Computer-Aided Design 53 (2014) 113

Contents lists available at ScienceDirect

Computer-Aided Design
journal homepage: www.elsevier.com/locate/cad

Review

A comprehensive study of three dimensional tolerance

analysis methods
Hua Chen a , Sun Jin a,b, , Zhimin Li a , Xinmin Lai a,b
a

Shanghai Key Laboratory of Digital Manufacture for Thin-walled Structures, Shanghai Jiao Tong University, Shanghai, China

State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University, Shanghai, China

highlights

Introduce four major 3D tolerance analysis models briefly.

Make a comprehensive comparison and discussion between them.
Expound the connotation of 3D tolerance analysis.
Present a perspective overview of the future research about 3D tolerance analysis.

article

info

Article history:
Received 29 October 2013
Accepted 27 February 2014
Keywords:
3D tolerance analysis
T-Map
Matrix
Unified JacobianTorsor
Direct linearization method
Comparison

abstract
Three dimensional (3D) tolerance analysis is an innovative method which represents and transfers
tolerance in 3D space. The advantage of 3D method is taking both dimensional and geometric tolerances
into consideration, compared with traditional 1/2D tolerance methods considering dimensional
tolerances only. This paper reviews four major methods of 3D tolerance analysis and compares them based
on the literature published over the last three decades or so. The methods studied are Tolerance-Map
(T-Map), matrix model, unified JacobianTorsor model and direct linearization method (DLM). Each
of them has its advantages and disadvantages. The T-Map method can model all of tolerances and
their interaction while the mathematic theory and operation may be challenging for users. The matrix
model based on the homogeneous matrix which is classical and concise has been the foundation of
some successful computer aided tolerancing software (CATs), but the solution of constraint relations
composed of inequalities is complicated. The unified JacobianTorsor model combines the advantages
of the torsor model which is suitable for tolerance representation and the Jacobian matrix which is
suitable for tolerance propagation. It is computationally efficient, but the constraint relations between
components of torsor need to be considered to improve its accuracy and validity. The DLM is based on the
first order Taylors series expansion of vector-loop-based assembly models which use vectors to represent
either component dimensions or assembly dimensions. Geometric tolerances are operated as dimensional
tolerances in DLM, which is not fully consistent with tolerancing standards. The results of four models
with respect to an example are also listed to make a comparison. Finally, a perspective overview of the
future research about 3D tolerance analysis is presented.
2014 Elsevier Ltd. All rights reserved.

Contents
1.
2.

Introduction........................................................................................................................................................................................................................
3D tolerance analysis models ............................................................................................................................................................................................
2.1.
T-Map (Tolerance-Map) model .............................................................................................................................................................................

2
3
3

Corresponding author at: Shanghai Key Laboratory of Digital Manufacture for Thin-walled Structures, Shanghai Jiao Tong University, Shanghai, China. Tel.: +86 021
34206306; fax: +86 021 34206306.
E-mail address: jinsun@sjtu.edu.cn (S. Jin).
http://dx.doi.org/10.1016/j.cad.2014.02.014
0010-4485/ 2014 Elsevier Ltd. All rights reserved.

H. Chen et al. / Computer-Aided Design 53 (2014) 113

3.
4.

2.2.
Matrix model..........................................................................................................................................................................................................
2.3.
Unified JacobianTorsor model.............................................................................................................................................................................
2.4.
DLM method...........................................................................................................................................................................................................
Discussion and comparison ...............................................................................................................................................................................................
Conclusion ..........................................................................................................................................................................................................................
Acknowledgments .............................................................................................................................................................................................................
References...........................................................................................................................................................................................................................

1. Introduction
The objective of tolerance analysis is to check the feasibility
and quality of assemblies or parts for a given GD&T scheme. The
results of tolerance analysis include worst case variations and
statistical distribution of functional requirement, acceptance rates,
contributors and their percent contributions, and the sensitivity
coefficients with respect to each contributor. Tolerance analysis is
an essential part for mechanical design and manufacturing because
it affects not only the performance of products but also the cost.
Tolerance analysis, including tolerance representation and tolerance propagation (tolerance transfer), can be classified into
many categories based on the analysis objective and analysis approach, as shown in Fig. 1. According to dimensionality, there
are one dimensional (1D), two dimensional (2D) and three dimensional (3D) tolerance analyses. Three approaches are applied
for 1/2/3D tolerance analysis, i.e., worst case (deterministic case),
statistical case and Monte Carlo simulation. Rigid and flexible tolerance analysis are two different models in the light of analysis objective. The former is surface-based and needs shape closure only,
such as engines tolerance analysis; the latter is point-based and
needs shape and force closure simultaneously, such as auto-bodies
tolerance analysis where the finite element method (FEM) is used
to take the deformation into consideration [14]. The division into
part level and assembly level is another classification. The stackup effect of assembly can be described by virtue of assembly function explicitly or implicitly, depending on the assembly method
and sequence, as well as the property of components [5]. Tolerance analysis runs through the whole process of the product, including design, process planning, manufacturing, inspection, but
the objective may be different in each phase. For example, the tolerance scheme, i.e., conventional (parametric) and geometric tolerance will be selected and specified, and then tolerance analysis for
functional requirement will be carried out in design phase. Meanwhile, besides manual analysis, computer aided tolerancing softR
R
R
ware (CATs), such as VisVSA
, 3DCS
and CETOL
are applied to
tolerance analysis successfully [69]. To be sure, the classification
of tolerance analysis will be more and more complicated with the
development of mechanical design and manufacturing.
Over the last thirty years, a large amount of fundamental research efforts has been given to explore the mathematical basis
for tolerance analysis. For tolerance representation, the models or
concepts include variational geometry [1012], variational class
[13,14], virtual boundary [15,16], feasibility space [17,18], vectorial approach [19], virtual joints [20], degree of freedom (DOF)
[2123], Tolerance-Map (T-Map) [24,25], topologically and technologically related surfaces (TTRS) [26], infinitesimal matrix [27],
matrix [2830], small displacement torsor (SDT) [31,32], and proportioned assembly clearance volume (PACV) [33,34]. Similarly,
for tolerance propagation, the approaches or methods consist of
the linearization method [35], system moments [36,37], quadrature [3840], reliability index [41,42], the Taguchi method [43,44],
Monte Carlo simulations [45,46], network of zones and datums
[47], kinematic formulation [48], the direct linearization method
(DLM) [49,50], Jacobian matrix [51,52], state space [53,54], and the
variational method [55]. It is worth noting that the partition of two
categories mentioned above is approximate and based mainly on

