Review of dimensioning and

tolerancing : representation
and processi ng
U Roy, C R Liu* and T C Woo ¢

cost. Statistical tolerances are sometimes used to

The paper surveys the current state of knowledge of minimize the product cost, with a slight deterioration
techniques for representing, manipulating and analysing in the product quality within an allowable limit. In this
dimensioning and tolerancing data in computer-aided statistical-tolerancing method, statistical distributions
design and manufacturing. The use of solid models and are specified for the nominal dimensions to allow more
variational geometry, and its implications for the flexibility into the manufacturing system.
successful integration of CAD and CAM, are discussed. Computer-aided analysis for D&T made its debut
The topics explored so far can be grouped into four in the 1980s, when the TOLTECH(TOLerance TECHnology) 1
categories: (a) the representation ot dimensioning and system was developed at the Production Engineering
tolerancing (D& T), (b) the synthesis and analysis of Laboratory NTH-SINTEFin Trondheim, Norway. This was
D& T, (() tolerance control, and (d) the implications of an early system designed to perform tolerance
D& T in CAM. The paper describes in detail the recent calculation. It mainly addresses the problem of assigning
work in each group, and concludes with speculation on tolerances so as to achieve minimal cost of manufacture.
a general framework k)r future research. Its analysis is limited to linear combinations of
dimensioning, toleran(ing, CAI~).'(-AM, ~olid models, variational dimensions, i.e. to dimensions considered as a 1D
~('om('try problem. The system has the capability of executing
the tolerance-distribution module, and it can calculate
Geometric dimensioning and tolerancing (D & T) is used the tolerance chains of an individual part.
to define the true functional limits of acceptable The solid-model-based investigation of D&T was
part geometry, including the description of essential initiated by Requicha 2 of the Production Automation
functional relationships. Until now, these relationships Project, University of Rochester, USA. He started the
have been expressed through standardized symbols. investigation by relating several representational issues
The D&T method has been drawing-based, and of D&T in a CSG-based solid modeler, PADL-1. He
has needed human intervention for its calculation, built up the object procedurally, incorporating the
representation and interpretation. It has largely dimensional values into the definition of the object
depended on the designer's experience and on itself, and he produced the 'dimensioned drawing' from
thumb-rule-based heuristics that may be very tedious
and error-prone. With advances in CAD/CAM tech- Geometric
niques, efforts have been made to automate these D &T 3 x 40.250 -+ 0.005 /tolerance
procedures.
Two types of tolerances, conventional and geo- i cl ooo
metrical, are usually represented in product dimension-
ing. Conventional tolerances specify upper and lower
limits for dimensions, while geometric tolerances relate
to the category of tolerances used to control the form,
profile, orientation, location and runout of the product
(see Figure 1 ). Tighter tolerances ensure better product lJ_Io.oolz
i I
quality, but they increase the manufacturing cost to a
considerable extent. Therefore, a tradeoff is needed
Icl
between the product quality and its manufacturing

Department of Mechanical and Aerospace Engineering, Syracuse

University, Syracuse, NY 13244, USA ..... ['~ I j~/-Conventional
* School of IndustrialEngineering,PurdueUniversity,West Lafayette. 3.250 4- 0.010 ~ ] tolerance
Indiana 47907, USA
t Department of Industrial and Operations Engineering, University
of Michigan, Ann Arbor, Michigan 48109-2117. USA Figure 1. Conventional and geometrical tolerances in
Paper received: 2 November 1987. Revised: 6 April 1990 engineering drawing

466 0010 4485:91070466 18 :(' 1991 Butterworth-Heinemann Ltd computer-aided design

this information. Following this work, several research
workers explored this field of D &T, and studied several --t
aspects of its implications for the successful integration
of CAD and CAM. Major achievements in the field can
be grouped into four categories: (a) the representation
of dimensioning and tolerancing, (b) the synthesis and
[_-Z
analysis of D &T, (c) tolerance control (how to control
the tolerance at different phases of manufacturing),
and (d) the implications of D & T in downstream
computer-aided manufacturing activities.

DIMENSIONS AND TOLERANCES

a
While the ANSI Y14.5M standard on dimensioning and
tolerancing 3 is not as expressive as a university
textbook, it does contain definitions. A dimension is
said to be 'a numerical value.., to define the size or
geometric characteristic, or both, of a part or part
feature'. To accommodate variations in a dimension,
tolerance is defined as 'the total amount by which a
specific dimension is permitted to vary'.
It helps intuition if a dimension is thought of as a
I
vector (with a reference, a direction and a magnitude),
rather than as merely a numerical value. The radius of
a hole, for example, is meaningful only with respect to
its center. Dimensioning for size can thus be established.
The center of the hole should then, in turn, be
established with respect to another reference - by b
'locating' via further dimensioning. As successive
references are made, tolerances accumulate.
Tolerances accumulate in a less-than-obvious manner,
for two reasons: the choice of reference, and the
direction dictated by the shape of the design. Consider
a block with k notches, of equal sizes and equally
spaced. Suppose that the size of a notch is dimensioned
with respect to an adjacent notch ('chain' dimension-
ing), as shown in Figure 2a. (For simplicity, again assume
the tolerances to be equal). Then, the variation for the
entire block obtained by successive referencing is k
times the tolerance of the notch. The left side of the C
block may be referred to as the 'datum'. Now, suppose Figure 2. Relationships between dimensions and
the datum is chosen to be in the middle of the block, tolerances; (a) datum on left side o[ block, (b) datum in
as shown in Figure 2b. Clearly, the accumulation is only middle o[ block, (c) nonlinear accumulation
k/2 times the tolerance. This illustrates that dimensioning
and tolerancing are coupled: the accumulation of
tolerances is a function of a given dimensioning designer has two choices: reduce some of the
scheme, and, conversely, to avoid undesirable tolerances, or redimension the design. Notice the
accumulation, a design may have to be dimensioned hidden combinatorial complexity in either choice.
appropriately. To compound the problem, tolerances Suppose that there are n dimensions. The designer may
do not always accumulate linearly (by adding). The choose k dimensions out of n. The designer may reduce
effect of bending the block is shown in Figure 2c. the tolerances of the k chosen dimensions, or chain
Trigonometric functions are involved, and hence the them differently. The iteration continues. The question
accumulation is nonlinear. 'are the tolerances appropriate for the dimensions?' is
To be sure, commercial software packages 4 exist not asked, and nor is the more fundamental question
for tolerance analysis. However, the iterative nature of 'what ensures the convergence of this iteration of
dimensioning and tolerancing should not be under- dimensioning and tolerancing?'.
estimated. That which occurs in dimensioning and
tolerancing, but is often avoided, is as follows. The
designer assigns the dimensions. Tolerances are looked Concurrency in design
up from a table if the dimension is standard. The needs for tolerance analysis, and, more to the
Otherwise, the assignments are based on 'experience'. point, tolerance synthesis, are not limited to the design
Accumulation (or 'stack-up') analysis is then performed. phase in the life cycle of a product. In design,
If the accumulation exceeds a certain specification, the functionality is of concern. Thus, ideally, tolerances

