You are on page 1of 7

Evaluating automated

manufacturing technologies:
Part I I - a methodology for
evaluation
SABAH U RANDHAWA and TOM M WEST

Justification methodology
Abstract: In Part I of this paper we provided a review of the
techniques used in the evaluation of automated manufacturing An overview of the proposed methodology is shown in
technologies. In this paper we provide a framework to integrate Figure I. The primary input classes are: a description of
the economic, technical, strategic, and qualitative factors in the the system to be analysed; operational requirements
investment decision. The framework is based on combining a and environmental restrictions; and management pre-
simulation model with a multi-attribute decision model. The ference structure reflected in the selection of attributes,
simulation model is used to estimate the operational and assessment of qualitative attributes, and in assessing
economic attributes for the investment problem. The multi-
the relative importance of attributes in the decision
attribute decision model is the linear additive scoring model
environment.
because of its ease of use and its ability to structure
multi-dimensional problems. The basic components of the system shown in Figure
1 include:
Keywords: automated manufacturing, implementation, justifi- (1) Attribute selection procedure: the decision environ-
cation, simulation, multiattribute evaluation, scoring models
ment is best described by a list of attributes or
characteristics. Therefore, the initial task in the
justification process is to develop a list of relevant
n Part I of this paper we developed a classification attributes for the situation being analysed. Some of
I scheme for CIM evaluation techniques and briefly
reviewed some of these techniques. It was pointed
out that evaluating investments in automated technolo-
these attributes will be quantitative (both economic
and non-economic) in nature; others will be
qualitative.
gies is a complex decision due to the large number of
attributes and their complex interactions. The task is (2) Simulation module: T h e simulation module is used
further complicated by the necessity of including the to obtain estimates for quantitative attributes, t h e
non-financial and qualitative aspects of manufacturing simulation module consists of three components:
performance. Due to these factors, no single modelling
(a) a resource library for describing the objects of
technique is adequate for analysing all aspects of the
the physical system (equipment, piece parts,
manufacturing evaluation. This paper describes a
etc.) and their interactions:
methodology based on integrating some of the tech-
(b) a library for decision strategies that describe the
niques discussed in Part I. The two techniques
policies for equipment layout, scheduling,
primarily utilized are simulation and scoring models. inventory control, etc.; and
The next section provides a brief overview of the
(c) a decision support library that contains models
justification methodology. This is followed by a
for estimating modelling factors such as fore-
detailed description of each component.
casts and economic parameters, and assisting in
the analysis of simulation output.
Department of Industrial and ManufacturingEngineering, Oregon An integral part of the simulation analysis is an
State University,Corvallis, Oregon97331, USA integrated capability for sensitivity and risk analysis.
0951-5240/92/040276-07 (~ 1992 Butterworth-Heinemann Ltd

276 Computer-Integrated Manufacturing Systems


S U RANDHAWA AND T M WEST

Quantitative attributes Qualitative attribute estimates


- economic ( NPV, IRR)
- non -economic ( utilization, lead time, throughput )
~ Simulation module J
(estimating L ,
quantitative attrioutes j

Qualitative
attributes

Strategic evaluation module Qualitative Multiattribute versus simulation module


attribute levels
.~ Evaluate J JCompute aggregate J
qualitative J ---~,cores for I
attributes J Jalternatives J
- Alternative
i evaluation

_•
System description scores
Operational requirements Develop J - Ranked
Management preferences
Quantitative +
qualitative
attributes
attribute
weights
Attribute
weights