5
7
9
10
11
12
12

their strong suits, because there is no boundary between the tolerance representation and propagation for these models, such as the
TTRS [56].
As new generations of tolerancing standards, i.e., ASME Y14.52009 [57] and ISO 1101 [58] were released and popularized, geometric tolerances are generally accepted as industry practices.
The traditional 1/2D tolerance analysis models are insufficient to
meet the ever-tightening and increasingly complex requirements
of tolerance analysis in various fields [59]. More specifically, variations of a feature caused by geometric tolerances are three dimensional, which cannot be considered by 1/2D methods. Researchers
and engineers need a new method that can analyze how those geometric tolerances are represented and propagated in three dimensional space urgently. It is the 3D tolerance analysis method. Let
us take a combustion engine as an example, as shown in Fig. 2.
The translational and rotational variations of piston accumulated
by geometric and dimensional tolerances of crank-link parts have a
significant impact on the compression ratio. In addition, tolerances
of parts affect not only the dimensional quality of assembly, but
also other qualities such as frictional work [60,61] and sealing.
Finding out the mapping relationship of tolerance between parts
and functional requirements and performance indexes is important to engine design. 3D tolerance analysis methods will offer a
significant clue for understanding the role of every tolerance of
parts in the variation stream (gray boxes in Fig. 2).
The 3D tolerance analysis is an innovative method which represents and transfers tolerance in 3D space. Geometric tolerances and
dimensional tolerances, as well as the interaction between them
in the tolerance zone can be taken into consideration by 3D tolerance analysis methods. Moreover, abundant results, i.e., the translational and rotational variations of target feature are obtained in
these methods. Many models have been developed for 3D tolerance representation and propagation since 1990s. Portman [27]
introduces a spatial dimensional chain where the individual
error is represented as an infinitesimal matrix to model the tolerance propagation. Fleming [47] illustrates the geometric relationships by a network of zones and datums connected by arcs to
which constraints are assigned. The effects of these constraints are
calculated through the network between nodes. Rivest et al. [48]
propose a kinematic formulation which exploits the kinematic
character of a toleranced feature relative to its datum. These three
methods are preliminary explorations of 3D methods. Laperrire
and Lafond [20,51] use virtual joints for tolerance representation and the Jacobian matrix for tolerance propagation. Davidson
et al. [24] present a T-Map representing all possible variations of
size, position, form, and orientation for a target feature. Desrochers
and Rivire [29] represent the variations of a feature with a displacement matrix and transfer them with a homogeneous matrix.
An SDT model introduced by Clment et al. [31] uses six small displacement vectors to represent the position and orientation of an
ideal surface in relation to another ideal surface in a kinematic way.
Desrochers et al. [62] put forward a unified JacobianTorsor model
which combines the advantages of the torsor model and the Jacobian matrix. Chase et al. [50] introduce a DLM based on the first
order Taylors series expansion of vector-loop-based assembly
models which use vectors to represent either component dimensions or assembly dimensions. Some models mentioned above
have been applied extensively by virtue of CATs.

H. Chen et al. / Computer-Aided Design 53 (2014) 113

Fig. 1. Categories of tolerance analysis.

2. 3D tolerance analysis models

2.1. T-Map (Tolerance-Map) model
R
(Patent No. 6963824) model developed by DavidThe T-Map
son et al. [24] is a hypothetical Euclidean volume of points, the
shape, size, and internal subsets of which represent all possible
variations in size, position, form, and orientation of a target feature. A T-Map is a convex set resulting from a one-to-one mapping
from all the variational possibilities of a feature within its tolerance zone, constructed from a basis-simplex and described with
areal coordinates.
The areal coordinates use areal parameters to describe the
position of a point in a reference triangle. Given three fixed points
1 , 2 , 3 , called basis points that are chosen in Euclidean space,
the position of any point is uniquely determined by the linear
equation:

= 1 1 + 2 2 + 3 3

Fig. 2. The mapping relationship of tolerances between parts and functional

requirements and performance indexes in engines.

The purpose of this work is to discuss four typical methods of

3D tolerance analysis, i.e., the T-Map model, the matrix model, the
unified JacobianTorsor model and DLM, all of which are research
hotspots recently. The TTRS theory is also introduced because it is
the basis of the matrix and torsor models. Although several review
papers have introduced these methods briefly [6370], it is the first
time to put them together to make a comparison in 3D context. It
should be noted that the concept of 3D here emphasizes tolerance
representation and propagation in three dimensional space, other
models, such as the state space model which focuses on tolerance
transfer with a state space method, the variational method which
pays close attention to variation propagation of fixtures with total
deterministic locating (321 scheme), are not discussed because
they are point-based tolerance schemes where variations caused
by geometric tolerances are not three dimensional.
The rest of the paper is organized as follows: Section 2 gives
an overall introduction of these 3D tolerance analysis models.
Section 3 makes a comprehensive discussion and comparison
between them. Finally, conclusions with a perspective overview
of the future research about 3D tolerance analysis are given in
Section 4.

(1)

where 1 , 2 , 3 are areal coordinates about , and have the

relation 1 + 2 + 3 = 1. Either one or two of these coordinates
will be negative when is chosen outside 1 2 3 .
In order to illustrate the modeling process of T-Map, here we
consider a cross-section of a round bar with a size tolerance t on
its length, as shown in Fig. 3. Fig. 3(a) shows the tolerance zone
ABCD where all points of the end face must lie in, and the Cartesian
coordinate system. Assuming a perfect round plane with diameter
d and no thickness, the possible displacement of this plane is the
tolerance zone, i.e., the volume limited by the planes 1 and 2 , and
the cylindrical diameter of the bar. The plane would translate along
the z axis and rotate around the x axis freely on two dimensional
occasions where the plane is always parallel to the x axis. The
plane 3 in Fig. 3(a) represents the greatest clockwise tilt of the
perfect plane in tolerance zone. After that, the two dimensional
set of planes in Fig. 3(a) defined by three basis planes 1 , 2 , 3
are mapped to the areal coordinates, as shown in Fig. 3(b). Any
end plane of the round bar that satisfies the size tolerance and is
parallel to the x axis will be represented by expressed as Eq.
(1). Especially, the points on the line-segment 1 2 in Fig. 1(b)
represent the parallel planes that are perpendicular to the z axis in
Fig. 1(a) and lie between 1 and 2 . The points on the line-segment
1 3 in Fig. 1(b) correspond to these planes in Fig. 1(a) that are
parallel to the x axis, pass through point B, and lie in the tolerance
zone between 1 and 3 . Similarly, the points on the line-segment
2 3 in Fig. 1(b) correspond to these planes in Fig. 1(a) that are

H. Chen et al. / Computer-Aided Design 53 (2014) 113

Fig. 3. Modeling process of the T-Map for a round bar with a size tolerance.
Fig. 5. (a) A half section of the T-Map for a round bar with a form tolerance and a
size tolerance. (b) A half section of the T-Map for a round bar with an orientation
tolerance and a size tolerance.

Fig. 4. A plane in the Cartesian frame of a tolerance zone.

parallel to the x axis, pass through point C , and lie in the tolerance
zone between 2 and 3 .
The three dimensional tolerance zone at the end of the round
bar is obtained by a full sweeping operation around z axis with the
rectangle ABCD in Fig. 3(a). Therefore, the three dimensional T-Map
is modeled by revolving the triangle in Fig. 3(b) a full turn around
the line 1 2 , it is a right-circular dicone shown in Fig. 3(c). It
should be mentioned that the areal coordinates can also be used to
identify points in three dimensional space where four basis points
are non-coplanar. The points 1 , 2 , 3 , 4 in Fig. 3(c) are selected
to establish a tetrahedron of reference
for three dimensional area
coordinates in the T-Map. Setting
i = 1, any end plane of the
round bar that satisfies the size tolerance will be represented by
the linear equation:

= 1 1 + 2 2 + 3 3 + 4 4 .

