Res Eng Des (t991) 3:23-37 Research in EngineeringDesign

© 1991 Springer-VerlagNew York Inc.

A Survey of Research in the Application of Tolerance Analysis to

the Design of Mechanical Assemblies
K e n n e t h W. Chase* and Alan R. Parkinson
Mechanical Engineering Department, Brigham Young University, Provo, Utah, USA

Abstract. Tolerance analysis is receiving renewed em- loose tolerances which makes parts easier and less
phasis as industry recognizes that tolerance management expensive to produce. Therefore, tolerance specifi-
is a key element in their programs for improving quality, cations become a critical link between engineering
reducing overall costs, and retaining market share. The and manufacturing, a common meeting ground
specification of tolerances is being elevated from a menial where competing requirements may be resolved.
task to a legitimate engineering design function. New en-
In the last few years, numerous companies have
gineering models and sophisticated analysis tools are be-
ing developed to assist design engineers in specifying tol- established comprehensive programs in quality
erances on the basis of performance requirements and management. Noteworthy among them are the ef-
manufacturing considerations. This paper presents an forts of Motorola, IBM, and Xerox, who have initi-
overview of tolerance analysis applications to design with ated formal, corporate-wide programs for improved
emphasis on recent research that is advancing the state of tolerance specification, monitoring, and control.
the art. Major topics covered are (1) new models for toler- Their success in reducing waste while cutting cost
ance accumulation in mechanical assemblies, including and development time and reclaiming lost market
the Motorola Six Sigma model; (2) algorithms for allocat- share has received national praise [Placek 198%,
ing the specified assembly tolerance among the compo- Placek 1990, Kendrick 1991].
nents of an assembly; (3) the development of 2-D and 3-D Another indication of the growing interest in tol-
tolerance analysis models; (4) methods which account for erancing is the Mechanical Tolerancing Workshop
non-normal statistical distributions and nonlinear effects;
sponsored by NSF and ASME in 1988 which
and (5) several strategies for improving designs through
the application of modern analytical tools. brought together an international group of experts
in tolerancing to discuss the state of the art and
identify research opportunities [Paleck, 1989b,
ASME 1990]. This has been followed by special
1 Introduction
theme sessions at several ASME conferences, such
as the Design Technical Conference in Montreal
Interest in tolerance analysis is rapidly increasing in (1989), the Design Show in Chicago (1990), and the
industry. The quest for quality has focused atten- Computers in Engineering Conference in Boston
tion on the effects of variation on cost and perfor- (1990).
mance of manufactured products. Excess cost or
poor performance will eventually show up as a loss
of market share. Therefore, the specification of tol- A Critical Link Between Design and Manufacturing
erance limits on each dimension and feature of engi-
neering drawings is considered by many to be a vital 'an,72ng To,erancee
design function. Tolerance requirements have a far-
reaching influence that touches nearly every aspect Resultant Dimensions Production Cost
of the manufacturing enterprise as shown in Fig. 1. Fit and Function ComDetina Process Selection
Both engineering design and manufacturing per- Design Limits Reouirementa Machine Tools
sonnel are concerned about the effects of toler- Performance Tight Loose Operator Skills
ances. Engineers like tight tolerances to assure fit Sensitivity Tooling, Fixtures
Robust to Variation Inspection Precision
and function of their designs. Manufacturers prefer
Assernblability
Scrap and Rework
* Offprint requests: Mechanical Engineering Department,
Brigham Young University, Provo, UT 84602, USA Fig. 1. The effects of assigned tolerances are far-reaching.
24 Chase and Parkinson: Application of Tolerance Analysis

2 Models for Tolerance Accumulation the number of parts in the assembly sum increases,
the component tolerances must be greatly reduced
The basis for rational tolerance specification is to in order to meet the assembly limit, requiring higher
create an analytical model to predict the accumula- production costs. In the R S S model, the low proba-
tion of tolerances in a mechanical assembly. Critical bility of the worst case combination occurring is
clearances or fits or other resultant features of an taken into account statistically, assuming a Normal
assembly are generally controlled by the stackup or or Gaussian distribution for component variations.
sum of several component tolerances. A number of Tolerances are commonly assumed to correspond
analytical models exist with varying levels of so- to six standard deviations (60- or ---3o'). Component
phistication as shown in Fig. 2. tolerances may be increased significantly, since
Common models for predicting how component they add as the root sum squared (RSS).
tolerances sum are Worst Case (WC) and Root Sum Statistical distributions may be used to predict
Square (RSS) as shown in Eqs. 1 and 2 [Fortini the yield of an assembly, that is, the number or
1967]. fraction of assemblies which are likely to lie inside
the spec limits. RSS analysis generally predicts too
One-dimensional Two- or three-dimensional few rejects when compared to real assembly pro-
assemblies: assemblies: cesses. This is due to the fact that the Normal distri-
bution is only an approximation of the true distribu-
Worst Case tion, which may be flatter or skewed. The mean of
dU = E Ti - TASN
;l°fi )
dU = X \[0--~il Ti ~ TASM (1)
the distribution may also be shifted from the mid-
point of the tolerance range. To account for these
Root Sum Square uncertainties, a more general form of the RSS
model is frequently used:
dU = [Z T2] I/2 "< TASM dU = [E (0f~2
\~ii /
T2]vz
---<TASM
dU = CfZ [Y (0f] 2 Ti 2 1/2
(2) \0~i/ (Zi)] <- TASM (3)

where Xi are the nominal component dimensions, Ti where Z is the number of standard deviations de-
are the component tolerances, dU is the predicted sired for the specified assembly tolerance and Zi
assembly variation, TASM is the specified limit for describes the expected standard deviations for each
dU, and f(Xi) is the assembly function describing component tolerance. Cf is a correction factor
the resulting dimension of the assembly, such as the added to account for any nonideal conditions. Typi-
clearance or interference. The partial derivatives Of/ cal values for Cf range from 1.4 to 1.8 [Bender 1968,
0xi represent the sensitivity of the assembly toler- Levy 1974, Gladman 1980].
ance to variations in individual component dimen- Another conservative estimate of assembly toler-
sions. For a one-dimensional tolerance stack the ances assumes the component dimensions are uni-
sensitivity is -+ 1.0. formly distributed over the specified tolerance
The WC model assumes all the component di- range. In this case, the value of Zi is V~. If trun-
mensions occur at their worst limit simultaneously. cated normal distributions arise due to inspection of
It is used by designers to assure that all assemblies component parts, then choose V ~ < Z~ < 3, as
will meet the specified assembly limit. However, as described by Spotts [1983]. Spotts also suggested
calculating the WC and RSS assembly tolerance
and simply averaging the two as a safe estimate
[Spotts 1978].
I TOLERANCEANALYSIS",,]
I ] )
i wors, Case I I Sta,istica, I I samplod 1 2.1 Estimated Mean Shift Model
Root Sum Squares Monte Carlo
Simple RSS analysis assumes that the variation of
Mean Shift
each component dimension is symmetrically dis-
Six Sigma
tributed about the mean or nominal dimension.
Hasofer-Lind
However, in real processes, the mean is shifted due
Method of Moments
to setup errors or drifts due to time-varying parame-
Integration
ters such as tool wear. Ignoring mean shifts can be
Fig. 2. Mathematical models of tolerance accumulation. very detrimental, resulting in large errors in esti-
 Chase and Parkinson: Application of Tolerance Analysis 25