sensitivity
analysis
I ~-- ordered
alternatives

Figure 1. Methodology for evaluation

(3)Strategic evaluation module: this module uses possible significant impact and familiar with the
management's judgement and evaluation to operational and other technical specialties should be
(a) develop estimates for qualitative attributes, and asked to contribute ideas. The experience of these
(b) develop attribute weights, or relative import- experts should augment the literature review process,
ance of the attributes in the decision environ- and help to minimize any omissions and duplication
ment. among attributes. Additionally, a simple model-
building approach, in which by building a model of the
(4) Multiattribute evaluation module: this module per- manufacturing problem and identifying relevant input
forms two functions: and output variables, suitable attributes may be
(a) computing the aggregate score for each manu- identified, can be used in the attribute identification
facturing alternative consisting of total weigh- process. A list of typical attributes for a CIM
ted attribute values for an alternative, and application is shown in Figure 2.
(b) Performing sensitivity analyses on attribute There is no one best set of attributes. The attributes
weights and attribute values. chosen to serve as a basis of comparison are dependent
on the characteristics of the specific manufacturing
The output from the system is a set of measures for
each manufacturing alternative. The measures reflect environment. However, it is generally desirable to keep
the attribute set as small and concise as possible.
the performance of an alternative with respect to the
Additionally, it is necessary to make every attribute
defined attributes, and incorporate tradeoffs among
mutually exclusive of every other attribute. There are
attribute weights and attribute values.
primarily two reasons for keeping the attribute set to a
minimum. First, most of the available techniques for
Attribute selection procedure combining the impact of attributes assume independ-
A manufacturing alternative is defined by a particular ence or near independence among attributes. This
collection of attribute values. The development of a list implies that the attributes should be mutually exclusive
of relevant attributes may be as simple or as extensive to avoid double counting of impacts. In practice, it is
as the analyst wishes. Often this activity is physically often impossible to obtain an independent set of
limited by the time and resources available. It is always attributes. However, independence is easier to justify
desirable to thoroughly consider all the attributes in the and approximate with a smaller attribute set than with a
decision environment. However, the time and effort larger set. Correlation among attributes may be
required for such complex analyses are usually minimized by using statistical techniques such as factor
prohibitive, and it is often possible to reduce the analysis. Second, a large number of attributes require
dimensionality of the problem without a significant loss considerable effort in obtaining attribute values for
of information and accuracy. A multistep iterative each alternative.
procedure is usually used to identify relevant attributes. Additional attribute sets for analysing manufacturing
The procedure starts by identifying an initial set of environments are given elsewhere 1-5. A procedure for
attributes through literature search and questionnaires, identifying attributes and minimiz!ng duplications
and developing operational definitions for the attri- through factor analysis is given in Randhawa and
butes identified. Persons knowledgeable in the areas of Bedworth 2.

Vol 5 No 4 November 1992 277


Evaluating automated manufacturing technologies: Part H
nature of the operational and design variables. The
Economic (quantifiable)
System acquisition and installation costs approach described below considers the economic
Startup costs analysis and the operating environment as integrated
Variable production costs (raw material, machine components, and evaluates the operational and the
operation, direct and indirect labour and utility costs) economic viability of the system at the same time.
Work-in-process inventory (and shortage) costs A number of operational factors need to be
Quality costs (inspection, manufacturing accuracy and
precision, scrap and rework, replacement) considered in evaluating a manufacturing system.
Maintainability costs (system diagnostics, repair and These include different types of products with differing
upkeep) production sequences; scheduling priorities; equipment
Operational (quantifiable) outlays; effect of machine breakdown and failures on
Equipment utilization output; production schedules; and interaction among
Throughput rate system components based on schedules, breakdowns,
Production lead times intermediate buffers, spatial interference, and other
Product accuracy and precision such factors. These complex interactions and the
Shop floor information requirements
Space requirements and utilization resulting economic and non-economic consequences
are very difficult to represent analytically. Simulation
Qualitative (non-quantifiable)
Flexibility (design change accommodation, changes in provides a mechanism by which these interactions can
product mix or market responsiveness, routing and be represented and analysed.
scheduling flexibility, capacity growth) The prerequisite for simulation analysis is an
Compatibility (with existing or new machines, tools, accurate representation of the manufacturing process.
computer control and communication) Simulation involves developing a model of the system
Safety of operationsto avoid injuries and accidents and then conducting computer-based experiments with
Training availability and quality
Worker morale (or acceptability and enthusiasm for use the model to describe, explain, and predict the
of the system) behaviour of the system. The system model generally
Technical and managementsupport includes: logical components, as for example, defining
the material flow in the system; heuristic components,
Figure 2. Attributes for CIM application as for example, defining scheduling policies on specific
machines to determine the order in which jobs are to be
processed on the machine; analytical components, as
for example, a cost estimation model to determine
Simulation module revenues and operating costs; an economic model to
evaluate net present worth or internal rate of return,
The objective of the simulation module is to estimate growth or forecasting model for generating future
operational and economic attributes for the investment demand, and a quality evaluation model to analyse
problem. Manufacturing systems are complex combin- product quality, scrap loss, etc. The simulation model
ations of computer control, communications, produc- represents an integration of these three components.
tion processes, and human workers. As such, the Since alternative manufacturing processes may differ
operational (or technical) and economic components of significantly, simulation models designed for a specific
the system are integrated and interdependent upon technology may not be readily adaptable for the
each other. Operational factors such as equipment analysis of an alternative process. Developing a 'new'
utilization, inventories, work in-process queuing, and model for each alternative to be evaluated is expensive,
system throughput rates affect the cash flows (revenues in terms of both time and cost. This is also one of the
and costs) are used to estimate economic attributes (net main reasons for relatively low acceptability of
present value or internal rate of return). The oper- simulation in evaluating manufacturing technologies.
ational factors, in turn, are affected by factors such as The key is to develop a simulation modelling frame-
equipment characteristics (e.g. speed and capacity), work that can be extended or modified to support the
equipment layout, and scheduling policies. For exam- analysis of multiple alternatives. It is not feasible to
ple, component interaction in one type of layout and develop a generic simulation model; instead, we
scheduling priorities may result in low work in-process propose that the conceptual framework shown in
inventories, but high equipment utilization; however, Figure 3 be used as a basis of developing simulation
changing either the layout or the scheduling priorities models in analysing manufacturing alternatives.
may increase inventories and increase utilization. The modelling framework contains three structures
Most evaluation approaches currently use single for representing and analysing a manufacturing system.
parmeter economic analysis to select a manufacturing Each of these structures is represented as a software
alternative. Subsequently, the operational feasibility of library in the simulation framework. The resource
the system is established by analysing combinations of library represents the physical objects (processing
system components. These economic analysis techni- stations, material handling equipment, and products in
ques are usually insensitive to operational considera- various stages of manufacturing) in the system. The
tions; they ignore the fact that cash flows are random layout for a manufacturing system is a combination of
variables, the randomness resulting from the stochastic objects from this library in a specific order.