(2)

The transformation from T-Map in Fig. 3(c) to Cartesian coordinates in Fig. 3(a) is worth mentioning. Presuming a plane in the
Cartesian coordinates, as shown in Fig. 4, its position can be described by the equation px + qy + rz + s = 0. p, q, r are the direction cosines in which r is approximately equal to 1 because the
rotation displacement of the plane shown in Fig. 3(a) around the
z axis is smaller than other two rotation displacements in the tolerance zone, and s is the absolute distance from the plane to the
origin of coordinates. Therefore, the planes in the tolerance zone
are distinguished by the coordinates p, q and s only. q in Fig. 3(b) is
obtained by assigning dimension or length on q in Fig. 3(a) because
q is dimensionless. The lateral dimension t of q axis in Fig. 3(b) is
t = d(t /d) = d tan() = dq = q . p is obtained in the same way
as q . Consequently, the transformation of any plane in the tolerance zone of Fig. 3(a) from areal coordinates to Cartesian coordinates is:

p
0
q 0
r = 1
s
t /2

0
0
1
t /2

0
t /d
1
0

t /d
1
0 2
.
1 3
4
0

(3)

It should be stressed that a T-Map for a single part is always

convex. This property allows the usage of fundamental principles
of convex sets, such as Minkowski sums and differences.
We continue to use the round bar shown in Fig. 3 as an objective
to illustrate the modeling process of form and orientation tolerance
for T-Map, but now a flatness tolerance t1 and a parallelism
tolerance t2 which is relative to a datum plane (another end of
bar) are specified besides the size tolerance t. According to the

Fig. 6. An assembly of two round bars for the T-Map.

tolerancing standards, the flatness tolerance zone is defined by two

parallel planes with the distance t1 . Its position and orientation
are not constrained within the size tolerance zone. The internal
triangular sub-set drawn with a dashed line in Fig. 5(a) represents
the flatness tolerance while the hatched triangle corresponds
to positions that the flatness tolerance zone can occupy. The
Minkowski sum of the sub-set of the dashed triangle and the
hatched triangle is the tolerance t. Similarly, the parallelism
tolerance zone is defined by two parallel planes which are parallel
to planes 1 and 2 with the distance t2 , represented by the
dashed triangle in Fig. 5(b). Because the parallelism tolerance zone
can only translate up or down within the tolerance zone t, the
Minkowski sum of sub-set of dashed triangle and its translational
zone shown in Fig. 5(b) is a truncated map along the q axis.
In order to illustrate the tolerance propagation of T-Map, a
simple assembly composed of two parts is shown in Fig. 6 as an
example. Part 1 is a round bar. Part 2 consists of two round bars
separated by an offset b. They are assembled end to end coaxially.
The functional surface is the upper face of the bar with diameter d2
in part 2. The dimensions and tolerances of assembly can be seen
in Fig. 6. The coordinate frames are similar to Fig. 3.
The variations of orientation in the tolerance zone of part 1 are
amplified and cause positional variations of the functional surface
because of the offset b. The accumulation T-Map depends on the
offset b and the diameters d1 and d2 since functional requirements
for the target face in the assembly control the assignment of
tolerances to the individual parts. Here we only consider the

H. Chen et al. / Computer-Aided Design 53 (2014) 113

Fig. 7. (a) q -s sections of the accumulation and functional requirement T-Maps for
the assembly when b = 0. (b) q -s section of two operand dicones for the assembly
when b = 0.

Fig. 8. A planar feature in a tolerance zone.

condition where d1 > d2 . The T-Maps of part 2 and functional

requirements in which the aspect ratios are unity are selected
to be constructed on the basis tetrahedron 1 2 3 4 in Fig. 3(c).
When b = 0, the half-section of the accumulation (solid line)
and functional requirement (dashed line) T-Maps are shown in
Fig. 7(a) in which the accumulation T-Map is inscribed in the
functional T-Map in order to avoid specifying excessively tight
tolerances on individual parts. ta and tf which are equal to t1 +
t2 represent accumulation tolerance and functional requirement
tolerance along the length direction. ta which is equal to t11
(d2 /d1 ) + t2 represents orientation tolerance of functional surface,
which is tighter than functional requirement.
However, it is somewhat complicated if b = 0. Assuming a rotational angle q1 in the tolerance zone of part 1, s of T-Map of
part 1 becomes s + bq1 which results in point 3 in Fig. 3(c) moves
downward by the amount bt11 /d1 and the opposite point 7 moves
upward by the same amount. That is an oblique elliptic dicone
truncated by the orientation tolerance t11 , shown by the dashed
line in Fig. 7(b). When b (d1 d2 )/2, the point 3 in T-Map of part
1 is still within the functional T-Map, and the accumulation tolerance in the axial direction is t1 + t2 . When b > (d1 d2 )/2, the point
3 in T-Map of part 1 touches the line 1f 3f , and the accumulation
tolerance along the axial direction is t11 ((2b + d2 )/d1 1)+ t1 + t2 .
Besides round bars, T-Maps have been developed for other features, such as polygonal bars [25], axes [7173], angled faces [74],
pointline clusters [75], and planar and radial clearance with a statistical way [7678]. T-Map can also be used for tolerance allocation or synthesis, the detailed discussion is not within the scope of
this paper.
2.2. Matrix model
The matrix model introduced by Desrochers and Rivire [29]
uses a displacement matrix D to describe the small displacements
of a feature within the tolerance zone and the clearance between
two features. Using the theory of the set of displacements by
Herv [79], Clment et al. [26,31,32] have proven that there are

only seven elementary surface types, as shown in Table 1. The combinations of surfaces are called TTRS when two surfaces belong to
the same part or Pseudo-TTRS when two surfaces belong to two different parts [80]. The reference elements are classified by the concept of minimum geometric datum element (MGDE) [81]. Thirteen
relative positioning constrains of basic components of the MGDE,
i.e., point, line and plane are defined, as shown in Table 1. The concept of functional requirements is also declared in the wake of the
TTRS [82].
The 4 4 homogeneous matrix D including a 3 3 rotational
matrix and a 3 1 translational matrix is chosen here to represent
the relative displacement of a feature within tolerance zone:
C C
S C
S
0

S C + C S S
C C + S SS
C S

S S + C SC
C S + S S C
CC
0

v
w

(4)

where , , are rotational displacements around the x, y, z axes

respectively; u, v, w are translational displacements along the
x, y, z axes respectively; C is the abbreviation of cos() and S is sin().
Tolerance is only meaningful in directions other than those
that leave a surface invariant with respect to itself [83], so every
parameter in the matrix can be seen as a micro DOF that leaves
the feature non-invariant. Let us take a planar surface shown in
Fig. 8 as an example, the ideal plane without form tolerance and
thickness can translate along the x axis and rotate around the y and
z axes in the tolerance zone. In other words, a plane has three noninvariant degrees, i.e., , and u within its tolerance zone. Another
three displacements (, v and w ) are invariant and set to zero.
Therefore, Eq. (4) can be simplified for the planar non-invariant
matrix D (see Eq. (5)). Non-invariant displacement matrix of the
other five surfaces can be seen in Table 1.
C C
S C
S
0

S
C

0
0

C S
S S
C
0

u
0
.
0
1

(5)

The matrix representation is completed by a set of inequalities

defining the bounds of every component in the matrix. These
inequalities depend on the type of surface and tolerance. If more
than one tolerance is specified on the same feature, their effects
are calculated through the principle of effects overlapping. Taking
the four vertexes (A, B, C , D) into consideration, the bounds of the
planar surface with t shown in Fig. 8 can be expressed by:
tSL u(A,B,C ,D) tSU
t /a t /a
t /b t /b

with

uB

uC

uD

=D

CB

CC

(6)

CD

where CA,B,C ,D are coordinates for points A, B, C , and D.