mates of the number of assemblies within spec lim- world competition in the electronics industry, Mo-
its [Spotts 1978, Evans 1975b]. torola has mandated +_6o- quality for all processes
Further modifications to the RSS model have (.002 defects per million). However, they also rec-
been proposed to take into account mean shifts or ognize that shifts and drifts in the mean of the pro-
biased distributions. Mansoor [1963] proposed that cesses are expected, so they have introduced a
the tolerance accumulation be represented as a WC modified process model which includes an allow-
sum plus a RSS sum. A similar model by Green- ance for accumulated mean shifts. The result
wood [1987], Greenwood and Chase [1987], and is a net quality level of +4.5o- (3.4 defects per
Chase and Greenwood [1988] introduced an esti- million).
mated mean shift factor mi ( a number between 0 and A versatile feature of the Six Sigma model of a
1.0) which quantifies the expected mean shift as a process is the ability to distinguish between short-
fraction of the specified tolerances, Eq. 4 illustrates term and long-term process capability. The process
the resulting expression. capability quantifies the spread of the process. It is
defined as 6.0 times the standard deviation of the
Worst Case Sum Statistical Sum process, 6o-~. Over the long term, however, the
mean of a process will drift due to tool wear or the
0f2 2 1/2
setup will vary from tot-to-lot, resulting in an appar-
ent increase in the process capability. The resulting
(4) modified standard deviation of a component pro-
cess may be estimated from Eq. 5
This is a versatile model. If all the mi are set to 1.0,
the result is a WC model. If all the m~ are set to 0, Ti
the result is a RSS model. By selecting mi between °-i = 3Cpi(1 - mi) where Cpi UL6O'Proc
- LL
(5)
0. and 1.0, the resulting variation will always lie
between the WC and RSS predictions. And any UL and L L are the upper and lower tolerance lim-
combination of mi may be chosen to account for the its, respectively, Cpi is the Process Capability Ratio,
degree of uncertainty in individual process charac- or the ratio of the specified tolerance range to the
terizations. process capability. Variable mi is the Mean Shift
Factor. When mi = 0, o-~ describes the short-term
variation in the process. When 0 < mi < 1.0, cri
2.2. Motorola Six Sigma Program
approximates the long-term variation. For the stan-
A new Tolerance accumulation model that has dard Six Sigma model, the target values of Cpi = 2
caught the attention of industry was developed by and mi = 0.25 result in tolerance limits of ---4.5o-i
the Motorola Corp. as a basic element of their over the long term. Note that other Cpi and mi val-
award-winning Six Sigma quality program which is ues may be selected to account for the degree of
now being adopted by other leading companies [Pla- uncertainty in individual process characterizations.
cek 1989a]. The Six Sigma model for tolerance accumulation
The basic premise of the Six Sigma Program is: in is shown in Eq. 6. It accounts for process mean shift
order to achieve high quality in a complex product variations by using an effective standard deviation
comprised of many components and processes, as expressed in Eq. 5. If a value for O"i for the pro-
each component and process must be produced at cess is known from prior experience, it may be sub-
significantly higher quality levels in order for the stituted in the equation.
composite result to meet final quality standards.
Stated statistically, suppose there were 1000 dimen-
sions or other characteristics of your product, any
0Xi 3Cpi(1 - mi)/ J -< TASM (6)
one of which could lower the quality of the finished
product. If each characteristic were produced to
---3o- quality (99.73% acceptable parts or 2700 de- Of course, tolerance analysis of assemblies is
fects per million), the resultant assemblies would be only one component of the complete Motorola qual-
only (.9973) 1°°° = .067 or 6.7% defect-free. To have ity management system, but the Six Sigma toler-
99.73% defect-free assemblies, you would need to ance analysis model is a significant contribution. It
produce each component to a quality of (.9973) '°°1 = is more realistic and versatile than the models com-
.9999973 or 99.99973%, which is 2.7 defects per mil- monly used for design. It should have a major im-
lion. pact on reducing production costs and improving
To achieve the high-quality levels required for quality [Harry et al. 1987, 1988, 1990].
26 Chase and Parkinson: Application of Tolerance Analysis

3 2-D Tolerance Analysis be solved by iterative means. If the nominal values

may be determined from, say, a precise CAD lay-
Modeling tolerance accumulation in 2-D assemblies out, this step may be omitted.
is much more difficult than for 1-D. Component di- Second, linearize the equations for small varia-
mensions are joined together as 2-D vector chains tions about the nominal by Taylor's series expan-
or loops. The loops pass from mating part-to-mating sion, retaining first-order derivatives. Substitute the
part, passing through the points of contact between design tolerances and solve this system of linear-
parts. Vector loops may be open or closed. Each ized equations for the corresponding variation in
open loop describes a resultant assembly clearance the kinematic variables and assembly resultants
or interference. Closed loops describe assemblies [Marler 1988, Chase et at. 1989].
containing some adjustable element, such as a
spring-loaded part or fastener, which takes up the
3.2 Geometric Tolerances
slack and assures closure.
Complex assemblies may require several open Geometric tolerances are being used increasingly by
and closed loops to define assembly relationships. aerospace and other industries to assure form and
Each vector loop results in three scalar equations function of mechanical parts. They are distinct from
which describe the response to manufacturing vari- size or dimensional tolerances. They control the
ations. The resulting nonlinear system of equations form and orientation (flatness, roundness, perpen-
must be solved simultaneously for the assembly re- dicularity, etc.) and location (position, concentric-
sultants [Fortini 1967, Chase and Greenwood 1988, ity, etc.) of surfaces and other features as defined in
Marler 1988, Bjorke 1989]. the standard, ANSI Y14.5 [ASME 1983]. Recom-
mended practice is to establish geometric toler-
ances on the basis of "Maximum Material Condi-
3.1 Kinematic Adjustments
tion" (MMC), which is essentially a WC analysis,
Manufacturing variations propagate through an as- resulting in tight component tolerances. The stan-
sembly by small kinematic adjustments between dard does not include statistical considerations or
mating parts. Thus, kinematic constraints must be tolerance accumulation effects [Foster 1986, Levy
applied to the vector loops. Each kinematic con- 1974].
straint introduces kinematic degrees of freedom However, geometric deviations provide sources
into the model, such as the location of a point of of variation that can accumulate and propagate
contact on a sliding plane or the relative angle be- through an assembly the same as size tolerances.
tween two mating parts which are joined by a pin Depending on the number of components and the
joint. The nominal values of the kinematic variables geometry, they can have a significant influence on
are not known. They are not dimensioned on any the resultant assembly variations. Efforts are being
engineering drawing. They must be determined by made to analyze geometric tolerances statistically
assembling an ideal assembly for which all manufac- and include geometric variations in vector loop
tured dimensions are at their nominal values. They models in order to compute their effects on complex
may then be calculated by solving the set of vector assemblies along with dimensional variations [Chun
loop equations and kinematic constraints describing 1988, Chase et al. 1989].
the system.
The variations in the kinematic variables are also
3.3 Sample 2-D Problem
unknown. Small changes in the manufactured di-
mensions produce small changes in the kinematic Figure 3a illustrates a simple 2-D problem described
variables. Thus, the kinematic variables are depen- by Fortini [1967]. It is a drawing of a one-way me-
dent variables. The variation in the kinematic vari- chanical clutch. This is a common device used to
ables may be determined by specifying the manu- transmit rotary motion in only one direction. When
facturing variations (design tolerances) and solving the outer ring of the clutch is rotated clockwise, the
the vector loop equations for the resulting adjust- rollers wedge between the ring and hub, locking the
ments in the assembly. two so they rotate together. In the reverse direc-
Thus, there are two steps in solving the vector tion, the rollers just slip, so the hub does not turn.
equations. First, set all the manufactured dimen- The angle ~1 between the two contact points is crit-
sions to their nominal values and solve the system ical to the proper operation of the clutch. If ~1 is
of vector loop equations for the nominal values of too large, the clutch will not lock; if it is too small
the kinematic variables and assembly resultants. the clutch will not unlock.
Vector equations are generally nonlinear and must The primary objective of performing a tolerance
 Chase and Parkinson: Application of Tolerance Analysis 27