278 Computer-Integrated Manufacturing Systems


S U RANDHAWAAND T M WEST

The other two libraries represent conceptual con- analysis. In the majority of evaluation techniques which
structs. The decision strategies library contains alterna- consider parameter variability, the system is first
tives for equipment layout, sequencing, scheduling, analysed using deterministic values. Sensitivity analyses
and similar physical attributes. The decision support are then performed on the base model through equally
(or model) library contains a set of analytical models to likely variations in attribute values to identify the more
be used in forecasting, economic evaluation, inventory critical attributes in the system. This is followed by risk
management, and analysing simulation output, and analysis or dermining the probability distributions
constructs required for executing simulation such as associated with outcomes of critical attributes. These
random number generators and event schedulers (for resulting distributions describe the variability associ-
generating and controlling the execution logic). ated with these attributes 6. Irrespective of the evalua-
The framework of Figure 3 shows a modular tion technique, advanced sensitivity and risk analyses
structure, independent of the implementation tech- generally require some form of simulation analysis.
nique. Thus, manufacturing alternatives to be analysed In simulation analysis, the variability associated with
can be constructed by selecting components from the input parameters (costs, product demand, economic
resource library, and interfacing these components indicators such as interest and inflation rates, and
using conceptual constructs from the decision strategies machine performance parameters including capacities
and decision support libraries. and rates) are defined in terms of probability distri-
An advantage of using simulation is that sensitivity butions; the variance or the standard deviation of the
and risk analyses are integral components of the output measures then indicate the degree of uncer-
analysis, rather than representing 'post-optimality' tainty of risk associated with predicting the final result.
Determining distributions for input parameters
depends on available information. If adequate histori-
cal data are available, then a standard distribution may
be fit to this data and verified using 'goodness-of-fit'
Resource Library tests. If no standard distribution fits the observed data,
Workstations then the factor may be modelled using an empirical
Processing Stations
Machining distribution computed from existing or synthesized
Drilling data. If no system data is available, as is not uncommon
Turning in evaluating new systems, a triangular distribution
with a minimum, maximum, and most likely value may
Transfer Stations be used 7. The triangular distribution recognizes the
variability of data, and is easy to use as the parameters
Storage Queues (Buffers) may be estimated based on input from system experts.
Material Handling In general, if an objective or subjective distribution
Robots could be fit, this is preferable. Distributions allow
Conveyors 'extreme events' to be generated which fall outside the
AGVs range of observed data. Extreme events while rare
often may have a significant impact on the design of the
Storage/Retrieval system.
AS/RS In addition to sensitivity and risk analysis, certain
important characteristics such as flexibility associated
Entities with advanced manufacturing technologies can be
Raw Material evaluated using simulation. Different types of flexi-
bility may be associated with a manufacturing system8.
Finished Products These include process flexibility, product flexibility,
volume flexibility, and expansion flexibility. Although
Decision Strategies Library
all forms of flexibility are hard to quantify, the effect on
Equipment Layout economic and non-economic attributes can be incor-
Sequencing (or Routing) Policies porated in the simulation analysis by designing and
Scheduling Policies analysing different scenarios.
Inventory Control Strategies