Two types of datum reference frame (DRF) need to be identified
for tolerance transfer in the matrix model when the assembly
graph is created, i.e., global DRF (R) which is the evaluation
reference of the functional feature and local DRF (Ri ) of each part
feature. The homogeneous matrix P which depends only on the
nominal geometry represents the transformation of the R to the Ri .
Given two points in Ri , the theoretical point M and the same point
M after a displacement in the tolerance zone, [MM ] represents
the total displacement of the M because of matrix Di , which can be
expressed by:
MM

= PR1 Ri MM R
i

1
= PRRi M R [M]Ri
i

1
PR
Ri

(Di I ) [M]Ri .

(7)

Line + plane

Point + line

(Point + line) or
(Line + plane)

Line

Plane

Ty,z
Rx,y,z

Tx,y,z
Ry,z

Tx,y,z
Rx

Ty,z
Ry,z

Tx
Ry,z

Tx,y,z

Prismatic

Revolution

Helical

Cylindrical

Planar

Spherical

Ci/Cj

General

TTRS

Ci/Cj

Ci/Cj

Prismatic

Ci/Cj

Ci/Cj

Ci/Cj

Ci/Cj

Ci/Cj

Ci/Cj

Ci/Cj

Revolution Helical

Ci/Cj

Ci/Cj

Ci/Cj

C11/C12/C13

Ci/Cj

Cylindrical

Ci/Cj

Ci/Cj

Ci/Cj

C6/C7

C8/C9/C10

Ci/Cj

Planar

Ci/Cj

Ci/Cj

Ci/Cj

Ci/Cj

C1/C2

C3

C4/C5

Spherical

0
0

0
0
1
0
0

C C
S C
S

C C
S C
S

C C
S C
S
0

C C
S C
S

0
0
1
0

0
0
u

v
w

S
C

0
0

S
C

0
0

S
C

0
0

C S
S S
C
0

C S
S S
C
0

C S
S S
C
0

C S
S S
C
0

u
0

0
1

v
w

p
2

S S + C SC
C S + S S C
CC

S S + C SC
C S + S S C
CC

v
u =
w

v
w

S
C

S C + C S S
C C + S SS
C S

C C
S C
S

S C + C S S
C C + S SS
C S

C C
S C
S

Non invariant displacements matrix

v
w

v
w

Notice:
1. T and R represent the translation and rotation respectively, the subscripts of x, y, z represent the orientation.
2. C is the abbreviation of cos( ) and S is the abbreviation of sin( ).
3. u, v, w are three translational displacements along the x, y, z axes respectively.
4. C(onstraints)1: pointpoint, coincidence; C2: pointpoint, distance; C3: pointplane, distance; C4: pointline, coincidence; C5: pointline, distance; C6: planeplane, parallel, distance; C7: planeplane, angle; C8: planeline,
perpendicularity;
C9: planeline, parallel, distance; C10: planeline, angle; C11: lineline, coincidence;

C12: lineline, parallel, distance; C13: lineline, angle and distance.

5.
Ci/Cj is the combination of C1C13 depending on the MGDE of two surfaces. For example, the
Ci/Cj corresponding to revolution and spherical surface are the combinations of point and line, i.e., C1/C2, C4/C5, C11/C12/C13.

Point

Point + line + plane

General

MGDE

Tx,y,z
Rx,y,z

Table 1
Constraints of TTRS and matrices.
Surface
Non invariant
displacements

6
H. Chen et al. / Computer-Aided Design 53 (2014) 113

H. Chen et al. / Computer-Aided Design 53 (2014) 113

The equation above gives the range of displacements of the ith

feature in the global DRF that is allowed by the tolerance zone.
More points are necessary to specify the constraints that keep the
feature inside the bounds of the tolerance zone. Taking the planar
surface shown in Fig. 8 as an example, the additional constraints of
points A, B, C and D are:
tSL (Di I ) MA,B,C ,D

Ri

x tSU

(8)

where x represents the directional vector of tolerance zone.

It should be mentioned that other geometric tolerances, such
as orientation tolerance, also can be modeled by Eq. (8) through
the arithmetic operation of some vertex points according to the
tolerancing standards. Moreover, a cylindrical clearance zone can
be viewed as an equivalent concentricity tolerance in the matrix
model.
For convenience, the assembly in Fig. 6 is also used to demonstrate the tolerance analysis of the matrix model, as shown in Fig. 9
where three local DRFs (R1 , R2 , R3 ) and one global DRF (R0 ) are
specified. The assembly graph is shown in Fig. 10. The functional
requirement is the displacement of point a along the z axis in the
global DRF, which is equal to za/D1 + za/D2 . Two matrices D1 and
D2 which are similar to Eq. (5) are established for tolerance representation of the upper surface relative to the under surface of two
round bars. With Eq. (7), we can obtain the two displacements of
point a:
za/D1 = u1 + b cos(1 ) sin (1 )
+ l2 (cos(1 ) cos(1 ) 1)
za/D2 = u2 .

Fig. 9. The assembly of two round bars for the matrix model.

Fig. 10. Assembly graph of the matrix model.

(9)
(10)

The constraints of the components of D1 under the tolerance t1 and

t11 with respect to four vertexes of part 1 are:

u1 + (d1 cos(1 ) sin(1 ))/2

t
t1
u1 d1 sin(1 )/2
1

u1 (d1 cos(1 ) sin(1 ))/2
2
2

u1 + d1 sin(1 )/2

d1 cos(1 ) sin(1 )
t11
t11 .
d1 sin(1 )

(11)

(12)

The constraints of the components of D2 under the tolerance t2

with respect to four vertexes of part 2 are:

u2 + (d2 cos (2 ) sin(2 ))/2

t
t2
u2 d2 sin(2 )/2
2

u2 (d2 cos(2 ) sin(2 ))/2

2
2

u2 + d2 sin(2 )/2

Fig. 11. Small displacement torsor of a surface.

The torsor, also known as small displacement torsor (SDT), is

a small displacement screw used to represent the position and
orientation of an ideal surface or its feature (axis, centerline, plane)
in relation to another ideal surface with a kinematic way [34]. As
shown in Fig. 11, at a given point O on nominal surface S0 , the SDT
of variational surface S1 from S0 can be expressed as:

(13)

where t1 /2 u1 t1 /2, t2 /2 u2 t2 /2, t11 /d1 1

t11 /d1 , t11 /d1 1 t11 /d1 , t2 /d2 2 t2 /d2 , t2 /d2 2
t2 /d2 .
Standard optimization algorithms, such as simplex, can be
applied to solve this problem.
The statistical method of the matrix model for tolerance analysis can be found in [30]. It is worth mentioning that some CATs
based on the matrix model have been developed and applied sucR
cessfully, such as CATIA.3D FDT
[6] and FROOM [68,69].
2.3. Unified JacobianTorsor model
The unified JacobianTorsor model introduced by Desrochers
et al. [62] combines the advantages of the torsor model which
is suitable for tolerance representation and the Jacobian matrix
which is suitable for tolerance propagation. Tolerance analysis of
this model is in a kinematic way because of the concept of torsor
which is also on the basis of TTRS.

T = 1/0

1/0

v
w

(14)

where 1/0 is the rotational vector around the axis including three
specific vectors, i.e., , and indicate the vectors around the
axes x, y and z in the local reference system respectively; likewise,
1/0 is the translational vector and u, v, w are three specific vectors
along the axes x, y and z respectively.
The relative translation between two surfaces at any point M in
Euclidean space can be obtained by a linearization rule in terms of

1/0 and a cross product of 1/0 and vector OM.

u
y + z

dM = 1/0 + 1/0 OM = v + x z .
w
x + y

(15)

The SDT is well suited to not only 3D tolerance representation

but also 3D metrology [84,85]. It should be noted that the SDT
for tolerance representation is the first order approximation of
the matrix model introduced in the previous subsection [86]. The
rotation angles in Eq. (4) can be simplified by sin( ) and
cos( ) 1 because the angles are very small in the tolerance zone.