ONE-WAY CLUTCH ASSEMBLY KN

I EMATJC
ION
I TS
Sf ~ ~ R i n P r i n g ~ . g ~ i rs
°° ........

b Dependent
Variables:b, d~.,~2

VECTOR LOOP ASSEMBLY GEOMETRIC FORM TOLERANCES

~ Hub~ ~ ' ~Ring

Fig. 3. Analysis of a one-way mechanical
clutch: (a) 2-D clutch assembly; (b) 2-D
kinematic joints applied to contact points;
(e) vector loop model of clutch; and (d)
t addition of geometric form tolerances.

analysis on the clutch is to determine how much the The predicted variation in assembly resultants may
angle ~1 is expected to vary due to manufacturing then be estimated by modifying Eq. 2 to include
variations in the clutch components. The indepen- form variations:
dent manufacturing variables are the hub dimension
a, the cylinder radius c, and the ring diameter e. The
dependent assembly resultants are the location of
d U = IN (0f~2 T2 +
\axi/
F.(Of]2da~]~/2
\0o~i/
(7)
contact point, b, and the two angles ~ and q~z.
In this example, the designer begins the con- where dai are the form variations and dU is the
struction of a vector kinematic model by creating a resultant assembly variation d~l.
precise CAD model of the assembly, to which the Once an expression for d ~ l has been obtained,
vector model will be added as an overlay. On the tolerance design may proceed. By substituting rea-
CAD model, he defines the kinematic joints at the sonable values for the component tolerances Ti and
contact locations between parts (Fig. 3b). When the form tolerances doq into Eq. 7, the designer must
parts are properly constrained by the joints, they verify that the predicted variation in ~1 will be less
will have zero kinematic degrees of freedom. than design requirement TASM. Alternately, to re-
Next, the joints are connected by vectors form- duce production cost, one might set d ~ l equal to
ing a closed vector loop and making sure all critical TASMand solve for the largest possible values for Ti.
dimensions have been included in at least one loop
(Fig. 3c). The geometry from the loops and joints is 4 3-D Tolerance Analysis
used to calculate the sensitivity of the dependent
4.1 3-D Solid Models
kinematic dimensions to changes in each indepen-
dent dimension. This information is used for com- An alternative approach to vector assembly models
putation of the expected variation in assembly re- is to create a solid model of the assembly on a CAD
sultants and possible re-allocation of tolerances. system. The solid model serves as the assembly
Geometric form tolerances may be added at each function. Small changes can be simulated and their
contact joint (Fig. 3d). The sensitivity of the depen- effects will propagate realistically, provided each
dent assembly dimensions b, ~1, and ~2 to form part is located relative to its adjacent parts and pro-
and orientation variations may then be calculated. vided that kinematic adjustments are permitted.
28 Chase and Parkinson: Application of Tolerance Analysis

rameters [Martino and Gabriele 1989a, 198b]. Cur-

Variables I Component Dimensions rent research efforts by Turner include the addition
Model Solid Model ATi
of kinematic constraints [Turner 1990, Turner and
•M i ~ ,, of Assembly I Assembly Resultants
Srikanth 1990, Srikanth and Turner 1990].
............. AUk
A number of researchers are taking an axiomatic
Fig. 4. Computing tolerance sensitivities from the solid model of approach to 3-D tolerance representations in solid
an assembly. models. Requicha represents the model variations
as a pair of "offset boundaries," or offset surfaces,
which bound each ideal surface. The set of offset
Solid models are precise representations of as- boundaries form a tolerance zone which bounds the
sembly geometry. They are constructed from sur- entire part [Requicha 1983, 1986]. A similar defini-
faces or solid primitives. The main obstacle to the tion creates a "virtual boundary" formed by taking
use of commercial solid modelers for tolerance into account the combined effects of all applicable
analysis is the lack of conventional dimension and size and form tolerances [Jayaraman and Srinivasan
tolerance data. This imposes a great hardship upon 1989, Srinivasan and Jayaraman 1989]. Several
software application developers. The trend toward problems remain to be resolved, including potential
feature-based solid modelers may help to overcome conflicts with existing standards, incorporation of
this deficiency. statistical models, and the lack of kinematic assem-
To obtain the sensitivities required for calculat- bly interactions. For commentaries on these issues,
ing an assembly tolerance sum, the relationship be- the reader is referred to Faux [1986] and Etesami
tween the parameters defining the model surfaces or [1987].
solid primitives and the dimensioned quantities ap- Variational geometry is another fundamental ap-
pearing on an engineering drawing of the parts must proach. It requires the formulation of analytical
be determined. This may be accomplished by mak- equations describing the geometric relationships
ing small changes in each of the model variables, which must be maintained in an assembly. Con-
measuring the resultant change in the component straints such as perpendicular surfaces or surfaces
dimensions and assembly resultants as shown in in sliding contact are defined in terms of dimen-
Fig. 4 and computing the corresponding sensitivi- sional parameters. If the design is modified, the sys-
ties. The sensitivities are used to form the linearized tem of equations may be solved to adjust the free
expressions, Eq. 8 shows the relationship between variables in keeping with the constraints. The ad-
the variations in the component dimensions and as- vantages of this method are the ease of design itera-
sembly resultants to variations in the model param- tion and the realistic propagation of manufacturing
eters. Finally, a linear programming problem is set variations by kinematic adjustments. However, the
up to find a set of model variations, dMj, which, resulting system of nonlinear equations can become
when substituted into Eq. 8, satisfies the specified very large and must be solved simultaneously.
component and assembly tolerance limits [Turner Also, geometric form and feature tolerances must
and Wozny 1987, 1990, Turner et al. 1987]. still be taken into account [Light and Gossard 1982,
Gossard et al. 1988, Chung and Schussel 1990].
Component tolerances:
4.2 3-D Vector Models
Of
dXi = X ~ dMj ~ Ti Vector loop models of assemblies may be applied to
(8) 3-D assemblies [Wirtz 1988]. The vectors are not
Assembly tolerances: confined to a plane and the kinematic conditions
d U i : [•( Og ]e dM~] '/2 --~ TASM become more complex. The system of equations is
twice as large, since each 3-D vector equation
yields six scalar equations. Form and orientation
Solid modelers are CPU intensive. Changing a tolerances may be added. Their interaction with the
single parameter for a sensitivity calculation re- kinematic joint axes must be carefully modeled. Re-
quires regeneration of the entire CAD geometry. A cently, vector models of an assembly have been
detailed model of an assembly may have thousands overlaid on the corresponding 3-D solid model and
of model parameters, resulting in a substantial wait associated with that model such that changes in the
for a complete sensitivity calculation on all but the solid model are automatically reflected in the vector
most powerful computers. However, significant model [Robison 1989].
progress is being made in reducing the enormous One major advantage of 3-D vector models of
number of sensitivity calculations by prior examina- assemblies over solid models is that the geometry is
tion of the model to eliminate noncontributing pa- reduced to only those parameters required to per-
 Chase and Parkinson: Application of Tolerance Analysis 29

form a tolerance analysis. Sensitivity analysis is 6 Nonlinear Analysis, Non-Normal Distributions