volume 23 number 7 september 1991 467

 should be as close t o zero as is possible. However, is a tolerance. Borrowing lrom elementary physics,
manufacturing places constraints on the ideal. In fact, dimension may be associated with the notion ()f
manufacturing cost is inversely proportional to tolerance position, and tolerance with velocity. This then suggests
(as comn]on sense dictates: the tighter the tolerance, a differential paradigm for representing tolerances.
the higher tile cost). This dichotomy (of tight tolerance Indeed, in his pioneering work, Hillyard ~ treated a
for design, and loose tolerance for manufacturing) polyhedron as a mechanism with many degrees of
represents the classic conflict of interest between these freedom. Tolerance, or variations of the geometry,
two phases, which perhaps partly explains the extra* could be viewed as the result of 'damping' the
ordinary amount of leadtime needed when a new mechanism. As the length of an edge can be expressed
design is put to manufacture. At the assembly stage, as the distance between its two vertices, dimensions
the picture is reversed again. Tight tolerances ensure D can be expressed as functions F of the geometry G :
interchangeability, without which parts would have to
D = E(G) (1)
be sorted by tolerance and stored separately. (Two
mating parts, each with three grades of tolerances, The differential form of Equation 1 involves the
would require six part numbers and storage bins). While Jacobian ]:
assembly and design have similar tolerance require-
A D = JAG (2)
ments, these two phases cannot be coupled without
the manufacturing phase - the phase that transforms In other words, tolerances can be represented as (small)
a logical description of a product into a physical form, changes AD in the dimensions. To display the effect,
as in changes AG in the geometry can be computed through
the inverse of J.
life cycle: design ~ manufacturing ~ assembly Statistically, a dimension is never exactly at its
tolerance: tight loose tight nominal value. Treated as a random variable x, the
One way in which to break the apparent deadlock is nominal value of a dimension may be associated with
to view the conflicting requirements for tolerances as the mean, and the tolerance with the variance.
constraints in optimization. The very act of considering Borrowing from statistics, it can be seen that the mean
constraints from different phases simultaneously alters and the variance are related by a moment-generating
the flow of the life cycle. By taking the traditionally function. In particular, the mean is the first moment
'downstream' considerations as constraints at the and the variance is the second moment.
design stage, it effectively achieves what is understood
as concurrent design. In the fourth section of this [" xf(x) dx (3)
paper, various techniques for tolerance analysis (using J
simulation), and tolerance synthesis (using optimization),
are presented. 0"2 = f (x --/~)2f(x) dx (4)
To simulate or to optimize, there must be
mathematical models and computer representations.
While tolerances may be treated as a field in a record, where f(x)is the density function. This integral paradigm
for example, such a simplistic view only perpetuates suggests that dimensions and tolerances are related in
the manual method of drafting: tolerances are a way that is opposite to that suggested in Equations
attributes to dimensions 3. It should be clear from 1 and 2.
research in solid modeling that a dimensioned drawing Leaving aside the question of the meaning of other
is but one way of rendering an internal representation moments (or higher-order derivatives), a third paradigm
of an object. It is equally clear from solid modeling is encountered, the constructive paradigm. While his
that dimensions do not have to be stored explicitly ground-breaking work ~ on tolerance representation
either as attributes or relationships in a data structure. was differential in paradigm, Requicha later proposed
While the internal representation of, and computation the notion of 'offset surfaces '7 constructed from the
for, dimensions are well understood, the same cannot nominal form (or 'true position', in ANSi terminology).
be said for tolerances. The volume of uncertainty enclosed by the two offset
surfaces (one for the maximum tolerance limit, and the
other for the minimum limit) forms a 'tolerance zone'.
Determinism versus uncertainty
It may be noted from the preceding overview that,
Figure 2 established the observation that tolerancing in the differential paradigm, tolerance is represented as
is a function of dimensioning, and vice versa. The a single instance of a variation in the geometry. In the
section immediately above established the practical integral paradigm, on the other hand, all the statistical
need for considering the different views of tolerances characteristics of nominal dimension and tolerance are
concurrently at the design stage. This section establishes brought to bear. If it is assumed that the distribution
the need for the representation and interpretation of characterizing the random variables is not central to
tolerances by the examination of some foundation the issue of representation, then the continuum (of
paradigms. distribution) may be reduced to two limits (for the
Intuitively, dimensions are geometries in the large, maximum and minimum tolerances). This is the essence
and tolerances are geometries in the small. For example, of the constructive paradigm.
a radius is a dimension, and circularity (a measure of While compactness may be a desirable feature for
the variation of the radius at different instances) representation, one should be equally concerned with

468 computer-aided design

the accuracy of computation. To be sure, representing between the extracted dimensions are not explicitly
a tolerance by its worst case (as two tolerance limits) given, proper connections need to be found so that a
discretizes the formulation of synthesis and analysis as dimension chain may be formed), fourth, verification
optimization or simulation problems. At the same time, of the consistency of the obtained dimensions (the
one should be cognisant of the fact that the result dimensions should be fully described with respect to
from the analysis (or synthesis) would be unnecessarily the given geometry), and fifth, graphical visualization
pessimistic. With these caveats, the discussion moves (or representation) of the output. In an attempt to solve
on to the third section of the paper, on representation, the problem, knowledge engineering has also been used
and the fourth section, on synthesis and analysis. The to acquire knowledge of dimensioning from standard
third section discusses the range of research work on drafting rules.
the recognition of implicit dimensioning information Yuen, Tan and Yu 9 have also presented a general
from a solid-model-based object database, and several scheme for the automatic dimensioning of objects from
procedures for attaching the tolerance information to their boundary representations, and have reported its
the solid model. The fourth section reviews the implementation on the CSG-based modeler PADL2. In
currently available technical tools for CAD-based this scheme, the interpretation of the semantics of
tolerance synthesis and analysis. conventional dimensioning practices has been viewed
as that of 'metric' relationships between geometric
REPRESENTATION OF D & T entities. Dimensions are defined by distances and
angular relationships between a pair of entities (e.g.
The information completeness of solid modelers makes points, lines or surfaces). Simple algorithms are used
it possible to have a complete D &T representation that to extract necessary information regarding straight-line
is interpretable by the computer. Several solid modelers endpoints in pair, endpoints in arc etc. from PADL2's
are now available, and many of them have some way boundary representation, and the linear and angular
of representing dimensions and tolerances. dimensions are then derived from the 'metric
information'. To represent adequate dimensioning for
Dimensioning the whole object, a 'dimension tree' is constructed so
that all the boundary surfaces of the object are present
In a traditional CAD system, dimensions are specified in the tree. Under- or overdimensioning is thus easily
manually, and this is very time-consuming work. As a detectable from the dimensioning tree. It is reported
geometric model already provides all the necessary that the current implementation does not fully agree
geometric data of an object, it is required to process with engineering-drawing standards, and that the
the information automatically to generate dimensions automatically generated dimensioning link may require
on the engineering drawing. Minagawa, Okino and manual modification.
Kakazu ~ report some success in the development of a Another piece of interesting research work on the
fully automatic dimensioning system based on a treatment of dimensions in 'product'-model construc-
CSG-based solid modeler, TIPS-1and AUTDIM.The AUTDIM tion (i.e. of a computer-understandable model that has
system structure is shown in Figure 3. The system works information that is semantically equivalent to the
in five steps: first, recognition of the geometric pattern representation on drawings) has been discussed by
from the object (CSG-model) data structure, second, Suzuki 1°. He introduces a dimension-description frame-
extraction of the dimensions of each feature (depending work for solid models - a dimension model based on
on the types of primitives), third, the connection of the solid models. This model has a shape-description
extracted dimensions to each other (as the connections function that uses dimensions as parameters. It regards
dimensions as surfaces' constraints, and they are
described in WFF (well formed formula in 1st-order
Primitive predicate) form. For example, two parallel surfaces A
and B separated by the distance L are described
Attribute by the predicate DISTANCE (A B L). By dimension
data modification, designers are able to modify the geometry
I I I of the solid object. The model helps to represent and

Preproce~
[ Shape-descriptiondata I
[ View
to manage the information about a product throughout
the design and manufacturing activities.

Tolerancing

~
Drawing
processor~
<1°utu I
IGeometry database]

.....
--- ~Dipr~een;ioOrning
Computer-based design and manufacturing processes
need their information to be logically 'unambiguous'
and free from redundancy. D &T representation usually
Anoth!r view _.~ Y contains a great deal of implicit information; this may
be 'obvious' to the 'intelligent' and 'experienced'
production engineer, but is not good for computer
understanding. The information content must not only
be explicit, but also unique for design, production and
Figure 3. AUTDIMsystem structure assembly groups. The situation has led researchers to