Decision Support Library Strategic evaluation module


Forecasting Module
Economic Evaluator Measuring qualitative attributes
Statistics Analyser Effective representation of qualitative attributes re-
Random Number Generator quires the approximate quantification of qualitative
Event Scheduler attributable values. The most common quantification
method is to develop a cardinal scaling procedure
Figure 3. Framework for simulation modeling where the various applicable qualitative levels are

Vol 5 No 4 November 1992 279


Evaluating automated manufacturing technologies: Part H
associated with a numeric scale. The choice of a 10- or Table 1. Attribute weights through palrwise compari-
lO0-point scale is usually logical, since most people son (importance scale 0 to 2; 0 = less important, 1 =
seemingly have an intuitive feel for such ranges. The same importance, 2 = more important)
most obviously identifiable values of an attribute are
the end points on the scale. For example, on a 0-10 Cost Compatibility Flexibility Sum Weight
scale, the 0 point represents the minimum attribute
value that is physically or practically attainable, while Cost 0 2 1 3 0.500
the maximum attribute value that is physically or Compatibility 0 0 1 1 0.167
Flexibility 1 1 0 2 0.333
practically attainable corresponds to the 10 point of the
scale.
To illustrate the scoring of qualitative attributes,
assume that manufacturing alternatives are to be complement of itself). This simplifies the assessment
evaluated on three attributes - quantitative attributes, procedure as only half the matrix values need to be
cost, and the qualitative attributes, compatibility and filled.
flexibility. Then, for compatibility, the 0 on the discrete This procedure may be further expanded through the
scale may correspond to 0% compatibility, and 10 may use of opinions from several 'judges' or managers.
correspond to 100% compatibility, where compatibility Concordance or logic tests may be performed on the
refers to integration of hardware and software compo- results and for large groups of input. Analysis of
nents of the system. The intermediate values on the Variance (ANOVA) may be performed to obtain
0-10 scale then represent different levels of compatibil- estimates of significant difference between attribute
ity, meaningful to the decision maker. weight estimates is.
The values on the discrete scale employ certain
mathematical relationships9. More importantly, it is Multiattribute evaluation m o d u l e
assumed that a scale value of 8.0 is twice as favourable
as 4.0. Additionally, the use of values across attributes Combining attributes
implies that the quantifiable difference between 'high' The next step in the analysis consists of computing an
and 'low' impacts for two different attributes is aggregate weighted value for each manufacturing
identical. alternative. This value is most easily obtained using the
Application of scaling techniques permit the quantifi- additive linear model that reduces the n-attribute
cation of qualitative variables but the numerical values problem to a single dimension using the following
for quantitative factors (e.g. cost) obtained using expression:
simulation analysis are usually not on a comparable
n
scale. The simplest procedure is to use a 0-10 scale with
the lower and upper values of the scale representing wj = w, j (1)
extreme but practically attainable values. Thus, for the i=1
attribute cost the 0 may represent a very high cost (an where Wj is the aggregate score for alternative j, wi is
FMS system, integrated and automated) while a 10 may the weight for attribute i, 3~jis the score of attribute i on
represent a very low cost value (manually controlled alternative j, and n is the total number of attributes in
job shop system). the model.
To use the above expression, the values for
quantifiable attributes (economic and noneconomic)
Attribute weights must be reduced to a comparable scale as that for
In any decision environment, not all attributes are qualitative attributes (see Table 2). The procedure is to
likely to be considered equally likely. The function of
weights is to express the importance of each attribute
relative to other attributes. Several techniques are Table 2. Computing aggregate scores of alternatives on
available for developing weights based upon judge- attributes based on 0 to 10 scale; 0 = unsatisfactory
mental or subjective opinions 1°'11. These include performance, 10 = superb performance)
ranking, rating, partial paired comparison 12, paired
comparison and Churchman-Ackoff-Arnoff
Alternative
method 13.
Attribute
The procedure proposed here for determining
Attribute weight A B C
attribute weights uses pairwise comparison of attri-
butes 14. Using the three attributes discussed in the
previous section, the comparison values in Table 1 Cost 0.500 6 5 9
show the importance of row attributes relative to the Compatibility 0.167 8 6 9
column attributes. Thus, cost is twice as important as Flexibility 0.333 9 7 3
compatibility. Also, the values below the principal Aggregate score on 7.33 5.83 7.00
diagonal are complement of the values above the production attributes
diagonal (0 is the complement of 2 and 1 is the