much simpler. It is very efficient computationally
6.1 Nonlinear Analysis
and well suited to design iteration. Describing man-
ufacturing variations with vectors and kinematics is The linearized models for tolerance accumulation in
also a medium that engineering designers are al- an assembly, as expressed in Eqs. 1-4, assume that
ready familiar with. the sensitivity, evaluated at the nominal, is constant
over the tolerance limits. That is, if you evaluated
5 Tolerance Analysis of Mechanisms the assembly function as you varied one parameter
over its tolerance range, the slope of the function
Since tolerance analysis of assemblies involves the would be nearly constant. This is usually a reason-
creation of a kinematic model, the principal differ- able assumption, when the tolerances are very
ences from the classical kinematic analysis of mech- small compared to the nominal dimensions or when
anisms are the inputs and the magnitude of the there are a large enough number of components to
motions. A mechanism will have one or more mask the effects.
kinematic variables as a prescribed input. The non- In a highly nonlinear assembly, the sensitivity
linear system of kinematic equations will be solved may not be symmetric over this range and the distri-
for the remaining kinematic variables. The resulting bution of the assembly resultant will be skewed or
motions will be orders of magnitude larger than asymmetric. This can happen even though all of the
those due to manufacturing variations. component distributions are symmetric. A linear-
For tolerance analysis of a mechanism, the di- ized model, however, will always yield a symmetric
mensions of the various components will be held resultant from symmetric inputs.
fixed while the nominal position and orientation of Analysis methods which can treat nonlinear ef-
each component is determined by kinematic analy- fects are shown in Table 2. While Hasofer-Lind re-
sis. Then a tolerance analysis may be performed by tains nonlinear effects, it is limited to Normal distri-
introducing manufacturing variations while holding butions. The relative CPU efficiency values are
the mechanism stationary. This may be repeated at only estimates, based on the author's own experi-
selected positions considered of interest or the ence. Actual values could differ substantially de-
mechanism may be incremented at regular intervals pending on problem complexity.
to see the results of manufacturing variations over a
range of motion. 6.2 Non-Normal D&tributions
Designers seldom have sufficient data by which to
specify the distribution of the manufacturing pro-
Table I. Referencesto mechanism toleranceanalysis cesses. Data has not been gathered because the
parts have not yet been made. Tooling has not been
Mechanism Analysis Authors ordered nor certified. The processes may not have
4-bar function Tolerance [Garrettand been selected. It is customary in such cases for de-
generator estimating Hall 1969] signers to assume a Normal distribution. If there is
Tolerance [Dhandeand uncertainty about the process, then a Uniform dis-
allocation Chakraborty1973] tribution may be assumed. Generally, deviations
Tolerance [Agarwal1981] from Normal are slight and vary from batch to
allocation
4-bar path Tolerance [Baumgartenand batch, making it difficult to predict. If the compo-
generator estimating Van der Werff 1985] nent distributions have a strong central tendency,
Adjustable [Schade1982] and there are five or more components contributing
linkages to an assembly sum, the result is likely to approxi-
Tolerance [Mallikand
mate a Normal distribution regardless of the com-
allocation Dhande1987]
Slider-crank Tolerance [Fu et al. 1987, ponent distributions. If there are mean shifts, one of
allocation Schade1980] the models discussed earlier may be used.
General 2-D linkages [Fentonet However, in some cases, where the process has
Tolerance al. 1989] been well characterized and is known to exhibit
allocation
3-D 4-bar function T o l e r a n c e [Dhandeand skewness (asymmetry) or kurtosis (peakedness), it
generator allocation Chakraborty1978] may be justifiable to estimate these parameters and
Tolerance [Beoharand apply an advanced analysis method, such as Monte
allocation Rao 1980] Carlo or Method of Moments. This is more likely to
Disk cams Tolerance [Rao and occur after production has begun and data has been
allocation Gavane1980]
gathered on finished parts.
30 Chase and Parkinson: Application of Tolerance Analysis

Table 2. Comparison of tolerance analysis methods

Efficiency
Assembly model Distributions Relative
Analysis method Linearized Nonlinear Normal Non-normal CPU time

WC X X NA NA 1
RSS X X 1
Hasofer-Lind X X 6
Method of Moments X X X 10
Integration X X X 60
Monte Carlo X X X X 100

6.3 Monte Carlo Simulation Component Distributions

Monte Carlo Simulation is a powerful tool for toler-
ance analysis of mechanical assemblies, for both
/b,.
nonlinear assembly functions and non-Normal dis-
tributions. It is based on the use of a random num-
ber generator to simulate the effects of manufactur-
ing variations on assemblies. Figure 5 illustrates the
method, which consists of the following steps:
1. A critical assembly resultant is identified and de-
sign limits are specified.
2. The component dimensions which contribute to
Fig. 5. Assembly tolerance analysis by Monte Carlo simulation.
the critical resultant are identified and tolerances
are specified for each dimension.
3. Also specified is the statistical distribution for
the variation in each component dimension. The manufacturing processes. For example, for a -+30-
distribution may be described algebraically or assembly spec, a sample of I00,000 assemblies
empirically. should yield 135 rejects at each limit, but could be
4. An assembly function is formulated relating the 10-20% off [Shapiro and Gross 1981]. Design itera-
component dimensions to the resultant assembly tions get pretty tedious when 100,000 simulations
dimension. must be generated for each trial design. The de-
5. A set of component dimensions for a single as- signer is not likely to have the patience to search for
sembly is selected using a random number gener- the optimum design.
ator to apply a small variation to each dimen- References to nonlinear or non-Normal analysis
sion. The resultant assembly dimension is are listed in Table 3.
calculated by means of the assembly function
and compared to the assembly limits to deter-
mine if it is within spec.
6. Step 5 is repeated until a sufficient number of Table 3. References to nonlinear or non-Normal Analysis
assemblies has been simulated to plot a histo- Method Authors
gram and estimate the percent of assemblies that
would be rejected based on the specified toler- Worst Case [Greenwood and Chase 1988a]
Hasofer-Lind [Parkinson 1982, 1985, Lee and Woo 1990,
ances. A variation on this step is to fit a distribu-
Greenwood and Chase t990]
tion to the histogram and use the distribution Method of [Evans 1970, 1974, 1975a, 1975b, Cox 1979,
function to calculate the percent rejects. Moments 1986, Shapiro and Gross 1981, Greenwood
1987]
The biggest disadvantage of the Monte Carlo Integration [Evans 1967, 1971, 1972, Sorensen 1991]
method is that it requires large samples to achieve Monte Carlo [Grossman 1976, Shapiro and Gross 1981,
reasonable accuracy. The number of simulated as- DeDoncker and Spencer 1987, Craig 1989,
semblies must be on the order of 100,000 to 400,000 Doepker and Nies 1989, Early and
Thompson 1989]
to predict the small percentage rejects of modern
 Chase and Parkinson: Application of Tolerance Analysis 31

7 Design Improvement
!
Design improvement is the principal aim of toler- ce

ance analysis. Rather than just predict the effects of ¢:

.2 ~d
variations, the goal is to systematically select toler- ice
II)
ances throughout an assembly to assure that design E
requirements will be met. The ideal assigned toler-
ances not only assure acceptable performance, but dnts
also assure that parts can readily be produced and
assembled, resulting in high process yields and re-
duced costs. Several strategies for design improve- a Dimension 1
ment exist. A good general discussion is presented
in the excellent book by Spence and Soin [1988].
Although the book is applied to electronics design,
the methods are just as applicable to mechanical
systems. ¢N
LI
The main topics in design improvement are: i-
o
ce