H. Chen et al. / Computer-Aided Design 53 (2014) 113

Then the 3 3 rotational matrix can be written as:

the expression of which can be written as follows:

(16)

And the small displacement of any point M can be obtained by:

u
1

x
v +
1
y
w

1
z

u
y + z
= v + x z .
w
x + y

dM =

(17)

Generally, the constraints of components of SDT depend on

the type of tolerance and feature. The SDT used for tolerance
representation of planar and cylindrical surfaces, as well as their
joints (clearance) has been studied deeply. Seven types of surface
and their screws are listed in [86]. The concept of non-invariant of
the matrix model is also effective in the SDT model. , v and w in
Eq. (14) that have no effect on tolerance analysis of the plane shown
in Fig. 8 can be set to zero to reduce the scale of computational
work.
Tolerance representation with SDT is concise and intuitionistic,
but it is difficult for tolerance transfer [64,65]. Therefore, Jacobian
matrix is introduced into tolerance analysis.
The Jacobian method is a linear arithmetic formulation applied
in series robot system, mapping the displacement or velocity of
joints to end joint. Laperrire and Lafond [20,51] bring it into
tolerance analysis by introducing the concept of functional pairs
expressed by a set of virtual joints.
There are two types of functional pairs in the assembly, i.e., internal pair and contact or kinematic pair. The former is composed
of two functional elements (FEs) on the same part; the latter is
made up of two FEs on different parts if there is a physical or potential contact between them.
The Jacobian matrix for ith FE can be expressed as:

[J ]FEi

i
[R0 ]33 [RPti ]33

[0]33

..
.
..
.
..
.

[Win ]33 ([Ri0 ]33 [RPti ]33 )

(18)

i
([R0 ]33 [RPti ]33 )

where R0i represents the local orientation of ith frame with

respect to 0th frame that is the global reference system; [RPTi ] is
a projection matrix designating the unit vectors along local axes
respectively for tolerance
zone tilted according to the direction
of tolerance analysis; Win is a skew-symmetric matrix allowing
the representation of the vector
among
the ith and nth frame (end
point), defined in Eq. (19); R0i Win reflects the leverage effect
when the small rotations of FE are being multiplied by terms of the
Jacobian matrix.
0
dzin
dyni

Win 33

dzin
0
dxni

dyni
dxni

(19)

where dzin = dzn dzi , dyni = dyn dyi , dxni = dxn dxi .
The SDT model is suitable for tolerance representation while the
Jacobian matrix is suitable for tolerance propagation. The unified
JacobianTorsor model combines the advantages of both methods,

u, u

v, v

w, w

FR

u, u

v, v


w, w

FE1

..
= [J]FE1 [J]FEn

u, u

v, v

w, w

(20)

FEn

where FR represents the functional requirement; (, ) is the

tolerance interval where must lie in, other vectors follow the
same way. The interval arithmetic is incorporated into Eq. (20)
to allow tolerance analysis to be performed on a tolerance zone
basis rather than on a point basis.
The assembly graph of Fig. 9 with the unified JacobianTorsor
model is shown in Fig. 12, including two internal FEs and one
contact FE. The target feature is the upper face of the bar with
diameter d2 in part 2. All coordinate frames are in the middle of
the tolerance zone or contact zone. The contact pair between part
1 and part 2 is considered as zero because there is no clearance
between two contacting planes. With Eq. (20), the final expression
of the unified JacobianTorsor formulation about the assembly in
Fig. 9 is:

(u, u)

(v, v)

(w, w)

(, )

(, )

( , )

FR

0
0

l2

0
0

l2

(0 , 0 )

(0 , 0 )

(t1 /2, t1 /2)

(t11 /d1 , t11 /d1 )

(t11 /d1 , t11 /d1 )

(0 , 0 )
FE1 .

(0 , 0 )

(0 , 0 )

(t2 /2, t2 /2)

(t11 /d2 , t11 /d2 )

(t11 /d2 , t11 /d2 )

(0 , 0 )

FE1

FE2

(21)

FE2

As can be seen, the orientation tolerance mainly limits the

rotational displacements of the upper surface of part 1. We only
focus on the value along the z axis of FR in this paper. The result
shows w must lie in the interval of [((t1 + t2 )/2 + bt11 /d1 ), ((t1 +
t2 )/2 + bt11 /d1 )]. It should be pointed out that the bt11 /d1 of the
result is the so called leverage effect caused by the small rotational
displacement t11 /d1 and the offset b.
The statistical method of the unified JacobianTorsor model
where Monte Carlo simulation is applied has been developed
[8789]. Moreover, the unified JacobianTorsor model can also be

H. Chen et al. / Computer-Aided Design 53 (2014) 113

Fig. 14. The effects of flatness tolerance t at the kinematic joint b of Fig. 13.

Fig. 12. Assembly graph of the unified JacobianTorsor model.

Fig. 13. An example of kinematic joints.

used for redesign of assembly tolerance where the contribution

of each FE can be calculated [90], and geometrical variations
management in a multi-disciplinary environment [91].
2.4. DLM method
The DLM (Direct Linearization Method) proposed by Chase
et al. [49,50] is based on the first order Taylors series expansion
of the assembly kinematic constraint equations with respect to
both the assembly variables and the manufactured variables in
assembly. The assembly equations expressed by the vector-loopbased assembly models which use vectors to represent either
component dimensions or assembly dimensions take three main
sources of variation into account in a mechanical assembly. They
are dimensional variations and geometric feature variations which
are the results of the natural variations in manufacturing processes,
and kinematic variations are small adjustments between mating
parts that occurred at assembly due to the dimensional and
geometric variations in manufacturing phrase.
It is the kinematic variations which result in implicit assembly
functions. As shown in Fig. 13, the kinematic variable F depends on
the variables , R, t and H, and the position of contact points a and
b which are called kinematic joints (dashed rectangles). Kinematic
joints describe motion constraints at the contact points between
mating parts. There are six common joints in 2D assemblies and
twelve common joints in 3D assemblies.
The vectors in a matrix form are arranged in chains or loops
representing the accumulation of variations mentioned above in
vector-loop-based assembly models. Firstly, ignoring the geometric tolerances, the assembly constraints with the vector-loopbased assembly models can be expressed as a concatenation of
homogeneous transformation matrices:
R1 T1 R2 T2 Ri Ti Rn Tn Rf = H

(22)

where Ri is the rotational transformation matrix between the vectors at node i; Ti is the translational matrix of vector i; Rf is the

final closure rotational transformation matrix with the global DRF;

