
heat treatment is the cost of the process itself plus the burden on that part of the cost of
parts scrapped during heat treatment.

High analytical modeling Low nnalytical modeling


capability
4 Low cooling ntes
5 Shon lmsfer time Long tnnsfer time
Downloaded by [FU Berlin] at 14:58 02 May 2015

6 Good pmcess control Poor pmcess control


7 Less thermal coupling More thermal coupling
8 l e s s specialized equipment More specialized equipment
9 Repeatable distortion Variable distortion
10 Small residual stresses Large residual stresses
11 Small variability

TABLE
8: Repertory grid used for estimating the cost of the machining process. T h e cost
of the machining i s the cost of machining the part plus the cost associated with parts
scrapped during machining.

3 Simple part design


4 Easy material machinability Hard material machinability
5 Adequate machining data Minimal machining data
6 Loose dimensional tolerance Tight dimensional tolerance
, High analytical modeling
capability
Low analytical modeling

8 Smdl machining distortions Large machining distortions


9 Standard fixturing
10 Fast machining speeds Slow machining speeds
I1 Adequate machining data Minimal machining data
Adequate process
capability
on-std. post-machining
328 THE ENGINEERING
ECONOMIST1999 VOLUME 44 NUMBER
4

TABLE
9: Repertory grid used for estimating the cost of the miscellaneous group.

2 No lead time requirements Lead time requirements


3 Low overhead costs High overhead costs
4 Low capital
5 New processes Standard processes
Downloaded by [FU Berlin] at 14:58 02 May 2015

Research presented in this paper has been partially supported by the National Science
Foundation and Department of Air Force (Grant No. DMI-9525991). The authors express
their appreciation to both agencies. They also thank United TechnologiesIPratt & Whitney
Company for providing technical support for this research.

[ I ] BERLINER, C., and BRIMSON.J.A., (1989), Cost Management for Today's Advanced
Manufacturing. The CAM-I Conceptual Design, Harvard Business School Press.
[2] BILUIT,A., (1992), "From Fuzzy Set Theory to Non-additive Probabilities: How Have
Economists Reacted?'Fuzzy Sets and System, Vol. 49, pp. 75-90.
[3] CANADA, J.R., and SULUVM, W.G., (I 989). Economic and Multiattribute Evaluation
of Advanced Manufacturing Systems, Booknews, Inc.
[4] COOPER, R., (1987), "Does Yourcompany need A New Cost System?"Journal of Cost
Management, pp. 45-49, Spring.
[ 5 ] COOPER,R., and KAPLAN,R.S., (l988), "Measure Costs Right; Make the Right
Decisions," Harvard Business Review. pp. 96-103, September/October.
[6] DEAN,E.B., (1995). Design for Cost and Qualiry: The Robus1 Design Approach,
<http:Nmijuno.larc.nas ...p ap/robdes/robdes.htd>.
[7] DEAN,E.B., (1996), Design for Compelitive Advantage, chttp:Nmijuno.larc.nasa.gov/
ebd/mypubs.html>.
[8] EDWARDS, W., G ~ E N T A M., G , and SNAPPER, K., (1975), "Effective Evaluation: A
Decision Theoretic Approach," in C.A. Bennett and A. Lumsdaine (eds.), Evaluation
and Experiment: Some Critical Issues in Assessing Social Program., New York,
Academic.
[9] FISHBURN, P.C.. (1970), Utility Theoryfor Decision Making, John Wiley & Sons, Inc.,
New York.
[lo] FRENCH,M.J., and WIDDEN, M.B., ( 1993). "Unction-costing: A Promising Aid to Early
Cost Estimation," DesignforManufacturability, ASME. Design Engineering Division
(Publication), DE, Vol. 52, pp. 85-90.
[ I I] GUNTER,B., (1987), "A Perspective on the Taguchi Methods," Quality Progress,
pp. 44-52, June.
[12] HELLENDOORN, H., and THOMAS, C., (1993), "Defuzzification in Fuzzy Controllers,"
Journal of Intelligent & F u ~ Systems,
y Vol. 1, No. 2, pp. 109-123.
[13] INNES,J., and MITCHELL, F., (1995), "A Survey of Activity-based Costingin the U.K.'s
Largest Companies," Management Accounting Research., No. 6, pp. 137-153.
[I41 JOHNSON,H.T., (1992), ''It's Time To Stop Overselling Activity-based Costing,"
Management Accounting, pp. 26-35, September.
[15] JOHNSON, H.T., and KAPLAN,R.S.. (1987), 'The Importance of Long-term Product
Downloaded by [FU Berlin] at 14:58 02 May 2015

Costs," McKinsey Quarterly, Utumn, pp. 3 6 4 8 .