Yield Modification, Tolerance Allocation, Sensitiv-

ity Analysis, and Probabalistic Design and Design E
Optimization. ~5

7.1 Yield Modification

The yield of an assembly process may be increased b Dimension 1
by design centering. The procedure is illustrated in
Fig. 6a. The figure shows the effect of varying two
design parameters, p] and P2. The region defined by I Oriainal
shaded boundaries is called the feasible design ¢q
space and represents the limits set on the values of
p] and pz for acceptable performance. The rectangle O

represents the specified production limits on p~ and

P2, that is, the tolerance limits, which vary about a E
specified nominal value. As can be seen, the origi-
nal value of the nominal places most of the rectan-
gle outside of the acceptable performance region.
The resulting design would have a low yield. By
adjusting the nominal to the center of the feasible C Dimension 1
design region, nearly all of the assemblies will per-
form satisfactorily. Fig. 6. Illustration of design improvement methods: (a) design
centering; (b) variance reduction; and (e) tolerance allocation.
The second method of increasing yield is by vari-
ance reduction, that is, by tightening the tolerances,
as shown in Fig. 6b. The new, tighter tolerances
place all of the produced assemblies inside the ac-
ceptable region. Of course, tighter tolerances are fore any parts have been produced or tooling or-
more costly to produce. But the increased cost may dered. It involves first, deciding what tolerance
be partially offset by the reduction in waste and limits to place on the critical clearances and fits for
rework. The optimum tolerances may be found by an assembly, based on performance requirements;
minimizing the overall cost of tight tolerances, second, creating an assembly model to identify
waste and rework [Spence and Sion 1988]. which dimensions contribute to the final assembly
dimensions; and third, deciding how much of the
assembly tolerance to assign to each of the contrib-
7.2 Research in Tolerance Allocation
uting components in the assembly. Figure 6c shows
Tolerance allocation is a design function. It is per- the tolerance on dimension 2 reduced, allowing an
formed early in the product development cycle, be- increase in tolerance on dimension 1.
32 Chase and Parkinson: Application of Tolerance Analysis

Tolerance analysis, on the other hand, is a pro- allocation of the available tolerance [Harry and
duction function. It is performed after the parts are Stewart 1988].
in production. It involves first, gathering data on the
individual component variations; second, creating
an assembly model to identify which dimensions 7.2.2 Cube root of the nominal. This method is
contribute to the final assembly dimensions; and based on the rule-of-thumb that the difficulty in ob-
third, applying the measured component variations taining a specified tolerance increases as the cube
to the model to predict the assembly dimension var- root of the nominal size of the part. The rule is the
iations. basis for the early tolerance standards for cylindri-
A defective assembly is one for which the com- cal fits [Fortini 1967]. The procedure is to select
ponent variations accumulate and exceed the speci- initial tolerance values equal to the cube root of the
fied assembly tolerance limits. The yield of an as- nominal, substitute into the assembly tolerance sum
sembly process is the percent of assemblies which equation, then scale proportionally. The resulting
are not defective. In tolerance analysis, component tolerances will each be proportional to the cube root
variations are analyzed to predict how many assem- of their nominal size [Chase and Greenwood 1988].
blies will be in spec. If the yield is too low, rework,
shimming, or parts replacement may be required. In
7.2.3 Difficulty factors. This is an extension of
tolerance allocation, an acceptable yield of the pro-
the cube root method, with more categories of diffi-
cess is first specified and component tolerances are
culty, such as size, shape, material, process, etc.,
then selected to assure that the specified yield will
where each category refers to a property affecting
be met.
the cost of producing a tolerance. The designer as-
Often, tolerance design is performed by repeated
signs a difficulty factor to each component toler-
application of tolerance analysis, using trial values
ance based on nominal size, then assigns another
of the component tolerances. However, a number
factor to each component based on shape (inside
of algorithms have been proposed for assigning tol-
dimension, outside dimension, etc.), and repeats
erances on a rational basis, without resorting to trial
this process for each category, writing the factors in
and error. Several are listed in Fig. 7.
a table. The difficulty factors for each component
dimension are summed and used as weight factors
in the tolerance sum equation to drive the allocation
7.2.1. Proportional scaling. By this procedure,
[Fortini 1967, 1985].
initial values of the component tolerances are se-
lected, substituted into the assembly tolerance sum
equation, then scaled proportionally so the sum 7.2.4 Minimum cost. If an empirical function of
equals the assembly tolerance limit. Initial toler- cost-vs-tolerance (or process capability) can be ob-
ance values may be selected from charts of toler- tained for each dimension in the assembly sum,
ance capabilities for specified processes, from de- then an optimization algorithm may be used to sys-
sign rules, standards, etc. [Mansoor 1963, Chase tematically search for the combination of compo-
and Greenwood 1988, Bjorke 1989]. nent tolerances which results in the least overall
A variation on this method adds flexibility by production cost. Numerous researchers have pro-
specifying weight factors to certain component tol- posed different search algorithms and different
erances so those components will receive a greater forms of empirical cost functions, as summarized in
Table 4.
The constant coefficient A represents the fixed
costs, such as tooling, setup, prior operations, etc.
ToLERANcE ALLOCATION I The B term represents the cost of producing a single
I component dimension to a specified tolerance T. All
I I costs are calculated on a per part basis.
Design Rules I Search RulesJ
Standards Difficulty Factors
7.2.5 Minimum cost with process selection. Op-
Cube Root Cost Optimized
timization procedures have been extended to not
Process Limits Taguchi
only find the least cost set of tolerances, but to also
Nonlinear Programming
select the least cost process from a set of alternative
Fig. 7. Tolerance allocation methods. processes for each dimension for the assembly.
 Chase and Parkinson: Application of Tolerance Analysis 33

Table 4. Proposed cost-vs-tolerance models

Cost model Method Author

Linear A - BT Linear prog [Edet and Auer 19651

Reciprocal A + B/T Lagrange mult [Chase and Greenwood 1988]
Nonlin prog [Parkinson 19851
Reciprocal squared A + B/T 2 Lagrange mult [Spotts 1973]
Reciprocal power A + B/T k Lagrange mult [Sutherland and Roth 1975]
Multi/Recip powers B/T kl Nonlin prog [Lee and Woo 1990]
Lagrange mult [Bennett and Gupta 1%9]
Lagrange mult [Chase et al. 1990]
Nonlin prog [Andersen 1990]
Exponential Be -rot Lagrange mult [Speckhart 1972]
Geom prog [Wilde and Prentice 1975]
Graphical [Peters 1970]
Expon/Recip power Be-mX/T k Nonlin prog [Michael and Siddall 1981, 1982]
Piecewise linear A~ - BtT~ Linear prog [Bjork 1989, Patel 1980]
Empirical data Discrete points Zero-one prog [Ostwald and Huang 1977]
Combinatorial [Monte and Datseris 1982]
Branch & Bound [Lee and Woo 1989]

That is, the computer can decide which process is accumulation expressions of Eqs. 1-4. The toler-
the most economical to produce each part dimen- ance sensitivity tells the designer which assembly
sion while considering the tolerances of all of the parameter variations have the greatest effect on the
parts and their cost interactions [Ostwald and critical assembly features. Listing the parameters
Huang 1977, Lee and Woo 1989, Chase et al. 1990]. and their corresponding sensitivities in order of de-
creasing magnitude reveals which components to
focus on for design improvement. Alternately, one
7.2.6 Minimum cost--complex assembly models. could list the product of the sensitivities and their
Optimization procedures may also be applied to corresponding tolerances in descending order and
complex assemblies defined by 2-D or multiple vec- also calculate the percent contribution made by
tor loops. Further extensions have been studied as each to the assembly resultant. Then, starting with
listed in Table 5. the largest contributor, the designer could try to
decrease the overall variation by tightening toler-
ances on the most sensitive components or de-
7.3 Sensitivity Analysis crease the overall cost by loosening the tolerance
The third area of design improvement stems from on the least sensitive components [Eaton 1975].
examining the tolerance sensitivities, which are the Sensitivity reduction is another approach in
partial derivative terms appearing in the tolerance which the sensitivity itself is reduced by moving the
nominal values to a less sensitive portion of feasible
design space. Figure 8 illustrates this method. In the
figure, the contour lines represent lines of constant
Table 5. References to tolerance allocation for mechanisms assembly performance. Closely spaced contours in-
dicate a region of high variability in performance.
Method Authors
By moving the nominal design from a region of high
Nonlinear assemblies [Lee and Woo 1990] variability to a region of low variability, as shown in
2-D assemblies [Sutherland and Roth 1975, the figure, the design is made insensitive, or robust,
Monte and Datseris 1982, to manufacturing variations. A systematic proce-
Parkinson 1985, Andersen
dure for accomplishing this is the popular Taguchi
1990]
Multiple loop assemblies [Bennett and Gupta 1969, Lee method developed by the well-known Japanese ex-
and Woo 1990, Andersen pert on quality control [Taguchi 1986, Kacker 1986,
1990] Byrne and Taguchi 1987, Taguchi et al. 1989].
Process mean shifts [Andersen 19901 A fundamental element of the Taguchi method is
Non-normal distributions [Parkinson 1985]
the formulation of a Quality Loss Function which
34 Chase and Parkinson: Application of Tolerance Analysis