H is the resultant matrix which is equal to the identity matrix for a
closed loop, or the final gap or clearance and its orientation for an
open vector loop.
Eq. (22) describes a series of rotations and translations to transform the local coordinates from vector-to-vector until it has traversed the entire vector loop and returned to the starting point. It
is important to note that the rotational value of Ri is always relative
to the prior vector. It is a positive angle when the rotational direction is same as the prior vector. It is a negative angle, otherwise.
Although smaller than dimensional variations, the accumulation and propagation of geometric feature variations are similar to
dimensional variations. In the vector-loop-based assembly models, geometric tolerances are considered by placing at the contact
point between mating surfaces with zero length vectors having
specified variations or tolerances. In other words, the geometric
tolerance associated with each joint may result in an independent
translational variation or rotational variation or both. It should be
pointed out that the effect of feature variations in 3D depends upon
the joint types and which joint axis you are looking down. Fig. 14
shows the effects of flatness tolerance t at the kinematic joint b
of Fig. 13. Fig. 14(a) represents a translational variation in the xy
plane while Fig. 14(b) represents a rotational variation in the xz
plane. There are three variations in all at this joint where another
rotational variation around the z axis is not shown, which implies
the DOF of kinematic motions and the DOF of feature variations are
mutually exclusive.
All the possible combinations of geometric feature tolerances
with kinematic joint types in 2D space and 3D space can be seen
in [92].
Assuming a geometric feature tolerance is added to joint i,
assembly constraint equation (22) can be rewritten as:

R1 T1 R2 T2 Rig Tig Ri Ti Rn Tn Rf = H

(23)

where Rig Tig is the transformation matrix caused by geometric

tolerance of ith feature.
It is complex to solve Eqs. (22) and (23) because they are
nonlinear equations. But the approximate solution with the DLM
method is accurate enough for tolerance analysis. The first order
Taylors series expansion of assembly constraint equations for a
closed loop can be written as:

1H C = A 1X + B 1U + F 1 = [0] .

(24)

And for an open loop is:

1H O = C 1X + D 1U + G 1

(25)

where 1HC is the vector of clearance variations in a closed loop

and 1HO is the vector of assembly variations in an open loop;
1X is the vector of variations of dimensional variables; 1U is
the vector of variations of assembly variables; 1 is the vector
of variations of geometric feature variables; A and C are the first
order partial derivatives of the dimensional variables in the closed
loop and open loop respectively; B and D are the first order partial
derivatives of the assembly variables in the closed loop and open

10

H. Chen et al. / Computer-Aided Design 53 (2014) 113

loop respectively; F and G are the first order partial derivatives of

the geometric feature variables in the closed loop and open loop
respectively.
Among Eqs. (24) and (25), 1U is obtained by solving these two
equations. For the closed loop, 1U is given in Eq. (26) if B is a fullranked matrix and in Eq. (27) if B is a singular matrix.

1U = B1 A 1X B1 F 1

(26)

1U = (BT B)1 BT A 1X (BT B)1 BT F 1.

(27)

From Eqs. (25) to (27), we can obtain the 1U in the open loop
as:

1U = (C D B1 A) 1X + (G D B1 F ) 1

1 T
1U = C D BT B
B A 1X

T 1 T
+ GD B B
B F 1.

(28)

(29)

Tolerance accumulation of DLM can be estimated with a worst

case way and a statistical way, as shown in Eqs. (30) and (31).
TW =

|Sijd | Tijd +

j =1

|Sij | Tij

(30)

j =1

2
2
m
n

|Sijd | Tijd +
|Sij | Tij
TS =
j =1

(31)

j =1

where Sijd and Sij are sensitivity matrices of dimensional variables

and geometric variables respectively, which are the coefficients
of the 1X and 1 in Eqs. (26)(29); m and n are the number of
dimensional and geometric variables respectively.
Let us continue to take the assembly shown in Fig. 9 as the
example to demonstrate the computational process of DLM. That is
an open loop where the assembly constraint equations are explicit
because no adjustable elements exist. The geometric tolerance t11
involves a rotational variable around the x axis in part 1 which
is equal to t11 /d1 . According to Eq. (23), the resultant vector along
the z axis at point a is l1 + l2 cos() + b sin(). From Eq. (25),
the stack-up variation along the z axis is (t1 + t2 cos())/2 +
(b cos() l2 sin()) .
The second order tolerance analysis (SOTA) method where
the second order Taylors series expansion of the assembly
kinematic constraint equations is taken into account by Monte
Carlo simulation has been developed to enhance the accuracy of
DLM [93].
Because the derivatives of the assembly function with respect
to both the assembly and manufactured variables are more readily
from the vector model, the DLM is more computationally efficient
over other models for tolerance analysis. Owing to the long-term
research by association for the development of computer-aided
tolerancing software (ADCATS), DLM has been applied on CATs
R
successfully, such as CETOL 6 Sigma
.
3. Discussion and comparison
So far, we have listed four 3D tolerance analysis methods particularly based on the literature published over the last three decades
or so. And a simple example has been used to demonstrate the analysis process of these models. Each of them has its advantages and
disadvantages. This section discusses and compares them in detail.
The T-Map based on area coordinates cannot only model all
3D variations of a feature, such as size, orientation and form, but
also model completely and precisely the interactions of them. It is
completely compatible with the ASME/ISO standards for geometric
tolerance and suitable for tolerance synthesis. The size and shape

of the accumulation map are controlled by the dimensions and

shapes of target surfaces in assembly, which gives expression to
the connotation of 3D tolerance analysis. However, the Minkowski
operation for tolerance propagation is not straightforward and
not suitable for computation by hand, and the calculation of
clearance for two planar surfaces or the pin-hole assembly has
been developed in 1D situation only. In other words, T-Map has not
yet been fully developed. A mass of efforts is still needed to study
the algorithms of sensitivities of contributors and their percent
contributions, as well as the statistical arithmetic. Moreover, the
T-Map for axes is four dimensional, which is difficult for illustration
in 3D situation. A better method for the visualization of higher
dimensional maps is needed.
The matrix model uses a displacement matrix to describe the
small displacements of a feature within the tolerance zone and the
clearance between two features. This model, completed by a set of
inequalities defining the bounds of the tolerance zones, reproduces
the measurable or non-invariant displacements associated with
various types of tolerance. It is very efficient for computation
and can be integrated into CAD systems easily. The statistical
method for the matrix model has also been developed which brings
the constraint relations between the translational displacements
and rotational displacements into computation by Monte Carlo
simulation. But the analysis objective and constraint objective are
points, which lead to different results with respect to different
points. The optimization may be difficult when lots of inequalities
are obtained. In addition, it is unable to take the form tolerance into
account.
The unified JacobianTorsor model combines the advantages of
the torsor model which is suitable for tolerance representation and
the Jacobian matrix which is suitable for tolerance propagation.
To overcome the limitations and difficulties of point-based
approaches, the interval arithmetic is brought into the model to
allow tolerance analysis to be performed on a tolerance zone
basis rather than on a point basis. Tolerance analysis of this
model is in a kinematic way because of the concept of torsor.
It is more suitable for the representation and propagation of
clearance in 3D situation. The statistical method and tolerance
allocation for this model have also been studied. Nevertheless, the
constraint relations between the components of torsor need to be
considered to improve the accuracy of results and conform to the
tolerancing standards better. As with the matrix model, the unified
JacobianTorsor model cannot deal with the form tolerance too.
The DLM is based on the first order Taylors series expansion
of the assembly kinematic constraint equations with respect to
both the assembly variables and the manufactured variables in
an assembly. Three main sources of variations, i.e., dimensional
variations and geometric feature variations, as well as kinematic
variations, are distinguished and represented by the vector-loopbased assembly models. Although all types of tolerance can be
modeled, and the results of statistical case and worst case can
be calculated efficiently with the sensitivity matrix, this method
heavily depends on the users expertise and experience to obtain
correct results. More specifically, how to define the joint types and
the effects of geometric variations are dependent of users choices.
Meanwhile, the relationship between the geometric tolerance
and the dimensional tolerance needs continuous optimization to
coincide with the tolerancing standards better.
The difference and comparison between four models with six
items are listed in Table 2. The symbol represents unknown
or unable to calculate based on the published literature. and
 represent point-based and surface-based respectively. w and
represent the variation of target surface along the z axis and around
the x axis in the assembly depicted in Fig. 6 respectively.
The results of four models corresponding to the assembly
shown in Fig. 6 are listed in Table 3 where only the condition of

H. Chen et al. / Computer-Aided Design 53 (2014) 113

11

Table 2
Comparisons of four models.