280 Computer-Integrated Manufacturing Systems


S U RANDHAWAANDT M WEST
use the same scale and scaling relationships for both to attribute values requires a complete solution of the
types of attributes, as discussed in the section on decision problem for different values of each attribute
'measuring qualitative attributes'. value. With a large number of attributes, this may
The additive weighting model can be a very powerful involve quite a few iterations, and the results may not
tool in evaluating multi-attribute manufacturing be easy to interpret. In actual practice, values
systems. However, the validity of the results obtained is associated with attributes with high importance weights
dependent on the development of realistically indepen- are the most likely candidates for sensitivity analysis.
dent attributes, the determination of comparable Since the weights of the associated attributes are high,
attribute values, and the assignment of reasonable even slight changes in attribute values may result in a
importance weightings. change in the ranking of alternatives. Also, attribute
values which involve a high degree of uncertainty and
subjectivity in estimation should be varied in the
Sensitivity analysis sensitivity analysis to investigate the effect on the
The two most important sensitivity analyses to look at variation on the ranking of alternatives.
are sensitivity to changes in attribute weights and As an alternative to using simulation to evaluate the
single-attribute values. Of these two, perhaps the more effect of variation of attribute values on alternative's
important is sensitivity to attribute weights. This is weighted score, analytic procedures can be used to
because attribute weights are the essence of value develop variability associated with the alternative
judgements and because weights, being purely subjec- weighted value. The complete procedure is described
tive numbers about which decision makers disagree, by West 9. To illustrate, consider the attribute cost
are more likely to be in dispute than other parameters. scaled on 0-10 scale to 3.0, 3.5, and 5.0 representing
A sensitivity analysis involving attribute weights three levels (low, most likely, and high), with relative
consists of investigating the sensitivity of the rankings probabilities of 0.2, 0.7, and 0.10, respectively, for
of the alternatives to small changes in attribute weights. three different scenarios (e.g. pessimistic, most likely,
If the rankings remain unaffected as the weights are and optimistic). Then the expected value of cost is 3.55
changed by modest increments, this can be established [(3.0)(0.2) + (3.5)(0.7) + (5.0)(0.10)], and the
as evidence that small errors in the estimation of expected variance is 0.272 [((3.0)2(0.2) + (3.5)2(0.7) +
attribute weights are not important. Weights can be (5.0)2(0.1)} - (3.55)2]. Once estimates of this type are
changed two at a time or three at a time, if desired. If determined for each attribute, the mean aggregate
one or more of the weights prove to be very sensitive, score for an alternative is determined using expression
the analysis will indicate the accuracy of the estimating (1) and the variance (vj) associated with this score is
weights and the stability in the ranking of alternatives. determined by:
The other sensitivity to look at is sensitivity to
r/
single-attribute values. These values may represent
point estimates or probability distributions or may be V] = Z I;i W2 (2)
subjective assessments reflecting the judgement of i=1
experts. Hence, in some cases these values may not be
where vi is the variance for attribute i and wi is the
known with precision, and in others may represent
weight of attribute i.
values of intense controversy. A full sensitivity analysis
The results give the decision maker an additional
to attribute values requires a complete solution of the
insight as to the variability of the final values. Various
decision problem for different values of each attribute
statistical methods can now be used for further analysis
value. With a large number of attributes, this may
of the choice being considered. Probably the most
involve quite a few iterations, and the results may not
important benefit obtained from this type of analysis is
be easy to interpret. In actual practice, values
tha ability to utilize estimates of the extreme values
associated with attributes with high importance weights
which may result under actual conditions.
are the most likely candidates for sensitivity analysis.
Since the weights of the associated attributes are high,
even slight changes in attribute values may result in a
Conclusions
change in the ranking of alternatives. Also, attribute
values which involve a high degree of uncertainty and The paper presented an approach for identifying an
subjectivity in estimation should be varied in the attribute set for a manufacturing environment, evaluat-
sensitivity analysis to investigate the effect of the ing quantitative attributes, estimating qualitative fac-
variation on the ranking of alternatives. tors, and developing aggregate alternative impact
The other sensitivity to look at is sensitivity to scores. The methodology is based on the integration of
single-attribute values. These values may represent three modelling techniques - simulation, manage-
point estimates or probability distributions or may be ment's judgement, and multi-attribute scoring model.
subjective assessments reflecting the judgement of Our focus was to develop a methodology that integrates
experts. Hence, in some cases these values may not be economic, operational, and qualitative aspects of
known with precision, and in others may represent manufacturing decision making, yet is easy to develop
values of intense controversy. A full sensitivity analysis and implement. Given the large number of attributes in