design is essentially complete and all nominal val-

ues for the design have been determined. However,
by considering the effects of tolerances during the
selection of the nominal values of design variables,
it is possible to develop a "robust" design that is
¢- more tolerant of variation. Developing a robust de-
O
¢/1
sign by judicious selection of variable nominal val-
e.,
ues is an important part of the Taguchi philosophy.
E Taguchi's methodology develops a model of the
design problem by direct experimentation. How-
ever, when a computer model of the design exists,
an appropriate means of developing a robust design
is through nonlinear optimization techniques.
These methods can be used at two levels. The
Dimension 1
first level is to select design variable values such
Fig. 8. Sensitivity reduction by shifting the nominal values. that the design remains feasible (i.e. will still func-
tion properly) despite variations arising from toler-
ances. The basic approach here is to calculate the
variation caused by tolerances using either a first-
order Taylor's series (Eq. 1) or through Monte
quantifies the cost of deviating from the target value
Carlo simulation. The transmitted variation is then
of a given design parameter. The loss function is
subtracted from the allowable values of the con-
expressed as a parabola, with minimum cost at the
straints, causing a shift of the optimum design into
target value and increasing with the square of the
the feasible region--as shown in Fig. 9. A good
deviation. It can include the full spectrum of costs,
review of applications at this level is given by
including inferior performance, increased rework
Eggert and Mayne [1990].
and warranty costs, dissatisfied customers and lost
The second level is to explicitly consider the var-
market share. Taguchi also includes the "loss to
iation from tolerances as an objective or constraint
society," if it can be quantified [Taguchi 1986].
in the problem. At this level the designer seeks to
actively control variation by either minimizing it as
one of the objectives in the design problem or by
7.4 Probabalistic Design and Design constraining the design to have variation less than a
value specified by the designer. Active control of
Optimization
variation can be computationally expensive since it
Probabatistic design may be considered to be an requires second derivatives of model equations. A
extension of tolerance analysis methods to include good discussion with examples of this level is given
consideration of the variational effects of both geo- by Parkinson et al. [1990].
metric and engineering parameters on design per-
formance. Engineering parameters, such as the lim-
iting strength of a metal or the viscosity of a
lubricant, exhibit manufacturing variations which
can be characterized by statistical distributions. By Reduced Feasible
applying statistical analysis to the engineering per- De.sign Region
formance equations for stress, lubrication, etc., the o~ Modified Constraints
variation in critical performance resultants can be ~.~J__ Due to Variation
predicted [Haugen 1980, Mischke 1989].
Design optimization is a mathematical method "e.
>
t ~ ~ . K , ~ , Optimum
for improving a design by applying linear or nonlin-
= ,i',il ~ N " " , ~ _ . . ~ O r i g i n al
ear programming techniques to search systemati- "~ ,,,,,, OpUmum N):.' \ ~ Design
cally for the minimum of an objective function. The a ,'~ t Constraints
objective function is derived from the engineering
model and expresses some critical performance pa-
rameter which is to be minimized, such as the Design Variable 1
weight of a structure or cost of a fluid distribution Fig. 9. Change in optimum and decrease in f~asible region to
system. insure design will remain feasible to variation caused by toler-
Often tolerance analysis is performed after the ances.
 Chase and Parkinson: Application of Tolerance Analysis 35

8 Summary Chase, K.W. and W.H. Greenwood (1988), "Design Issues in

Mechanical Tolerance Analysis," Manufacturing Review,
ASME, v 1, n 1, pp. 50-59, March
There is probably no other design improvement ef-
Chase, K.W. and W.H. Greenwood, B.G. Loosli, and L.F.
fort which can yield greater benefits for less cost Hauglund (1990), "Least Cost Tolerance Allocation for Me-
than the careful analysis and assignment of toler- chanical Assemblies with Automated Process Selection,"
ances. Tolerancing provides a common meeting Manufacturing Review, ASME, v 3, n 1, pp. 49-59, March
ground for engineering and production personnel Chase, K., C. Sorensen, and C. Andersen (1989), "Recent De-
velopments in Tolerance Analysis Software for Mechanical
where effective communication can assure that
Assemblies, Part A: Two-Dimensional Modeling," Proceed-
their competing requirements are met in the most ings of the ASQC Western Regional Conference--1989, pp.
economical way, with the greatest customer satis- 73-80
faction. Chun, K. (1988), "Development of Two-Dimensional Tolerance
In the foregoing discussion, a number of research Modeling Methods for CAD Systems," ADCATS Report No.
88-7, Brigham Young University
areas have been surveyed where significant pro-
Chung, J.C.H. and M.D. Schussel (1990), "Technical Evaluation
gress is being made. As a result of current research, of Variational and Parametric Design," Computers in Engi-
powerful new design tools are becoming available neering-1990, ASME, 1, pp. 289-298
which incorporate improved methods for predicting Cox, N.D. (1979), "Tolerance Analysis by Computer," J. of
the effects of manufacturing variations on engineer- Quality Technology, v 11, n 2, pp. 80-87, April
Cox, N.D. (1986), "How to Perform Statistical Tolerance Analy-
ing performance and production quality. The effec-
sis," ASQC Series Vol 11, Basic References in Quality Con-
tive application of these concepts will assist manu- trol
facturing enterprises in competing in the worldwide Craig, M. (1989), "Managing Variation by Design Using Simula-
marketplace. tion Methods," Failure Prevention and Reliability--1989,
This literature survey has been so broad in scope ASME Publ. No. DE--Vol 16, pp. 153-163
DeDoncker, D. and A. Spencer (1987), "Assembly Tolerance
that only a few papers have been referenced in each
Analysis with Simulation and Optimization Techniques,"
area to permit some tutorial descriptions. We apolo- SAE Paper No. 870263
gize for any papers which may have been passed Dhande, S. and J. Chakraborty (1973), "Analysis and Synthesis
over in the selection process. A more complete bib- of Mechanical Error in Linkages--A Stochastic Approach,"
liography is available from the authors on request J. of Engineering for Industry, ASME, v 105, pp. 672-676
[Chase 1991]. Dhande, S. and J. Chakraborty (1978), "Mechanical Error Anal-
ysis of Spatial Linkages," J. of Mechanical Design, ASME,
v 100, pp. 732-738
Doepker, P.E. and D. Nies (1989), "Designing Brake Compo-
References nents Using Variation Simulation Modeling," Failure Pre-
vention and Reliability--1989, ASME Publ. No. DE--Vol
Agarwal, M. (1981), "Optimal Synthesis of Tolerance and Clear- 16, pp. 131-138
ance in Function Generating Mechanisms--A Parametric Early, R. and J. Thompson (1989), "Variation Simulation Model-
Programming Problem," ASME Paper No. 81-DET-5 ing--Variation Analysis Using Monte Carlo Simulation,"
Andersen, C.B. (1990), "General System for Least Cost Toler- Failure Prevention and Reliability--1989, ASME PubL No.
ance Allocation in Mechanical Assemblies," ADCATS Re- DE--Vol 16, pp. 139-144
port No. 90-2, Brigham Young University Eaton, F. (1975), "Computer Model of a Three-Dimensional As-
ASME (1990), "Research Needs and Technological Opportuni- sembly," ASME Paper No. 75-DE-21
ties in Mechanical Tolerancing," ASME Publication No. Edel, D.H. and T.B. Auer (1965), "Determine the Least Cost
CRTD-15 Combination for Tolerance Accumulations in a Drive Shaft
ASME (1983), "Dimensioning and Tolerancing," ANSI Seal Assembly," General Motors Engineering Journal,
Y14.5M-1982, ASME Publication Fourth Quarter, 1964, pp. 37-38, First Quarter, pp. 36-38
Baumgarten, J. and K. Van der Werff (1985), "A Probabilistic Eggert, R.J. and R.W. Mayne, (1990), "Probabilistic Optimiza-
Study Relating to Tolerancing and Path Generation Error," tion Using Successive Surrogate Probability Density Func-
J. of Mechanism and Machine Theory, v 20, n 1, pp. 71-76 tions, Proc. of ASME 16th Design Automation Conf., Chi-
Bender, A., Jr. (1968), "Statistical Toleraneing as it Relates to cago, IL, DE-Vol 23-1, p. 129, September
Quality Control and the Designer," SAE Paper No. 680490 Etesami, F. (1987), "On the Theory of Geometric Tolerancing,"
Bennett, G. and L.C. Gupta (1969), "Least Cost Tolerances--I Computers in Engineering--1987, ASME, 2, pp. 327-334
and II," International J. of Production Research, v 8, pp. Evans, D.H. (1967), "An Application of Numerical Integration
65-74 and 169-181 Techniques to Statistical Tolerancing," Technometrics, v 9,
Beohar, S. and A. Rat (1980), "Optimum Stochastic Synthesis n 3, pp. 441-456, August
of Four Bar Spatial Function Generators," ASME Paper No. Evans, D.H. (1970), " A Statistical Tolerancing Formulation," J.
80-DET-32 of Quality Technology, v 2, n 4, pp. 226-231, October
Bjorke, O. (1989), Computer-Aided Toleraneing, Second Ed., Evans, D.H. (1971), "An Application of Numerical Integration
ASME Press Techniques to Statistical Tolerancing, II--A Note on the Er-
Byrne, D. and S. Taguchi (1987), "The Tagnchi Approach to ror," Technometrics, v 13, n 2, pp. 315-324, May
Parameter Design," Quality Progress, pp. 19-26, December Evans, D.H. (1972), "An Application of Numerical Integration
Chase, K.W. (1991), "A Bibliography on Tolerance Analysis of Techniques to Statistical Tolerancing, III--General Distribu-
Mechanical Assemblies," ADCATS Report No. 91-2, tions," Technometrics, v 14, n 1, pp. 23-25, February
Brigham Young University Evans, D.H. (1974), "Statistical Tolerancing: State of the Art,
36 Chase and Parkinson: Application of Tolerance Analysis