T-Map
Matrix
Unified JacobianTorsor
DLM

Worst case

Statistical case

Sensitivity and
percent contribution

Geometric tolerance
Form

Orientation

Position

Objective

Application (CATs)

Table 3
Comparisons of the results of four models with a worst case.
Results
T-Map
Matrix
Unified JacobianTorsor
DLM

t1
t1
t1
t1

Residual between 3D and 2D

+ t2 + t11 ((2b + d2 )/d1 1)

+ t2 + 2b cos(t11 /d1 ) sin(t11 /d1 ) + 2l2 (cos(t11 /d1 ) sin(t11 /d1 ) 1)
+ t2 + 2bt11 /d1
+ t2 cos(t11 /d1 ) + 2(b cos(t11 /d1 ) l2 sin(t11 /d1 )) t11 /d1

b > (d1 d2 )/2 is considered. It should be noted that the result of

the matrix model is obtained by ignoring the constraint inequalities because the symbolic parameters impede the optimization.
Residual gaps between 3D and 2D which is t1 + t2 are approximated and simplified. As can be seen, the residuals reflect the property of 3D tolerance which takes the structure of assembly and the
geometric tolerance into consideration. This is a significant difference of tolerance analysis between 3D and traditional 1/2D methods. The result of T-Map depends on not only the shapes of target
surface but also the diameters of bars in the assembly. Because the
rotational variation aroused by the orientation tolerance t11 is very
small, the results of the matrix model and DLM can be approximated and simplified as ((t1 + t2 )/2 + bt11 /d1 ), which is the result
of the unified JacobianTorsor model.
The rotational displacement can be obtained by the TMap model and the unified JacobianTorsor model because these
two models are surface-based approaches. More specifically, the
accumulation maps illustrated in Fig. 7 express both translational
and rotational variations of functional surface. Similarly, Eq. (21)
of the unified JacobianTorsor model includes all of translational
and rotational variations of functional surface. However, the result
of the matrix model is the variation of point a on functional
surface. The pose of a point in space is described by position
rather than orientation. If another point of functional surface is
selected as the target objective, the result may be different. So does
the DLM. Compared with point-based approaches, the advantages
of surface-based models include: (1) the results reflect not only
position but also orientation of the analytic objective clearly in 3D
space; (2) the variations caused by selection of different points are
avoided.
4. Conclusion
The discussion and comparison have been given in the previous
section where our subjective judgment comes into play. More
details about these models, as well as the differences between
them are left to the readers. Generally, each of these tolerance
analysis models has its own strengths and weaknesses, and it is
up to the users to make the wise choice according to the analysis
objective and condition.
We would like to conclude this paper by presenting a perspective overview of the future research about 3D tolerance analysis
which is considered to be challenging but promising.
(1) Working conditions of the assembly, such as deformation because of force and temperature, and variation of joint due to
lubricating medium, need to be taken into account to improve
the reliability. The combustion engine is a typical example,

2bt11 /d1 + (d2 /d1 1)t11

2bt11 /d1
2bt11 /d1
2bt11 /d1

which endures high temperature and high pressure and forced

liquid lubrication when it works. There is no doubt that the geometric feature and joint (clearance) are different to the static
condition. Pierre et al. [94] integrate the thermomechanical
strains into the SDT model and Zhang et al. [95] take the working condition into the unified JacobianTorsor model, which
can be seen as the preliminary work. The interaction between
the temperature and force and lubrication aggravates the difficulties of the analysis process.
(2) The constraints between translational and rotational vectors of
3D tolerance analysis models should be optimized further in
order to conform to the tolerancing standards better and enhance the accuracy, because they depend on each other in the
tolerance zone. According to the envelope principle, and
must shrink to zero when u arrives at its limited position for a
planar feature in the tolerance zone, as shown in Fig. 8. Therefore, it should be wrong to take both the limited values of translation and rotation into computation simultaneously. The unified JacobianTorsor model and DLM cannot deal with these
constraints. Other 3D models which are not discussed in detail
in this paper also need to consider the constraints.
(3) The mathematic models of tolerance analysis, especially the
tolerance propagation for form tolerance and runout tolerance
which includes circular runout and total runout, still need a
large deal of research. The position and orientation of form tolerance are random in a tolerance zone (see Fig. 5(a)), which
means the form tolerance is not deterministic in tolerance
propagation. Meanwhile, the role of form tolerance heavily depends on the users expertise and experience, which is illustrated in DLM. Approximately, a runout tolerance can be seen
as the combination of form tolerance and positional tolerance,
which is more complex than form tolerance.
(4) The existing 3D models focus on connections in series mainly.
The solution of tolerance representation and propagation for
parallel connections in the assembly will greatly reduce the
gap between 3D models and reality. F shown in Fig. 13 depends on the variables , R, t and H, and the transfer route
which passes point a or point b or both. Most often, a series route passing one of two points is selected for tolerance
propagation. However, there are two routes that participate
in tolerance propagation actually. It is complicated for parallel
connections because there are interactions between them.
(5) The sensitivity and percent contribution of tolerance are very
useful for tolerance optimization, especially in a statistical
case. Some of 3D tolerance analysis methods, such as T-Map
and the matrix model, still lack suitable algorithm to calculate them. Moreover, for a feature specified by more than one