volume 23 number 7 september 1991 469

 consider and to devise a proper 'tolerancing theory' representation, rhe logical structure ot the VGraph is
that is acceptable within the CAD/CAM context. shown in Figure 5. The lowest nodes of the graph
Requicha 7'11 developed a theory based on the structure are NFace (nominal-face) nodes that are tile
'variational-class' concept. Variational classes are nominal faces of the object. VFace nodes represent
families of objects that are similar to a nominal object, subsets of NFaces, and tt]ey point to tile laces on the
are interchangeable in assembly, and are functionally object's boundary, while VEdge nodes represent tile
equivalent. By his definition, an object is considered to intersection of the two associated VFaces, and each
be in tolerance if its features' boundaries lie within the VEdge node points to two VFaces. SFeats and CFeats
specified range of the 'tolerance zone' (see Figure 4). nodes define groups of VFaces and VEdges, respectively,
The tolerance zone is again defined over a domain of to form 'features' to which attribute lists, denoted by
feasible region constructed by 'offsetting' (i.e. expanding AttList in Figure 5, are attached. The other kind of node
or contracting for plus/minus tolerances) the part's used in the VGraph is the DatSys node. These define
nominal boundaries. The tolerance information is datum systems, and contain an order set of datums.
specified as a set of geometric attributes of the surface These datums are usually represented by pointing to
features (2D subsets) of an object boundary, and it SFeats and CFeats nodes. The system does not use the
dictates the offsetting criteria for the boundary surfaces. B-rep for direct access to these NFaces. Besides, an
A formal theory for the 'offsetting' operations has been indexing scheme for the faces of each instance of the
discussed at length by Rossignac ~z~ in an attempt to primitive solids is consistently maintained in the system
combine them with other Boolean and 'rigid-motion' for the proper identification of any desired NFace to
operations in an extended CSG scheme. However, this which tolerance attributes are to be attached. This
kind of tolerance representation in terms of 'tolerance VGraph subsystem is still at the experimental stage,
zones' proposed by Requicha differs in some respects and it has a limited ability to describe design tolerances.
from the ISO systemU; moreover, as the handling of This representation scheme has not yet been checked
dimensions and tolerances in the general case requires for its suitability for tolerance analysis.
the ability to access the bounded entities of objects, Jayaraman and Srinivasan 1~17 have examined the
the CSG-based tolerance theory of Requicha raises issues of representing the geometric tolerances in solid
some manipulation problems during implementation. models from the perspective of functional requirements
Further research on an effective tolerance theory is related to the geometry of mechanical parts. Their
needed. research is mainly concerned with the positioning of
Using the theory of tolerancing as discussed above, parts with respect to each other in an assembly, and
Requicha and Chan is have implemented the represent- with maintaining material bulk in critical portions
ation of tolerances (which treats tolerances as properties of parts. They develop specific 'virtual boundary
or attributes of an object's features) and other requirements' (VBRs) to reflect the required functional
variational information in a CSG-based modeler, conditions of the assembly, and then discuss the
PADL-2. The variational information is associated with theoretical basis of the interpretation of those virtual
the solid model by means of a graph, called a VGraph, boundary requirements with the help of the theory of
or variational graph. This VGraph graph structure is solid-model-based offsetting, as proposed by Rossignac
linked with the nominal representation of PADL-2 via and Requicha.
NFaces (nominal faces of an object) that are associated There are also two other significant projects on the
with the faces of primitives in the object's CSG representational issue: one is funded by CAM ITM, and
the other by the United States Air Force under the IC/,M
project 1'~. Both pieces of research are based on a
Tolerance zone
B-rep-type data structure of the solid model.
(:AM I'S work m proposes the building of a separate
D&T modeler coupled with the geometric modeler.
This D&T modeler is used to create, modify and
interrogate the D &T model and its relationships to the
geometric modeler. The D&T modeler, the Evaluated
Dimensions and Tolerances (EDT) model, has been
designed to represent 'features of the classes' that are
dimensioned and toleranced in accordance with the
Axsl Y14.5M standard~. The model is applicable only
Positional tolerance for location and size tolerances, and it is limited to
zone geometric entities such as planar faces, cylindrical faces,
conical faces and spherical faces. As the model uses
the boundary representation, the designer needs to
create the B-rep of the part first, before he..she can
enter the D&T information. 'Templates' (a set of
procedures) are associated with the EDT model to
provide a method of correctly modeling the EDT.
Templates are used to check feature validity, to
establish the datum reference frame (DRF), and to
Figure 4. Tolerance zones compute the respective dimensional values.

470 computer-aided design

Datum systems DatSys
DatSys

FaceOp

EdgeOp
Features
and
attributes
FaceOp FaceOp

VEdge VEdge

f
VFace VFace I VF;ce I
Solid
ClasOp

Nominal
geometry
NFace NFace ,,. I " " °

Figure 5. General structure of VGraph

The designer designs the slot as a feature by selecting

the relevant faces and grouping them together. The EL
DRF . node connects the B-rep faces to the 'slot' data
structure (D/T node) within the dimension and
tolerance model. The D/T node is then associated with
~/T I~ a datum reference frame (the global datum reference
(slot)Z frame in Figure 6) that may already exist, or may be
created for this feature. Dimension and tolerance data
is then specified for the slot by the creation of separate

1
ED nodes, with one for each dimension, such as position
and size, as shown in Figure 6. A separate template
A is required for each type of dimension attached. These
ED nodes are then linked with the D/T node. CAMq is
I- currently in the process of implementing this work for
further manufacturing applications.
It is evident that Requicha's approach, as described
\ above, is similar in principle to the EDT model. The
difference lies in the way that the designer represents
Figure 6. EDT model for slot tolerancing the tolerancing information in the respective schemes.
In the EDT model, the complete B-rep of the object
should be available to the designer before he/she starts
The overall schema for the EDT model consists of tolerancing, whereas Requicha's VGraph is CSG-based,
four types of nodes: (a) a D/T (dimension and and it allows the designer to incorporate the tolerancing
tolerance) node, (b) an EL (entity-linking) node, (c) a information into the process of defining the CSG tree.
DRF (datum-reference-frame) node, and (d) an ED The principal disadvantage of Requicha's system is that
(evaluated-data) node. An EDT model for slot tolerancing all nonprimitive faces derived from the same primitive
is shown in Figure 6. face receive the same variations. For example, different

volume 23 number 7 september 1991 471

 size tolerances could not be assigned to the lengths Gossard, Zuffante and Sakurai ~'~' have reported a
AB and CD of the body in Figure 6. For those portions similar kind of feature-based design system that uses
to be toleranced, a separate modeling sequence is a B-rep solid modeler. They have also used a
required, i.e. instead of a 'minus' Boolean operation representational scheme that combines CSG and
being performed, two 'union' Boolean operations are boundary representation. They represent feature and
required to build the 'slot' in the cube. surface information in a CSC-based graph structure
Recently, Elgabry 2° addressed a framework of the called the 'object graph', which uses 'relative-
tolerance representation for better analysis. His frame- position operators' (RPO) nodes to represent explicitly
work is based on the CSG model, and the creation of dimensions between features and elements of features.
a separate 'tolerance-model data structure', in addition These RPO nodes are used in a way that is similar to
to the CSG tree, is proposed. This data structure for the use of other regularized set-operator nodes (i.e.
tolerancing a part is also somewhat similar to that of union, difference and intersection nodes). The object
Johnson's EDT model TM. graph of the cube with a slot shown in Figure 6 is
In the GEOTOL system, Turner 2~ has attempted to shown in Figure 8. Each RPO node defines specified
associate the tolerancing information with the evaluated dimension and tolerance constraints between two
boundary representation of the part. All variations are feature surfaces to determine the position of an
applied to the part faces (currently limited to planar 'operand' face with respect to a 'reference' face. For
and cylindrical faces only) of the nominal model. For example, in Figure 8, the face a of the slot feature is
instance, if the designer specifies a size tolerance for constrained by the distance A from the face 5s of the
the distance between two parallel faces of the part cube by an RPO operator. The system has been
shown in Figure 6, the tolerance is simply attached to implemented on a polyhedral solid model, and it is
the two face nodes in the part's boundary represent- limited to the conventional tolerance representation
ation. A prototype representational module has been only.
built to provide IBM's Geometric Design Processor IGES(Initial Graphics Exchange Specification) research
(GDP) solid-modeling system with a 'generalized CSG' and development activities for the Product Data
(GCSG) architecture. Exchange Specification (PDES)fOCUS on data modeling
It should be noted that D & T representation is and representation, as well as the exchange of complete
feature-based, and it needs the proper identification of product-definition data. Part of the ~CES,PDES work
'geometric features'. Two kinds of features are involved: includes a tolerance model that is similar, but not
(a) lower-level features (such as points, lines, arcs, identical, to the PDDI model r . It addresses the
splines and surfaces), and (b) higher-level features, representation of tolerances for B-rep-based models.
which are combinations of the lower-level features (or An early system, APPAS28, exhibits a limited capability
the combination of other higher-level features, such as of incorporating tolerances and other surface inform-
holes, slots, pockets, countersinks or complex features), ation (which it takes from COFORMas the input in the
and which maintain certain relationships among form of a list of attributes) to derive the process
themselves. Lower-level features are basic topological sequence. SrOPP 29 also uses the same method of
entities, and, therefore, they are well defined and attaching the variational information (e.g. roundness
unique, whereas higher-level features are design- error, positional accuracy, straightness, size tolerance,
specific, and the choice for their selection depends on surface finish) with attributes for the 'hole specification'.
the function and the context of application. For D& T In C~MS/PRO~°, information about dimension, tolerance,
implementation, this features information must be surface finish and the approachability of tools is
extracted or recreated from the solid model. Roy and provided with each of the machining-shape elements.
Liu 22 showed the necessity of having a hybrid CJMS/PROinherits all this information from OMS/DEC. The
CSG/B-rep data structure for the tolerance represent- attempts to include tolerancing information, as described
ation so that the advantages of both CSG and B-rep
models can be exploited. The tolerance module is
. l A Io.oo21
attached at the top of this hybrid structure. The user
interacts with the solid model at each hierarchical level
of object construction for associating tolerance and
other technological information (such as material data,
f 2.000 + 0.005
surface roughness etc.), rather than waiting until the

i ooo!o
oo
entire part geometry has been defined. A process of
establishing tolerance information is shown in Figure 7B.
A nominal-part drawing including tolerance is shown
F
in Figure 7A, and Figure 7B shows how the specified
1.000 -+ 0.005
tolerance information has been attached to the hybrid l c Io.oos
CSG/B-rep module via a 'reference face list' (RFL). The
reference face list acts as the bridge for combining the I
I I
CSG and B-rep data structures. All kinds of conventional I I
and geometrical tolerancing representation are possible I I

in this system. This work on tolerance representation ~ t

has been implemented 23'24 on the Sun workstation

based on the polyhedral B-rep TWIN solid model 2s. Figure 7A. Nominal part drawing including tolerance