Vol 5 No 4 November 1992 281


Evaluating automated manufacturing technologies: Part H
a manufacturing environment and the complex interac- 6 Randhawa, S U and West, T M 'Uncertainty
tions among those attributes, simulation is a cost modeling in CIM investment analysis', CIM Rev.,
effective tool for analysing quantitative aspects of such Vol 6 No 1 (1989) pp 32-36
systems. A strategic component uses management's 7 Law, A M 'How to choose input probability
subjective jdugements in estimating qualitative attri- distributions for simulation', Simon Sez, Vol 8
bute levels, and in estimating attribute weights. Finally, (Summer 1988)
the individual attribute values and weights are synthe- 8 Browne, J, Dubois, D, Rathmill, K, Sethi, S P and
sized using a linear additive model. Stecke, K E 'Classification of flexible manufacturing
systems', The FMS Mag. (April 1980) pp 114-117
References 9 West, T M 'An integrated approach to the
1 Arbel, A and Seidman, A 'Performance evaluation evaluation of multiattributed alternatives in facility
of flexible manufacturing systems', IEEE Trans. siting studies', PhD Thesis, Department of Indust-
Syst. Man & Cybern., Vol 14 No 4 (1984) pp rial Engineering, Oregon State University (1977)
606-617 10 Bartlett, C V, Heel'man, E and Rettig, S 'Compari-
2 Randhawa, S U and Bedworth, D D 'Methodology son of six different scaling techniques', J. Social
identifies factors for comparing FMS with conven- Psychol. Vol 51 (1960) 343-348
tional manufacturing systems', Ind. Eng., Vol 17 11 Eckenrode R T 'Weighting multiple criteria',
No 6 (1985) pp 40-45 Manage. Sci. Vol 12 No. 3 (1965) pp 180-192
3 Varney, M W, Sullivan, W G and Cochran, J F 12 Buel, W D 'A simplification of Hay's method of
'Justification of flexible manufacturing systems with recording paired comparisons', J. Appl. Psychol.,
the analytic hierarchy process', Proc. Ann. Int. Ind. Vol 44 (1960) pp 347-349
Eng. Conf., Los Angeles, CA (May 19-23 1985) pp 13 Churchman, C W, Ackoff, R L and Aenoff, E L
181-190 Introduction to Operations Research, Wiley, New
4 Smith, M L, Ramesh, R, Dudek, R A and Blair, E L York (1957)
'Characteristics of US flexible manufacturing 14 West, T M and Randhawa, S U 'Multicriteria
systems - a survey', in K E Stecke and R Suri (ells), evaluation of manufacturing systems', in B Prasad
Proc. 2rid ORSA/TIMS Conf. on FMS: Operat. (ed.) CAD~CAM, Robotics and Factories of the
Res. Models & Applic., Elsevier, Amsterdam Future: Proc. 3rd Int. Conf. Springer-Verlag, Berlin
(1986) pp 477-486 (1989) pp 271-275
5 Falkner, C H and Benhajla, S 'Multiattribute 15 Dunn-Rankin, P 'Multiple comparisons in a simpli-
decision models in the justification of CIM systems', fied rank method of scaling', Educ. & Psychol.
Eng. Economist, Vol 35 No 2 (1990) pp 91-114 Measure., Vol 29 (1969) pp 315-329

282 Computer-Integrated Manufacturing Systems

You might also like