Part 1. Background," J. of Quality Technology, v 6, n 4, pp. Kackar, R., (1986), "The Power of Taguchi Methods," Quality
188-195, October Progress, pp. 21-29, December
Evans, D.H. (1975a), "Statistical Tolerancing: State of the Art, Kendrick, J. (1991), "Customers 'Win' in Batdrige Award Selec-
Part 2. Methods of Estimating Moments," J. of Quality Tech- tions," Quality, pp. 23-31, January
nology, v 7, n 1, pp. 1-12, January Lee, W. and T.C. Woo (1989), "Optimum Selection of Discrete
Evans, D.H. (1975b), "Statistical Tolerancing: State of the Art, Tolerances," J. Mechanisms, Transmissions, and Automa-
Part 3. Shifts and Drifts," J. of Quality Technology, v 7, n 2, tion in Design, v 111, pp. 243-251, June
pp. 72-76, April Lee, W. and T.C. Woo (1990), "Tolerances: Their Analysis and
Faux, L D. (1986), "Reconciliation of Design and Manufacturing Synthesis," J. of Engineepqngfor Industry, v 112, pp. 1t3-
Requirements for Product Description Data Using Functional 121, May
Primitive Part Features," CAM-I Report No. R-86-ANC/ Levy, S. (1974), Applied Geometric Tolerancing, TAD Products
GM/PP-01.1, CAM-I Inc, Arlington, TX, December Corp., Beverly, MA
Fenton R., W. Cleghorn, and J. Fu (1989), "Allocation of Di- Light, R. and D. Gossard (1982), "Modification of Geometric
mensional Tolerances for Multiple Loop Assemblies," J. of Models Through Variational Geometry," Computer-Aided
Mechanisms, Transmissions, and Automation in Design, v Design, v 14, n 4, pp. 209-214, July
111, pp. 465-470, December Malfik, A. and S. Dhande (1987), "Analysis and Synthesis of
Fortini, E. (1967), Dimensioning for Interchangeable Manufac- Mechanical Error in Path-Generating Linkages Using a Sto-
ture, Industrial Press chastic Approach," J. of Mechanism and Machine Theory, v
Fortini, E. (1985), "Dimensions and Tolerances~" Mechanical 22, n 2, pp. 115-123
Design and Systems Handbook, Second Ed., H. Rothbart Mansoor, E. (1963), "The Application of Probability to Toler-
editor, McGraw-Hill ances Used in Engineering Designs," Proc. of the Inst. of
Foster, L.W. (1986), Geometrics II--The Application of Geo- Mech. Engineers, v 178, pt 1, n i, pp. 29-51
metric Tolerancing Techniques. Addison-Wesley Marler, J. (1988), "Nonlinear Tolerance Analysis Using the Di-
Fu, J., W. Cleghorn, and R. Fenton (1987), "Synthesis of Di- rect Linearization Method," ADCATS Report No. 88-6,
mensional Tolerances of a Slider Crank Mechanism," lOth Brigham Young University
Applied Mechanisms Conference, Oklahoma State Univer- Martino, P.M. and G.A. Gabriele (1989a), "Application of Varia-
sity tional Geometry to the Analysis of Mechanical Tolerances,"
Garrett, R. and A. Hall, Jr. (1969), "Effect of Tolerance and Failure Prevention and Reliability--1989, ASME Paper No.
Clearance in Linkage Design," J. of Engineering for Indus- DE--Vol 16, pp. 19-27
try, ASME, v 91, pp. 198-202, February Martino, P.M. and G.A. Gabriele (1989b), "Estimating Jacobian
Gladman, C.A. (1980), "Applying Probability in Tolerance Tech- and Constraint Matrices in Variational Geometry Systems,"
nology," Trans. Inst. Eng. Austral. Mech. Eng., v ME5, n 2, Failure Prevention and Reliability--1989, ASME Paper No.
p. 82 DE--Vot t6, pp. 79-84
Gossard, D.C., R.P. Zuffante, and H. Sakurai (1988), "Repre- Michael, W. and J.N, Siddall (198i), "The Optimization Problem
senting Dimensions, Tolerances, and Features in MCAE Sys- with Optimal Tolerance Assignment and Full Acceptance,"
tems," IEEE Computer Graphics & Applications, v 8, n 2, Y. of Mechanical Design, ASME, 103, pp. 842-848, October
pp. 51-59, March Michael, W., J.N. Siddall (1982), "The Optimal Tolerance As-
Greenwood, W.H. (1987), "A New Tolerance Analysis Method signment with Less Than Full Acceptance," J. of Mechani-
for Designers and Manufacturers," (Dissertation), ADCATS cal Design, ASME, 104, pp. 855-860, October
Report No. 87-2, Brigham Young University. Mischke, C.R., (1989), "Stochastic Methods in Mechanical De-
Greenwood, W.H. and K.W. Chase (1987), "A New Tolerance sign, Parts 1-4," Failure Prevention and Reliability--1989,
Analysis Method for Designers and Manufacturers," J. of ASME Publ. No. DE-Vol 16, pp. 1-28
Engineering for Industry, ASME, v 109, pp. 112-116, May Monte, M.E. and P. Datseris, (1982), "Optimum Tolerance Se-
Greenwood, W.H. and K.W. Chase (1988a), "Worst Case Toler- lection for Minimum Manufacturing Cost and Other Crite-
ance Analysis with Nonlinear Problems," J. of Engineering ria," ASME Paper No. 82-DET-35, pp. 1-9
for Industry, ASME, v 110, pp. 232-235, August Ostwald, P.F. and J. Huang (1977), "A Method for Optimal
Greenwood, W.H. and K.W. Chase (1990), "Root Sum Squares Tolerance Selection," J. of Engineering for Industry, ASME,
Tolerance Analysis with Nonlinear Problems," J. of Engi- 99, pp. 558-565, August
neering for Industry, ASME, v 112, pp. 382-384, November Parkinson, A.R., C. Sorensen, J. Free, and B. Canfield, (1990),
Grossman, D. (1976), "Monte Carlo Simulation of Tolerancing in "Tolerances and Robustness in Engineering Design Optimi-
Discrete Parts Manufacture," Stanford Artificial Intelligence zation," Proc. of ASME 16th Design Automation Conf., Chi-
Laboratory, Memo AIM-280, Computer Science Dept. Re- cago, IL, DE-Vol 23-1, p. 121, September
port No. STAN-CS-76-555, May Parkinson, D.B. (1982), "The Application of Reliability Methods
Harry, M. (1987), "The Nature of Six Sigma Quality," Motorola to Tolerancing," J. of Mechanical Design, v 104, pp. 612-
Corporation 618, July
Harry, M. and J. Lawson (1990), "Six Sigma Producibility Anal- Parkinson, D.B. (1985), "Assessment and Optimization of Di-
ysis and Process Characterization," PubIication No. 60-3-03/ mensional Tolerances," Computer-Aided Design, 17, 4, pp.
88, Motorola Corporation 191-199, May
Harry, M. and R. Stewart (1988), "Six Sigma Mechanical Design Patel, A.M. (1980), "Computer-Aided Assignment of Manufac-
Tolerancing," Publication No. 6(r-2-10/88, Motorola Corpo- turing Tolerances," Proc. of the 17th Design Automation
ration Conf., Minneapolis, MN, June
Haugen, E. (1980), Probabilistic Mechanical Design, John Wiley Peters, J. (1970), "Tolerancing the Components of an Assembly
Jayaraman, R. and V. Srinivasan (1989), "Geometric Toleranc- for Minimum Cost," J. of Engineering for Industry, ASME,
ing: 1 Virtual boundary requirements, IBM J. of Research pp. 677-682, August
and Development, v 33, n 2, pp. 90-104, March Placek, C. (1989a), "Motorola, Westinghouse Nuclear Fuel
 Chase and Parkinson: Application of Tolerance Analysis 37