[I61 KACKAR,R., (1985), "Off-line Quality Control, Parameter Design, and the Taguchi
Method," Journal of Quality Technology, Vol. 17, No. 4, pp. 176-188.
[I71 KEENEY,R.L., MCDANIELS,T.L., and SWOVEW, C., (1995), "Evaluating
Improvement in Electric Utility Reliability at British Columbia Hydro." Operations
Research, Vol. 43, No. 6. November-December.
[18] KEENEY,R. L., and RAIFFA,H., (1976), Decision with Multiple Objectives: Preferences
and Value Tradeoffs. John Wiley & Sons. Inc., New York.
[ I 91 L o c ~ s c r oA..
, and THURSTON, D.L.. (1 993), "Multi-attribute Design Optimization with
Quality Function Deployment," Proceedings of the 3rd Industrial Engineering
Research Conlerence, Institute of Industrial Engineers, Norcross, Ga.
(201 MELIN,J.B., JR., (1994), "Parametric Estimation," Cost Engineering, Vol. 36, No. I ,
pp. 19-24, January.
[21] METR~,M., (1995). "Combining Semantic Networks with Multi-attribute Utility
Models: An Evaluative Data Base Indexing Method," Expert System With
Applications, Vol. 9, No. 3.
[22] MILLER,J.G., and VOUMAN,T.E., (1985), 'The Hidden Factory," Harvard Business
Review, pp. 142-150, September/October,.
[23] PHADKE,S.M., (1989), Qualify Engineering Using Robust Design, Prentice Hall,
Englewood Cliffs, N.J.
1241 REED,T.B., WHITE,C.C.. BOLTON,M.P., and HILLER,W.D., (1994), "Application of
Multiattribute Utility Theory to Public Transit System Design." Transportation
Research Record, No. 1451, December.
[25] SON. Y.K.. (1991). "Cost Estimation Model for Advanced Manufacturing Systems,"
International Journal of Production Research. Vol. 29, n. 3, pp. 4 4 1 4 5 2 , March.
[26] TAGUCHI, G.,ELSAYED, E., and HSIANG,T.,(1989), Quality Engineering in Production
Systems, McGraw Hill, New York.
[27] VON NEUMANN, J., and MORGENSTERN, 0.. (1944), Theory of Games and Economic
Behavior, Princeton University Press.
330 THE ENGINEERING ' 1999 VOLUME 44
ECONOMIST 4
NUMBER

[28] YAN, X.T., SHARPE,J.E., and CHAPLIN,R.V., (1994), "Parametric Approach to


Mechatronic Product Design," Methodologies, Techniques, and Tools for Design
Development, ASME, Petroleum Division (Publication), PD, Vol. 64, No. 5, pp. 1-8.
(291 YOSHIKAWA, T.,INNES, J., and MITCHELL, F., (1989), "Cost Management Through
Functional Analysis," Journal of Cost Management, pp. 14-19.

PO-KANG(DEAN) TINGis a Graduate Student in the Industrial Engineering Department at


Downloaded by [FU Berlin] at 14:58 02 May 2015

Florida A&M University-Florida State University College of Engineering. His research


interests include cost estimation for product and process design, applications of fuzzy
mathematics and complexity theory.

CHUN(CHUCK) ZHANGis an Associate Professor of Industrial Engineering at Florida A&M


University-Florida State University CoIlege of Engineering where he teaches courses in
computer-integrated manufacturing and concurrent engineering. He received his Ph.D, in
industrial engineering from the University of Iowa in 1993. Dr. Zhang's current research
interests include intelligent processing of composite materials, lolerancing and metrology
for precision machining, and integration of manufacturing execution systems. He has
published more than 70 technical articles. Dr. Zhang is a recipient of the 1999 SME
Outstanding Young Manufacturing Engineer Award.

BENWANGis Professor and Chairman of the Industrial Engineering Department at Florida


A&M University-FloridaStateUniversity CollegeofEngineering. Dr. Wang's chiefresearch
interests are integrated product and process design, predictive maintenance, and
environmentally conscious manufacturing. He also specializes in policy and process
development for environmentally conscious manufacturing. He holds four U.S.patents, and
is the author or coauthor of 90 refereed journal papers and three books. In 1995, he was
named US DOE Dr. Samuel P. Massie Chair of Excellence in Environmental Engineering.
He is a recipient of the 1998 IRVASEE Engineering Faculty Fellowship Award, the 1998
United Nations Engineer Award, and the 1991 University of lowa Old Gold lowa
Fellowship.

BARBARADUBROSKY is a Senior Materials Engineer in Pratt & Whitney, West Palm Beach,
Florida. Her major responsibilities include developing and fostering the use of processing
science tools, including affordability methodologies and lean principles, as well as in-
process control, and advanced sensor and measurement technologies.
ABHUrr DESHMUKH is an Assistant Professor of Mechanical and Industrial Engineering at
the University of Massachusetts. Dr. Deshmukh received his Ph.D. in industrial engineering
from Purdue University in 1993. His major research interests include production planning
and manufacturing systems analysis.
Downloaded by [FU Berlin] at 14:58 02 May 2015

Você também pode gostar