472 computer-aided design

 I" I c I o.oo ]

#6
,/~#4 "t #5

#1

#7
I

B-rep
\

"

RFL

Pointing to B-rep Pointing to RFL

S
,1/ -"
#5

#6
B-rep
B-rep

Center planes
CSG

Referenceface # 1
Referenceface #2
Referenceface #3
Pointing to RFL /
Referenceface #4
Referenceface #5
Referenceface #6
Referenceface #7
Referenceface Referenceface #8
list
Referenceface #9
Referenceface # 10
Referenceface #11
Referenceface #12
y
Referenceface #13 /

Figure 7B. Tolerance representation in CAD database

[ # 1, # 2.... : face numbers in CSG. 1, 2.... : nodes of B-rep graph; they also correspond to the faces of the object. Faces in CSG have
pointers to the g-rep, and the face nodes of the B-rep graph have pointers to the reference face list. RFL: reference face list.]

here, are mainly restricted to manufacturing applications and material information). The integral part model can
only; no effort is made to encompass the broader view be constructed by combining the technological and
of an informationally complete solid model. the material information with the information obtained
lwata 31 develops the conception of an integrated from a solid modeler. The system provides its own
product model, where the part model includes all of database management system, in which the product
the part information (including geometric, technological and part models are stored in the database. The model

volume 23 number 7 september 1991 473

 AB
Geometric t
reasoning
programs

Roo, br.och "-

t f o r m f e . t u re
GEOMAP411
Relationship solid modelling system

Distance (A, B, X)

Attribute//~
nominal = 10 Geometrical
calculations
COMET/DB Display functions
S2
Figure 9. General framework of ( OMEl,l)~ model
,,4 S5 -"' ~ :,4

@ /--30R

Figure 8. Object graph with RPO nodes

try: set-operator node, u: set-operator node, R: relative-position @ o
operator node. I

is capable of handling the model construction, the

model modification, and the retrieval of the required
data. Adding or changing the technological information
_"
_. 7 o.o,F__*.°
(including nongeometric information)is performed in
the model-construction stage. When the construction
is finished, the system checks the possibilities of
interference between the parts of the product. It is not
known whether the system also includes tolerance
analysis.
Other researchers 1°'{2 34 from the University of
Tokyo, Japan, also address a similar kind of object-
modeling scheme based on the solid modeler GEOMAP-III.
By the use of a database management system (DBMS),
various geometric and nongeometric data (such as
physical, technological or management data) can be
added, manipulated and retrieved by the definition of b
the appropriate types of attributes. Dimensions and Figure "10. Typical mechanical part and its dimension
tolerances are both considered to be the geometric hierarchy; ( a) example mechanical part, ( b ) dimensioning
constraints, and 1st-order predicate logic is used to hierarchy
represent them. The D&T model has been kept
separate from the solid model, and it is represented as
a kind of upper-level model on the top of the solid called a 'dimensioning hierarchy' (see Figure 10), and
model. A general framework of the model is shown in both conventional and geometrical tolerances are
Figure 9. A PROLOG-like system has been used for described. The system thus needs either an exhaustive
manipulation and to reason about the relationships and constraint describing input procedures, or a powerful
respective tolerances. It has been implemented as algorithmic approach to capture the essential geometric-
COMET/DB (COnceptual Modeling Experimental Tools/ constraint patterns between different geometric entities
Data Base) in a LISPenvironment. Tolerance evaluation from a system such as an engineering drawing. The
and analysis are also supported to a limited extent. input to Yu's system is the general information
Yu :~s'36 has proposed a variational solid-modeling obtainable from the conventional drawings. The user
system, v-SOUP,to incorporate industrial D & T practices specifies this information with the help of a design
into the boundary representation of an object, in a module, ED-DESIGN37. An engineering-drawing under-
way that is compatible with engineering drawing. In a standing system, EDUS, is developed to construct the
way that is similar to that used in the previously V SOLID from ED-DESIGN. [t iS possible to perform the
described product model, the D &T is also treated as selection of the process sequence for dimensional and
the geometric constraints in V-SOLID. These constraint geometric accuracy in a way that is based on this
relationships are explicitly encoded in a graph structure hierarchical definition of the dimensioning scheme.

474 computer-aided design

 Expert systems such as GAR138use the D&T as input determination of the economical solution of these
as they appear on the drawing. They are not connected equations. In this selection procedure, the allowable
with any solid modeler such as PAUL-2, ROMULUS,or variations in the design functions are first determined
GMSOLID. from the desired functional requirements, and then the
Another interesting approach 39to the representation independent dimension tolerances are established. This
of technological information is to define a grammar of whole developmental process of tolerance determin-
the part, and to present its structure via a technological ation is termed the 'tolerance-synthesis', or 'tolerance-
tree. The structure could be derived from the drawing. allocation', problem.
The implementation details of this concept are not It should be recognized that, in general, the
discussed. Milacic has published another paper 4° quantification of functional requirements, and the
on SAPT (the system for manufacturing process formulation of the required design functions, are often
planning), where he talks about building a knowledge very complex procedures. A straightforward approach
base containing information about (a) geometrical is to provide all the relevant and critical dimensions
characteristics: shape, dimension etc., (b) functional with their nominal values and upper and lower limits.
characteristics: geometric relationships between the This is called 'worst-case' analysis, and it is a very
forms of parts: parallelism, perpendicularity, coaxiality conservative method of specifying tolerances, as the
etc., and (c) other technological characteristics. The design-function variables are very unlikely to combine
representation technique for this knowledge base is in this fashion in any particular product. For
similar to that of GARI,and is based on the production- more realistic consideration, statistical tolerances are
system concept. determined to provide statistical distributions for the
A commercial system, PRO/ENGINEER (a product of design functions.
the Parametric Technology Corporation) 4~, supports Michael and Siddal142'43 propose a 'vector-space
tolerance representation in its feature-based, parametric formulation' for tolerance synthesis. In this formulation,
solid modeler. This solid model uses boundary a vector space is defined with its coordinates
representation for object modeling. It stores all corresponding to the independent dimensions.
functional, topological, geometrical and other feature 'Tolerance' and 'design' regions are then constructed
relationships in its database. It allows the user to in this vector space from identified tolerance variables
establish tolerances for each dimension at any time, (which define the dimensional tolerances) and design
and it also helps in analysing tolerance stackups. Some variables (which define the design functions), with their
other existing systems (such as those of Cognition, limits given. The objective is to find the upper and
Intergraph, Computervision and CIS) also provide the lower tolerance limits for the dimensions that minimize
facility of tolerance representation in their 2D drafting the overall manufacturing cost (which is also a function
modules. of the tolerance limits), subject to the constraint that
the 'tolerance' region be completely contained within
the 'design' region. They also extend the idea to
statistical tolerancing. These approaches are further
elaborated and incorporated in the GEOTOLprototype
TOLERANCE SYNTHESIS AND ANALYSIS CAD system by Turner 21. GEOTOLuses three methods
This section presents the current status of knowledge for tolerance synthesis and analysis: (a) linear
and the use of mathematical techniques of tolerance programming, (b) the Monte Carlo method, and (c) the
synthesis and analysis to highlight the findings of major least-squares method. These methods support both
research work. The subject is studied from the worst-case and statistical tolerancing problems. As
deterministic and statistical points of view. GEOTOLdeals with linearized tolerance variables only,
those variables are required to be selected carefully.
Farmer 14 has developed an interactive system for
tolerance selection. The procedure is iterative, and a
Tolerance synthesis number of possible solutions need to be checked before
Design tolerance the best one is selected. The system consists of three
D&T selection in the very first design stage includes sets of software: the drafting program, the tolerance-
the proper transformation of the functional require- technology program and the dataset program. The
ments of design to suitable dimensions and tolerances. drafting program is used for specifying the design
It evolves in four steps: (a) identification (description information of several manufacturing, assembly and
and quantification) of the functional requirements, inspection requirements, and for preparing drawings.
(b) the identification of datum-reference-frame (DRF) The dataset program is a collection of local standards,
features, and relationships between features that related to size and angular tolerances, different
influence functional and assembly requirements, (c) the limits and fits conditions, machining/manufacturing
development of the functional equations (with the process accuracy etc., that are required in tolerance
functional requirements as dependent variables, and analysis. The tolerance-technology module performs
the design sizes and tolerances of the dimensions the tolerance analyses of designs with the information
affecting the functional requirements as the independent available from the drafting and dataset programs.
variables (from here on, those independent variables Parkinson 44 proposes a statistical tolerance-synthesis
are termed the product dimensions, whereas the method, where the distributions of the dimensions are
dependent variable is the design function)), and (d) the derived from the sample data drawn from the intended