Unit, Globe Metallurgical Named Malcolm Baldridge Na- Spotts, M.F. (1978), "How to Use Wider Tolerances, Safely, in
tional Quality Award Winners," Quality, pp. 13-14, January Dimensioning Stacked Assemblies," Machine Design, pp.
Placek, C. (1989b), "Mechanical Tolerancing Workshop," Qual- 60-63, April
ity, pp. 16-17, December Spotts, M. F. (1983), Dimensioning and Tolerancing for Quan-
Placek, C. (1990), "Milliken and Xerox Garner 1989 Baldrige tity Production, Prentice-Hall.
National Quality Awards," Quality, pp. 13-14, January
Srikanth, S. and J. Turner (1990), "Toward a Unified Represen-
Rao, S., and S. Gavane (1980), '°Analysis and Synthesis of Me-
tation of Mechanical Assemblies," Engineering with Com-
chanical Error in Cam-Follower Systems," ASME Paper No.
puters, v 6, pp. 103-112
80-DET-22
Requicha, A.A.G. (1983), "Toward a Theory of Geometric Srinivasan, V. and R. Jayaraman (1989), "Geometric Toleranc-
Tolerancing," Int. J. of Robotics Research, v 2, no 4, pp. 45- ing: 2 Conditional Tolerances, IBM J. of Research and Devel-
60, Winter opment, 33, 2, pp. 105-124, March
Requicha, A.A.G. (1986), "Representation of Geometric Fea- Sutherland, G. H. and Roth, B. (1975), "Mechanism Design:
tures, Tolerances and Attributes in Solid Modelers Based on Accountingfor Manufacturing Tolerances and Costs in Func-
Constructive Solid Geometry," 1EEE J. of Robotics and Au- tion Generating Problems," J. of Engineering for Industry,
tomation, RA-2 no 3, pp. 156-166, September ASME, v 97, pp. 283-286, February
Robison, R.H. (1989), "A Practical Method for Three-Dimen- Taguchi, G. (1986), Introduction to Quality Engineering, Asian
sional Tolerance Analysis Using a Solid Modeler," ADCATS Productivity Organization
Report No. 8%3, Brigham Young University
Taguchi, G., E. Elsayed and T. Hsiang, (t989), Quality Engi-
Schade, R. (1980), "Probabilistic Models in Computer Auto-
neering in Production Systems, McGraw-Hill
mated Slider-Crank Function Generator Design," ASME Pa-
per No. 80-DET-48 Turner, J. (1990), "Relative Positioning of Parts in Assemblies
Schade, R. (1982), "A Probabilistic Model of Four-Bar Mecha- Using Mathematical Programming," ComputerAided De-
nisms," ASME Paper No. 82-DET-43 sign, v 22, n 7, pp. 393-400, September
Shapiro, S. and A. Gross (1981), Statistical Modeling Tech- Turner, J. and S. Srikanth (1990), "Constraint Representation
niques, Marcel Dekker and Reduction in Assembly Modeling and Analysis," Rens-
Sorensen, C., D. Nielsen, and K. Chase (1991), "Improved selear Design Research Center, Tech. Report No. 90027.
Methods for Tolerance Analysis of Mating Hole Patterns," Turner, J. and M. Wozny (1987), "Tolerances in Computer-
Proc. o f the Internat. Design Productivity Conf., Honolulu, Aided Geometric Design," The Visual Computer, n 3, pp.
Hawaii, February 214-226
Speckhart, F.H. (1972), "Calculation of Tolerance Based on a Turner, J. and M. Wozny (1990), "The M-Space Theory of Tol-
Minimum Cost Approach," J. of Engineering for Industry, erances," Advances in Design Automation--1990, v 1,
ASME, v 94, pp. 447-453, May ASME Publication No. DE--Vol 23-1, pp. 217-225
Spence, R. and R. Sion (1988), Tolerance Design of Electronic Turner, J., M. Wozny, and D. Hoh (1987), "Tolerance Analysis
Circuits, Addison-Wesley in a Solid Modeling Environment," Computers in Engineer-
Spotts, M.F. (1973), "Allocation of Tolerances to Minimize Cost ing--1987, ASME, v 2, pp. 169-175.
of Assembly," J. of Engineering for Industry, ASME, v 95, Wilde, D. and Prentice, E. (1975), "Minimum Exponential Cost
pp. 762-764, August Allocation of Sure-Fit Tolerances," J. of Engineering for In-
Spotts, M.F. (1975), "Probability Theory for Assemblies with dustry, ASME, v 97, pp. 1395-1398, November
Piece Part Errors Concentrated Near End of Tolerance Wirtz, A. (1988), "Vectorial Tolerancing," Tech. Paper CH
Limit," ASME Paper No. 75-DE-1 9470, New-Technikum Buchs, Gallen, Switzerland