volume 23 number 7 september 1991 475

 manufacturing processes. He develops a mathematical specified tolerance range for the part dimension)about
programming formulation, with the product cost as the the midpoint of the tolerance range, which is the
objective function, and the standard deviations of the probable location of the mean of the component
design variables, i.e. dimensions, as the decision dimension. On the estimation of the range of the mean
variables for the problem. This product cost has been shift for each component, the resulting assembly
modeled as a function of tolerance, and it is considered tolerance is calculated by the analysis model, which
to comprise two components: (a) the product includes both mean shift or bias, and part tolerance or
processing cost, and (b) the cost due to rejected variance.
products. ]he tighter the tolerances, the higher is the
manufacturing cost, with a lower number of rejected
products, and vice versa. Manufacturing processes thus Manufacturing tolerance
have a direct bearing on this optimization assignment. D &T, thus selected and specified in the design stage,
No constraints on tolerances are considered in is further revised according to a detailed process plan
this method. The particular product-cost model, as to obtain the manufactured dimensions and tolerances
the objective function, controls the occurrence of of a mechanical part. The calculated manufacturing
any infeasible optimum point (i.e. infeasible small D & T are not only functions of design D&T, but also
tolerances). If the infeasibility still occurs, it indicates depend on the capabilities of the manufacturing
an error in the product-cost modeling, and remodeling processes and of the manufacturing equipment
is necessary. (i.e. machining errors, jig- and fixture-setting errors,
Another concept of using stochastic procedures for workpiece-positioning errors, errors due to cutting-tool
tolerance synthesis has been introduced by Lee and wear etc.). This suggests that a suitable transformation
Woo4S'4~L They formulate the concept as a combinatorial from design D & T to manufacturing D & T is required.
optimization problem by treating the manufacturing Hoffmann 49 has tried to develop a 2D tolerance-
cost as the objective function, and stackup conditions allocation model for a part, considering the operations
in assembly as the constraints. As the constraints are and setting-up inaccuracies involved in a given process
nonlinear, they have developed a technique for plan. His model can handle only those problems that
approximating the volume under an n-dimensional fit into D models (i.e. parts made of D elements, namely
probability-density function bounded by nonlinear 1D points and 2D edges in Euclidean space). A system
functions. In Reference 46, a branch-and-bound of linear inequalities is first developed from the given
algorithm was developed for handling the automatic part-tolerance specification. A set of constraints is then
computation of discrete tolerances, which is identified from actual operations and setup inaccuracies
combinatorially complex. for the positions of several of the D elements inw)lved
Dong and Soom 47 have developed a unidirectional, (that are to be machined according to a sequence
tolerance-chain analysis system for axisymmetric, plan), and this is added to the previous system of
rotational parts represented in a 2D CAD database. inequalities. The numerical solution of the total set of
Their work, in the AU]OANA (AUtomatic TOlerance inequalities defines the resultant tolerances for the part.
ANAlysis) program, includes the automatic retrieval of The main difficulty with this model lies in solving this
the required toterancing information from the CAD set of inequalities.
database, and the analysis of relationships between the Bourdet ~°'~1 develops another model for the
tolerances in a specific design, including a proper optimization of unidirectional tolerance transfer from
tolerance-distribution scheme between all the related design to manufacture. The optimization procedure
dimensions. tries to find the optimal manufacturing dimensions from
In the case of assembly tolerance accumulation, both the given design dimensions, taking advantage of the
'worst-case' and statistical models show distinct maximum range of specified tolerance fields within the
limitations when applied to tolerance allocation. The capabilities of the available machining. In other words,
worst-case model results in component tolerances that it calculates the manufacturing dimensions to be as
are tight and expensive to produce. Currently available close as possible to the limiting design tolerances.
statistical models, on the other hand, may allow looser Daniel s° has implemented Bourdet's model to
tolerances. This is because these models assume that demonstrate the computerized tolerancing. With the
manufacturing variations follow a normal distribution. input of workpiece data (the drawing of the part and
No skewness and bias are taken into consideration, its initial dimensions) and of process-plan data (the
although all manufacturing processes exhibit bias 48. sequence and kinds of operations to be performed, the
Bias may be caused by machine-tool error, setup error initial values of the setting of the dimensional tolerances
or any other machining errors. Chase and Greenwood 72 etc.), the program carries out an iterative procedure
thus propose the 'estimated mean shift' (EMS) model for the process-plan optimization of tolerances.
to deal with this expected bias for component Bourdet's model is limited to 1D toterancing only. For
tolerances in an assembly tolerance accumulation. As this reason, for a given workpiece, the program needs
very little information is available on component to be executed in three directions independently.
distribution types in the early design stages, this EMS [he AUTOANA 47 system, as mentioned above, also
model provides a useful tool for tolerance allocation transforms the design D &T to the manufacturing D &T
in several components of an assembly. In this method, by simply translating the design coordinate system to
the designer estimates the bias for each component the manufacturing coordinate system when the two
by defining a zone (expressed as a fraction of the coordinate origins do not coincide.

476 computer-aided design

Tolerance analysis
In tolerance analysis, the objective is to find the
variability of a design function (which may be i•"-•"x 3 C=40+c
any dimension that is dependent on independent
dimensions, i.e. the product dimensions, as described ~'//D=13
+ 4\\ ~ l
above) and its validity. As with tolerance synthesis, \ B=80+b
research is carried out to develop both statistical and
deterministic ('worst-case') tolerance-analysis tech- F=53°+f X\ +e
niques. The commonly used statistical technique is the
Monte Carlo simulation, whereas the 'variational- '1•~ A=140+a --,
geometry' method, advocated by Hillyard s'S2, is the
widely accepted deterministic technique. Figure 11. Dimensioned pentagon 53
With Monte Carlo simulation, the design-function
values are generated from a large sample of dimensions
with a specified statistical distribution. These design- derived from the above constraint equations, describes
function values are, in turn, used to calculate the the exact tolerances of the pentagon.
statistical parameters for the design-function distribution. Hillyard's system deals mainly with polyhedral
The main problem with this method is that it needs a objects, and considers an object as a pin-joined, elastic
large sample size to obtain a reasonable estimate of wireframe covered by elastic membranes. It uses a
the distribution function. dimension-tree to represent the dimensioning scheme,
A commercial package called VSAS 4 is available that and includes the construction of a 'rigidity matrix' as
uses the Monte Carlo method for tolerance analysis. a function of geometry and dimensions. The inversion
VSAS needs the assembly information of the product of this matrix is then used to define geometry variations
and the dimensioning scheme for each part as input. in terms of dimensional changes. It can determine
It can automatically perform the tolerance analysis for whether a component is under-, over- or exactly
gaps, and generate samples of dimensions and the defined by a given dimensioning scheme. Tolerances
corresponding samples of the design function. The are treated as small dimensional changes, so that
disadvantage of the system is that it cannot extract the geometry resulting from specified maximum
information from an existing CAD database; the dimensional deviations can be determined. This method
designer has to model all the details of an assembly treats the size, orientation and flatness tolerances only.
directly in terms of relationships between the parts. A It uses the B-rep-based BUILDsolid modeler.
more recent work by Turner 21 uses this Monte Carlo The work of Gossard 26, Lin ~4 and Light ss'56 extended
method in the solving of tolerance-analysis problems and refined the above variational-geometry approach
for several simple assemblies. His work, with CEOTOL, is from the standpoint of the user interface and
geometric-model-based, and the system derives all its computational efficiency. Light' s 'symbolic-dimension-
necessary input geometric relationships from the ing' system has the capability of modifying a part
geometric model. geometry by means of altering the numerical values
In the deterministic tolerance-analysis approach, of the explicitly defined dimensional constraints. Its
Hillyard and Braid s,s2 developed first the concept of algorithm selects the minimum set of equations and
variational geometry for analyzing the inconsistency in unknowns required for the solution of a given change
the specification of dimensions and tolerances in in dimensions. It reduces the computational time
computer-aided mechanical design. The variational- considerably. Minnichelli s~ proposes a further extension
geometry method is a dimension-driven, constraint- of the above technique for tolerance analysis. He has
based technique. The geometry of an object is dictated considered the design of an assembly, and has
by specified dimensions. The method regards dimensions concentrated on solving a composite rigidity matrix for
as constraints between vertex geometries. The user the assembled parts. This method is also B-rep-based.
creates a part topology and a set of dimensions from The variational-geometry technique, as discussed
which the exact geometry is derived. The part geometry above, is the constraint-based approach that leads to
can thus be modified by changing its suitable dimension systems of equations and inequalities. The dimension
constraints. To clarify the variational-geometry technique, tree, as created by the dimensioning scheme of the
an example of dimensioning a 2D pentagon is shown part, defines the set of equations relating the geometry
in Figure 11. As the pentagon consists of five vertices of the part to the dimensions. Tolerances are
with two degrees of freedom (DOF) for each vertex, represented as allowable ranges, or as statistical
2 x 5, or ten, constraint equations are needed to distributions of the explicitly defined dimensions. As
constrain the part properly. The relationships between the dimensions are nonlinear functions of the vertex
the dimensions of the part and its geometry (i.e. vertex coordinates, they ultimately yield a system of nonlinear
coordinates, in this case) constitute the required set of equations to be solved. The main drawback of
constraint equations. With a given, user-defined the variational-geometry technique is the inherent
dimension scheme, the solution of the equations yields computational difficulty of solving simultaneously this
the exact vertex geometry for the pentagon. The large set of nonlinear constraint equations. It requires
solution of another system of equations relating the an explicit definition of the independent constraints of
variation in positions (or displacements) of the vertices, the model variables. The technique is thus limited in

volume 23 number 7 september 1991 477

 terms ol the practical model size, tile complexity of Assembly
the part geometry, and the type of tolerance supported. f . \,,

Different kinds of geometric tolerances are diffi(ult to

represent as a set of equations.
Aldefeld's ~: work on variation in 2D geometries
is another attempt to capture the drawing-based
dimensioning concept ill a geometric model. Unlike
the variational-geometry technique, this work is based
on geometric reasoning, in which symbol manipulation
and interferencing methodology have been used in
processing the geometric knowledge. The system is
able to detect the consistency of the geometric model
(whether it is under- or overdimensioned or redundant),
and it supports the automatic generation of several
variants of a given geometry with some assigned SubassE
dimension values.
In addition to statistical and deterministic techniques,
some researchers have also attempted to perform
tolerance analysis on procedural representation of parts
in CAD. Elgabry 2° postulates an effective tolerance-
analysis scheme based on Requicha's 'tolerance-zone'
representational technique. For any datum reference
frame, a 'tolerance shell', as Elgabry calls it, can be
defined that contains all possible locations of the part
tolerances. Although all part tolerances are required to
form the tolerance shell, only some of them really Assembl ements
contribute to the development of the shell, depending
on the tolerance hierarchy. Options are provided to Part
slice out the tolerance shell in any prescribed direction
to reveal its crosssection. Different kinds of fit, including
interference and clearance, among the assembled parts
Part_d
of a product can readily be checked in any arbitrary
direction with the help of the shell analysis. The c(
proposed model is for a CSG solid-based tolerance Position : vector
representation and analysis. A working version of the
Figure 12. General structure of assembly
model has not yet been implemented. Representing
geometric tolerances as the tolerance zones, Fleming ~n
studied uncertainty in assemblies of rigid parts. He functional and spatial relationships between the
introduces the concept of building networks of constituent parts of a product. The assembly structure
tolerance zones and datums for the analysis of allows the user to define several 'associations' between
toleranced parts and their assemblies. In the network, the subassemblies and the components. The nodes,
each arc represents a relationship implied by the marked as M, A, I and G in Figure 12, correspond
tolerance specification or by contact (in the assembled to 'membership', 'aggregation', 'interaction' and
condition) between the parts. It is shown how all the 'generalization' associations, respectively, and these
geometric constraints can be converted to an algebraic associations express different functional relationships
form, so that they can be easily manipulated to between the components of a typical assembly. With
determine whether the parts of an assembly can be the information about parts' mating relationships from
guaranteed to fit together. the CAD data models of single parts, this assembly
It should be noted that tolerance analysis in a module largely facilitates an automatic 'tolerance-
complex, 3D assembly environment is a difficult stackup' analysis and other tolerance-related checking
problem. Minnichelli's effort s~, as mentioned above, procedures for product validity. Turner's work 21 with
proves that a constraint-based approach in this case (;EOTOL is worth mentioning here. Instead of building
is very computationally intensive, and is almost a separate assembly database, 'relative-positioning
impractical for complex cases. On the other hand, very operators' are used in GEOTO[ to place several parts
little work has been done in the arena of procedure- relative to each other. They have been designated as
driven solid models. The main problem in this case is relative-positioning nodes in GEOTOL'Sgeneralized CSG
that currently available CAD systems represent only schema (see Figure 13). In general, the action of a
part models, and they do not provide any global relative-positioning GCSG node is to position a target
database that (:an support the assembly-specific subtree of a model relative to a reference subtree.
information of the complete product. Roy and Liu ~'~ These subtrees may represent a part feature, part or
have studied the subject in detail, and have proposed subassembly. Turner has extended the CSG's inherent
a solid-model-based assembly-data module to establish capability of representing 'bounded volumes' from
a unified framework (see Figure 12) for representing the single-part representation to assembly description. This

478 computer-aided design

 cannot meet the tolerance requirements of a part
design, some finishing operations are recommended,
__~nybinaryoperatOt
t~ with a subsequent increase in product cost. For
this reason, research into machine-tool accuracy

/
LT~ ~
~2~sstG°nn
M ode
iondge~xFORM
li
ng enhancement has been proposed that aims to reduce
the number of processes required 63. Donmez 64 and
Ferriera 65 have reported some significant achievements
in reducing machine-tool errors of one order of
magnitude or better.
/ ,. ....
IMPLICATIONS OF D&T IN CAM
This section briefly discusses the implication of
tolerance use with respect to the overall design/
manufacturing process. Related development areas of
CAM have been identified where tolerance information
is found to have profound effects on decision-making
Figure 13. Use of relative-positioning operator with any processes.
binary operator In the field of automatic process planning, the
selection of machining activities/processes is greatly
influenced by the constraint of preferential relationships
only works in a smaller domain of two-parts-at-a-time owing to the desired machined-surface quality and
assembly conditions, and its expressability is also product tolerance specification 66. Based on the
limited. geometrical tolerancing information, and the reference
and datum feature information, the machined features
(i.e. surfaces, holes, slots, pockets etc. to be produced
TOLERANCE CONTROL
by a particular machining operation) are ordered for
Tolerance control is an important component of machining. A modular system CEFPOS (GEnerative
CAD/CAM integration. As the selection of proper Frame-based Planner for Orthomorphic Shapes)67 has
design D&T affects the functionality, as well as demonstrated the critical role of tolerancJng information
manufacturability, of a product, the selection of manu- in the development of a generative process plan. This
facturing processes, and their sequence of execution, D&T information also affects the choice of the
affects process tolerance stacking. A close monitoring workpiece orientations during several machining
of the inaccuracies developed at each of the manu- conditions, and it thus indirectly influences the selection
facturing phases is absolutely necessary to avoid undue and analysis of flexible fixture design 6~'69 for an
rejection of the part in inspection. automated machining task.
A computer-aided tolerance-control (CATC) D &T information is the single most important factor
system 6°'61 has been developed by the Ohio State in automated inspection. On the receipt of a finished
University, USA, research group. Their method of product (or for online process monitoring), it is the
analyzing and controlling tolerances is based on the inspection module that decides on the acceptability of
'tolerance-chart' technique, developed in the early the product/process according to the given tolerancing
1950s. The tolerance-chart technique is the simplest information. The necessary functions of an inspection
method of selecting the tolerance for the resulting system include (a) the identification of the part
dimension. Machined dimensions are combined in feature(s) that are to be inspected, (b) the extraction
closed chains of n dimensions, the resulting one being of the relevant tolerancing information from the CAD
the result of a combination of the other n - 1 database, (c) the interpretation of its meaning in terms
independent working dimensions. This method is of the relationship between the tolerances and the
limited to simple 1D cases; it is almost impractical for geometry of the part, and (d) the verification of the
2D and 3D tolerance analysis. The CATC program is finished product measurements with respect to the
iterative in nature, and uses computer graphics for tolerance specification. Hopp 7° describes an ongoing
information display. CAD-directed inspection project at the US Automated
The same procedure has also been adopted by Ji62 Manufacturing Research Facility (AMRF), which has
in operational-dimensions calculation. The primitive established a framework for a direct interpretation of
dimensions of each operation being known, the the design data from the CAD model in terms of
operational tolerance of each of them is calculated inspection actions and the subsequent generation of
from the set of blueprint dimensions and allowances, inspection procedures. The development of another
using the linear-programming technique. CAD-based vision system for the planning and
An important area of tolerance control is related to execution of inspection has been reported by Park and
the computerized process control of machine tools. Mitchell 71. This vision system is able to recognize and
One of the basic reasons for using tolerances in locate the part and its features in the 2D image based
engineering design is uncertainty relating to the on the 3D CAD database. It generates an intermediate
inherent inaccuracies of manufacturing equipment. vision-based representation to facilitate the recognition
When the precision and accuracy of a machine tool of objects from images. The system inspects the surface

volume 23 number 7 september 1991 479

finish and shape of the object, measures each specified 2 Requicha, A A G 'Mathematical models ot rigid
dimension, and compares the readings with the solid objects' l echnkal l'vlen~o 28 Production
nominal dimension and tolerance. Automation Project, University of Ro( hester, USA
(1977)
3 'Dimensioning and tolerancing' ANSI Standard
REMARKS
Y14.5-1982 American Society of Mechanical
In conclusion, it is appropriate to review the work so Engineers, USA (1982)
far accomplished in the different fields of D&T, and to
identify the necessary directions for future cooperative 4 'The VSAS (Variation Simulation Analysis Software)'
study. In tolerance representation, it is obvious that Applied Computer Solutions, St. Clair Shores,
the capabilities of solid models must be enhanced Michigan, USA (1984) (marketing brochure)
to incorporate both conventional and geometrical 5 Hillyard, R C and Braid, I C 'Characterizing
tolerances. As discussed earlier, the simple solution for non-ideal shapes in terms of dimensions and
tolerance representation is to hold the tolerance as an tolerances' Comput. Graph. Vol 12 No 3 (1978)
attribute of the geometry. To achieve total tolerance pp 234-238
attribution to the geometry, every relationship between
the geometric entities would have to be defined and 6 Requicha, A A G 'Part and assembly description
toleranced. Therefore, future solid models must support languages. Part I: Dimensioning and tolerancing'
a data structure that will allow the user to access both Technical Memo TM19 Production Automation
lower-level (i.e. points, edges, faces) and higher-level Project, University of Rochester, USA (May 1977)
(i.e. features) geometric entities. It is also necessary to 7 Requicha, A A G 'Representation of tolerances in
develop a mathematical foundation for tolerancing that solid modelling: issues and alternative approaches'
is independent of any specific computer representation. in Pickett, M S and Boyse, J W (Eds.) Solid Modeling
Another area that deserves much attention is that by Computers: From Theory to Applications Plenum
of finding a common base for the interpretation Press, USA (1984) pp 3 22
of tolerancing information, especially in automatic
inspection. There are important computational issues
8 Minagawa, M, Okino N and Kakazu, Y 'Automatic
(path planning, surface fitting etc.) that remain open. dimensioning on 3D solid geometry' Proc. 5th ICPE
Tokyo, Japan (1984)
The development of a suitable method for defining
the optimal manufacturing tolerances to respect the 9 Yuen, M M, Tan, S T and Yu, K M 'Scheme for
requirements of the design tolerancing is a very automatic dimensioning of CSG defined parts'
important topic for study. Although some work has Comput.-Aided Des. Vo120 No 3 (1988) pp 151-159
been done in this area, it has mostly been limited to
10 Suzuki, H, Kimura, F and Sata, T 'Treatment of
1D tolerancing only. Further extension of the prototype
dimensions on product modelling concept' Int.
models (e.g. Bourdet's model)in 3D solid modeling is
Syrup. Design & Synthesis Tokyo, Japan (Jul 1984)
a necessity.
D&T control is another important issue that needs 11 Requicha, A A G 'Toward a theory of geometric
to be addressed. So far, work has been done on tolerancing' Technical Memo 40 Production Auto-
computerizing the old 'tolerancing-chart' technique for mation Project, University of Rochester, USA (Mar
tolerance control, which is only suitable for conventional 1977)
tolerancing. Solid-model-based control strategy, and its 12 Rossignac, J R 'Blending and offsetting solid models'
implementation issues, need to be addressed for online PhD Thesis Dep. Electrical Engineering, University
manufacturing processes to monitor all the machined of Rochester, USA (1985)
surfaces within the allowable tolerance limits.
13 Rossignac, J R and Requicha, A A G 'Offsetting
operations in solid modelling' Comput. Aided
ACKNOWLEDGEMENTS Geom. Des. Vol 3 (1986) pp 129 148

This work was supported by the US National Science 14 Farmer, L E and Gladman, C A 'Tolerance
Foundation under grant CDR 8803017 to the Engineering technology - computer based analysis' ManuL
Research Center for Intelligent Manufacturing Systems Tech. CIRP Ann. Vol 35/1/1986 (1986) pp 7-10
at Purdue University, USA. The authors sincerely 15 Requicha, A A G and Chan, S C 'Representation
appreciate the help of Professor D C Anderson of of geometric features, tolerances and attributes in
CADLAB, School of Mechanical Engineering, Purdue solid modellers based on constructive geometry'
University. The authors also wish to thank an IEEE J Robot. & Automat. Vol RA-2 No 3 (1986)
anonymous referee who provided many useful and pp 156-166
constructive suggestions for the improvement of the
paper. 16 Jayaraman, R and Srinivasan, V 'Geometric
tolerancing. Part I: Virtual boundary requirements'
IBMJ. Res. & Develop. Vo133 No 2 (1989) pp 90-104
REFERENCES 17 Jayaraman, R and Srinivasan, V 'Geometric
1 Bjorke, O Computer-Aided Tolerancing Tapir tolerancing. Part I1: Conditional tolerances IBM J.
Publishers, Norway (1978) Res. & Develop. Vol 33 No 2 (1989)

480 computer-aided design

18 Johnson, R H 'Dimensioning and tolerancing final 32 Kimura, F 'GEOMAP-III: designing solids with
report' Report R-84-GM-02.2 Computer Aided free-form surfaces' IEEE Comput. Graph. & Applic.
Manufacturing International, USA (1985) (1984)
19 Burkett, W C 'PDDI approach to dimensioning and 33 Sata, T, Kimura, F, Suzuki, H and Fujita, T 'Designing
tolerancing a solid model' Report P-85-ASPP-02 machine assembly structure using geometric
Computer Aided Manufacturing International, USA constraints in product modelling' CIRP Ann.
(1985) Proc. Dimensioning & Tolerancing Workshop Vol 34/1 (1985) pp 169-172
20 Elgabry, A K 'A framework for a solid-based 34 Kimura, F, Suzuki, H and Wingrad, / 'A uniform
tolerance analysis' ASME Int. Computers in approach to dimensioning and tolerancing in
Engineering Conf. Chicago, USA (1986) product modeling' Preprints CAPE'86 Vol 1 (1986)
pp 166-178
21 Turner, J U 'Tolerances in computer-aided geometric
design' PhD Thesis Rensselaer Polytechnic Institute, 35 Yu, Y C, Liu, C R and Kashyap, R L 'A variational
USA (1987) solid model for mechanical parts' ASME Winter
Ann. Meet. PED Vol 21 (1986) pp 237-245
22 Roy, U and tiu, C R 'Feature based representational
scheme of a solid modeler for providing dimension- 36 Yu, Yuan-chen 'A new solid modeling system for
ing and tolerancing information' Int. Conf. CAD/CAM integration through engineering drawing
Manufacturing Science & Technology of the Future understanding' PhD Thesis Purdue University, USA
Massachusetts Institute of Technology, USA (Jun (Dec 1987)
1987) (also published in Int. J. Robot. & Comput.-
37 Yu, Y C, Liu, C R and Kashyap, R L 'ED-DESIGN:
Integr. Manut. Vol 4 No 3/4 (1988) pp 335-345)
an engineering drawing based design model for
23 Roy, U, Pollard, M D, Mantooth, K and /iu, C R computer understanding: Parts I and I1' Int. J. Robot.
'Tolerance representation scheme in solid model: & Comput.-Integr. Manuf. (to appear)
Part I' Technical Report TR-ERC 89-1 Engineering 38 Descotte, Y and Latombe, J 'GARI: an expert system
Research Center for Intelligent Manufacturing for process planning' in Pickett, M S and Boyse, J W
Systems, Schools of Engineering, Purdue University, (Eds.) Solid Modeling for Computers: From Theory
USA (Jan 1989) to Application Plenum Press (1984)
24 Roy, U, Mantooth, K, Pollard, M D and Liu, C R 39 Milacic, V R 'Computer-based informatization of
'Tolerance representation scheme in solid model: manufacturing engineering activities' Int. J. Prod.
Part I1' Technical Report TR-ERC 89-2 Engineering Res. Vol 20 No 3 (1982) pp 369-408
Research Center for Intelligent Manufacturing
Systems, Schools of Engineering, Purdue University, 40 Milacic, V R 'SAPT - expert system for manufactur-
USA (Jan 1989) ing process plan' Computer-Aided~Intelligent Process
Planning, ASME Winter Ann. Meet. Miami Beach,
25 Mashburn, T A 'A polygonal solid modeling Florida, USA (1985)
package' MS Thesis School of Mechanical Engineer-
ing, Purdue University, USA (Dec 1987) 41 Pro~ENGINEER:Concepts and Capabilities Parametric
Technology Corporation, Waltham, MA, USA
26 Gossard, D C, Zuffante, R P and Sakurai, H (1987)
'Representing dimensions, tolerances, and features
in MCAE systems' IEEE Comput. Graph. & Applic. 42 Michael, W and Siddall, J N 'The optimization
Vol 8 No 2 (1988) pp 51-59 problem with optimal tolerance assignment and
full acceptance' ASME Trans. J. Mech. Des. Vol 103
27 Birchfield, E B and King, H H 'Product definition (Oct 1981) pp 842-848
data interface (PDDI)' Proc. 1985 United States Air
Force C/M Industry Conf. (1985) 43 Michael, W and Siddall, J N 'The optimal tolerance
assignment with less than full acceptance' ASME
28 Wysk, R A 'An automated process planning and Trans. J. Mech. Des. Vo1103 (Oct 1981 ) pp 842-848
selection program: APPAS' PhD Thesis Purdue 44 Parkinson, D B 'Tolerancing of component
University, USA (May 1977)
dimensions in CAD' Comput.-Aided Des. Vol 16
29 Choi, Byoung Kyu 'CAD/CAM compatible, tool No 1 (1984) pp 25-32
oriented process planning for machining centers' 45 Lee, W and Woo, T C 'Tolerances: their analysis
PhD Thesis Purdue University, USA (1982) and synthesis' Technical Report 86-30 Dep.
30 Iwata, K, Kakino Y, Ohba, F and Sugimura, N Industrial and Operations Engineering, University
'Development of non-part family type computer- of Michigan, USA (Dec 1986)
aided production planning system CIMS/PRO' Adv. 46 Lee, W and Woo, T C 'Optimum selection of
Manuf. Tech. (1980) pp 171-184 discrete tolerances' Technical Report 87-34 Dep.
31 Iwata, K and Arai, E 'Development of integrated Industrial and Operations Engineering, University
modelling system for CAD/CAM of machine of Michigan, USA (Dec 1987)
product in Advances in CAD~CAM North-Holland, 47 Dong, Z and Soom, A 'Automatic tolerance analysis
Netherlands (1983) from a CAD database' ASME 86-DET-36 Design

volume 23 number 7 september 1991 481

 Enp,ineerin~ Tech. Conf. Columbus, Ohio, USA(Oct calculations' Dep. Mechanical Engineering, Beijing
19861 Institute of Aeronautics and Astronat~ti(s, China
48 Mansoor, E M 'Application of probability to 63 Liu, C R and Barash, M M 'Tile science and
tolerances used in engineering design' Pro(:. I Mech. advanced technology for cost-effective manufacture
E Vol 178 Pt 1 No 1 (1963 64) pp 29 39 of high precision engineering products' (Apr 1982)
(proposal submitted to and funded by US Office of
49 Hoffmann, P 'Analysis of tolerances and process
Naval Research)
inaccuracies in discrete part manufacturing'
Comput.-Aided Des. Vol 14 No 2 (1982) 64 Donmez, M A, Blomquest, D, Hocken, R, Liu, C R
and Barash, M M 'A general methodology for
58 Daniel, F 'Development of a computer program
machine tool accuracy enhancement by error
for the optimization of tolerance transfer from
compensation' Precision Eng. Vol 8 No 4 (1986)
design to manufacturing' MS Thesis Technion -
Israel Institute of Technology, Haifa, Israel (Mar 65 Ferreira, P M and Liu, C R 'A method for estimating
1986) and compensating quasistatic errors of machine
tools' in ASME Volume: Quality: Design, Planning
51 Daniel, F, Weill, R and Bourdet, P 'Computer aided
and Control (Dec 1987)
tolerancing and dimensioning in process planning'
Manu/. Tech. CIRP Ann. Vol 35/1/1986 (1986) 66 Roy, U and Liu, C R 'An expert system approach
pp 381-- 386 to the machine tool sequencing and machining
sequencing problems in computer aided process
52 Hillyard, R C and Braid, I C 'Analysis of dimensions
planning' Technical Report TR-ERC88-'16 Engineering
and tolerances in computer-aided mechanical
Research Center for Intelligent Manufacturing
design' Comput.-Aided Des. Vol 10 No 3 (1978)
Systems, Schools of Engineering, Purdue University,
pp 161-166
USA (Jul 1988)
53 Minnichelli, M O 'A preliminary investigation of
67 Vissa, N N 'A frame-based generative process
computer-aided tolerance analysis techniques' ME
planning system for machining prismatic parts'
Thesis Rensselaer Polytechnic Institute, USA (May
MSIE Thesis Purdue University, USA (Dec 1987)
1983)
68 Ferreira, P M and Liu, C R 'Generation of workpiece
54 Lin, V C, Gossard, D C and Light, R A 'Variational
orientations for machining using a rule-based
geometry in computer-aided design' Comput.
system' Robot. 8, Comput.-Inte~r. Manu/. Vol 4 No
Graph. Vot 15 No 3 (1981)
3/4 (1988) pp 545 -555
55 Light, R A 'Symbolic dimensioning in computer-
69 Trappey, J F C and Liu, C R 'Issues of automatic
aided design' MS Thesis Massachusetts Institute of
fixture design in computer integrated manufacturing
Technology, USA (1980)
systems' Proc. Int. Ass()(. Sci. & Tech. Develop. Int.
56 Light, R A and Gossard, D C 'Modification of Symp. Robotics & Automation Santa Barbara, CA,
geometric models through variational geometry' USA (25 27 May 1988) pp 149 154
Comput.-Aided Des. Vol 14 No 4 (1982)
70 Hopp 'CAD - directed inspection' Presented 34th
57 Aldefeld, B 'Variation of geometries based on a CIRP Con. Ass. Madison, Wisconsin, USA (19-25
geometric-reasoning method' Comput.-Aided Des. Aug 1984)
Vol 20 No 3 (1988) pp 117-126 71 Park, H D and Mitchell, O R 'CAD based planning
58 Fleming, A D 'Analysis of uncertainties and and execution of inspection' IEEEComputer Vision
geometric tolerances in assemblies of parts' PhD & Pattern Recognition Conf. Ann Arbor, Michigan,
Thesis University of Edinburgh, UK (1987) USA (5 9 Jun 1988)
72 Chase, K W and Greenwood, W H 'Design issues
59 Roy, U and Liu, C R 'Establishment of functional
relationships between the product components in in mechanical tolerance analysis' ASME Ann. Winter
Meet. Advanced Topics in Manufacturing Technology
assembly data base' Technical Report TR-ERC 88-7
Engineering Research Center for Intelligent Manu- Boston, MA, USA (13-18 Dec 1987)
facturing Systems, Schools of Engineering, Purdue
University, USA (May 1988) (also published in BIBLIOGRAPHY
Comput.-Aided Des. Vol 20 No 10 (1988) pp
570 580) Chan, S C and Voelcker, H B 'An introduction to MPL
a new machining/process programming language'
60 Karolin, A V 'Computer aided process routing Proc. Int. IEEE Conf. Robotics & Automation (Apr 1986)
development for discrete part manufacture' MS pp 333 344
Thesis Ohio State University, USA (1983)
Darbyshire, I and Davies, B J 'EXCAP: an expert
61 Ahluwalia, R S and Karolin, A V 'CATC a generative process planning system' University of
computer aided tolerance control system'/. Manuf. Manchester Institute of Science and Technology, UK
Syst. Vol 3 No 2 (1984) (1984)
62 Ji, P 'Computer aided operational dimensions Eastman, C M and Preiss, K 'A review of solid shape

482 computer-aided design

modelling based on integrity verification Comput.- ASME Winter Ann. Meet. Miami Beach, Florida, USA
Aided Des. (1984) pp 66-80 (1985)

Grossman, D D 'Procedural representation of three Kakino, Y, Ohba, F, Moriwaki, T and Iwata, K 'A new
dimensional object' IBM J. Res. & Develop. Vol 20 No method of parts description for computer-aided
6 (1976) production planning' Adv. Comput.-Aided Manuf.
(1977) pp 197-213
Hillyard, R 'The Build group of solid modellers' IEEE
Comput. Graph. & Applic. (Mar 1982) pp 43-52 Matsushima, K, Okada, N and Sata, T 'The integration
of CAD and CAM by application of artificial intelligence
Iwata, K and Sugimura, N 'A knowledge based techniques' CIRP Ann. Vol 31 (1982)
computer aided process planning system for machining Requicha, A A G 'Representations for rigid solids:
parts' 16th CIRPInt. Sem. Manufacturing Systems Tokyo, theory, methods and systems' ACM Comput. Surv. Vol
Japan (1984) 12 No 4 (Dec 1980) pp 437-464
Iwata, K and Sugimura, N 'An integrated CAD/CAPP Wang, K K (Ed.) 'TIPS-1' Sibley School of Mechanical
system with know-hows on machining accuracies of and Aerospace Engineering, Cornell University, USA
parts' Computer-Aided~Intelligent Process Planning, (Oct 1984) (1983 version)

volume 23 number 7 september 1991 483