12

H. Chen et al. / Computer-Aided Design 53 (2014) 113

tolerance, the algorithm for separating and calculating each of

them has not been presented by far.
Acknowledgments
The work described in this paper is supported in part by
grants from the National Natural Science Foundation of China
(Grant Nos. 51121063, 51175340) and the National Science &
Technology Pillar Program during the 12th Five-year Plan Period
(Grant No. 2012BAF06B03). The authors are grateful for these
financial supports.
References
[1] Dizioglu B, Lakshiminarayana K. Mechanics of form closure. Acta Mech 1984;
52:10718.
[2] Takezawa N. An improved method for establishing the process wise quality
standard. Rep Stat Appl Res Union Japan Sci Eng (JUSE) 1980;27(3):6376.
[3] Liu SC, Hu SJ. An offset element and its application in predicting sheet metal
assembly variation. Int J Mach Tools Manuf 1995;35(11):154557.
[4] Liu SC, Hu SJ. Variation simulation for deformable sheet metal assembly using
finite element methods. Trans ASME, J Manuf Sci Eng 1997;119:36874.
[5] Chase KW, Magleby SP, Gao J. Tolerance analysis of 2-D and 3-D mechanical
assemblies with small kinematic adjustments. In: Zhang HC, editor. Advanced
tolerancing techniques. New York: John Wiley & Sons; 1997. p. 10337.
[6] Prisco U, Giorleo G. Overview of current CAT systems. Integr Comput-Aided
Eng 2002;9:37387.
[7] Salomons OW, Houten FJAM, Kals HJJ. Current status of CAT systems.
In: ElMaraghy HA, editor. Geometric design tolerancing: theories, standards
and applications. London: Chapman & Hall; 1998. p. 43852.
[8] Shen Z. Tolerance analysis with EDS/VisVSA. ASME J Comput Inf Sci Eng 2003;
3:959.
[9] Islam MN. Functional dimensioning and tolerancing software for concurrent
engineering applications. Comput Ind 2004;54:16990.
[10] Hillyard RC, Braid IC. Analysis of dimensions and tolerances in computer-aided
mechanical design. Comput-Aided Des 1978;10(3):1616.
[11] Lin VC, Light RA, Gossard DC. Variational geometry in computer-aided design.
Comput Graph 1981;15(3):1717.
[12] Light RA, Gossard DC. Modification of geometric models through variational
geometry. Comput-Aided Des 1982;14(4):20914.
[13] Requicha AAG. Toward a theory of geometric tolerancing. Int J Robot Res 1983;
2(4):4560.
[14] Requicha AAG. Representation of tolerances in solid modeling: issues and
alternative approaches. In: Pickett MS, Boyse WJ, editors. Solid modeling
by computers: from theory to applications. New York: Plenum Press; 1984.
p. 322.
[15] Jayaraman R, Srinivasan V. Geometric tolerancing: I. Virtual boundary
requirement. IBM J Res Dev 1989;33(2):90104.
[16] Srinivasan V, Jayaraman R. Geometric tolerancing: II. Conditional tolerance.
IBM J Res Dev 1989;33(2):10522.
[17] Turner JU, Wozny MJ. A mathematical theory of tolerance. In: Wozny MJ,
Mclaughlim JL, editors. Geometric modeling for CAD applications. Elsevier
Science Publishers, IFIP; 1988. p. 16387.
[18] Turner JU. A feasibility space approach for automated tolerancing. Trans ASME,
J Eng Ind 1993;115(3):3416.
[19] Wirtz A. Vectorial tolerancing a basic element for quality control. In:
Proceedings of the 3rd CIRP seminar on computer aided tolerancing. 1993.
p. 11528.
[20] Laperrire L, Lafond P. Modeling tolerances and dispersions of mechanical
assemblies using virtual joints. In: CD-ROM proceedings of 25th ASME design
automation conference. 1999.
[21] Bernstein N, Preiss K. Representation of tolerance information in solid models.
In: Proceedings of 15th ASME design automation conference, September 1989.
DE-Vol. 19-1. p. 478.
[22] Zhang BC. Geometric modeling of dimensioning and tolerancing. Ph.D. Thesis.
Arizona State University. 1992.
[23] Kramer GA. Solving geometric constraint systems: a case study in kinematics.
MIT Press; 1992.
[24] Davidson JK, Mujezinovi A, Shah JJ. A new mathematical model for geometric
tolerances as applied to round faces. ASME J Mech Des 2002;124:60922.
[25] Mujezinovi A, Davidson JK, Shah JJ. A new mathematical model for geometric
tolerances as applied to polygonal faces. ASME J Mech Des 2004;126:50418.
[26] Desrochers A, Clment A. A dimensioning and tolerancing assistance model for
CAD/CAM systems. Int J Adv Manuf Technol 1994;9:35261.
[27] Portman VT. Modelling spatial dimensional chains for CAD/CAM applications.
In: F. Kimura, editor. Proceedings of the 4th CIRP design seminar on computeraided tolerancing. 1995. p. 7185.
[28] Cardew-Hall MJ, Labans T, West G, Dench P. A method of representing
dimensions and tolerances on solid based freeform surfaces. Robot ComputAided Manuf 1993;10:22334.
[29] Desrochers A, Rivire A. A matrix approach to the representation of tolerance
zones and clearances. Int J Adv Manuf Technol 1997;13:6306.

[30] Whitney DE, Gilbert OL, Jastrzebski M. Representation of geometric variations

using matrix transforms for statistical tolerance analysis in assemblies. Res
Eng Des 1994;6(4):191210.
[31] Clment A, Desrochers A, Rivire A. Theory and practice of 3D tolerancing for
assembly. In: Proceedings of the CIRP seminar on computer aided tolerancing.
USA: Penn State University; 1991.
[32] Clment A, Rivire A. Tolerancing versus nominal modeling in next generation
CAD/CAM system. In: Proceedings of the CIRP seminar on computer aided
tolerancing. 1993. p. 97113.
[33] Teissandier D, Coutard Y, Grard A. Three-dimensional functional tolerancing
with proportioned assemblies clearance volume (U.P.E.L: Unions pondres
dspaces de libert). In: Proceedings of the 1996 engineering systems design
and analysis conference presented at the 3rd Biennial joint conference on
engineering systems design and analysis, ESDA (ASME, Petroleum Division),
PD-Vol. 80.8. P. 12936.
[34] Teissandier D, Coutard Y, Grard A. A computer aided tolerancing model:
proportioned assembly clearance volume. Comput-Aided Des 1999;31(3):
80517.
[35] Fortini ET. Dimensioning for interchangeable manufacture. New York:
Industrial Press; 1967.
[36] Evans DH. Statistical tolerancing: state of the art, part 2. Methods of estimating
moments. J Qual Technol 1975;7(1):112.
[37] Evans DH. Statistical tolerancing: state of the art, part 3. Shift and drifts. J Qual
Technol 1975;7(2):716.
[38] Evans DH. An application of numerical integration techniques to statistical
tolerancing. Technimetrics 1967;9(3).
[39] Evans DH. An application of numerical integration techniques to statistical
tolerancing, IIa note on the error. Technimetrics 1971;13(2).
[40] Evans DH. An application of numerical integration techniques to statistical
tolerancing, IIIgeneral distributions. Technimetrics 1972;14(1).
[41] Parkinson DB. The application of reliability method to tolerancing. ASME J
Mech Des 1982;104:6128.
[42] Parkinson DB. Reliability indices employing measures of curvature. Reliab Eng
1983;15379.
[43] Taguchi G. Performance analysis design. Int J Prod Des 1978;16:52130.
[44] DErrico JR, Zaino NA. Statistical tolerancing using a modification of Taguchis
method. Technometrics 1988;30(4):397405.
[45] DeDoncker D, Spencer A. Assembly tolerance analysis with simulation and
optimization techniques. SAE Trans 1987;96(1):10627.
[46] Grossman DD. Monte Carlo simulation of tolerancing in discrete parts
manufacturing and assembly. Technical report STAN-CS-76-555. Computer
Science Dep., Stanford University, USA; 1976.
[47] Fleming A. Geometric relationships between toleranced features. Artif Intell
1988;37:40312.
[48] Rivest L, Fortin C, Morel C. Tolerancing a solid model with a kinematic
formulation. Comput-Aided Des 1994;26:46576.
[49] Chase KW, Gao J, Magleby SP. General 2-D tolerance analysis of mechanical
assemblies with small kinematic adjustments. J Des Manuf 1995;5(4):26374.
[50] Gao J, Chase KW, Magleby SP. General 3-D tolerance analysis of mechanical
assemblies with small kinematic adjustments. IIE Trans 1998;30:36777.
[51] Lafond P, Laperrire L. Jacobian-based modeling of dispersions affecting
predefined functional requirements of mechanical assemblies. In: Proceedings
of IEEE international symposium on assembly and task planning. 1999.
[52] Laperrire L, EIMaraghy HA. Tolerance analysis and synthesis using Jacobian
transforms. CIRP AnnManuf Technol 2000;49(1):35962.
[53] Jin J, Shi J. State space modeling of sheet metal assembly for dimensional
control. Trans ASME, J Manuf Sci Eng 1999;121:75662.
[54] Mantripragada M, Whitney DE. Modeling and controlling variation propagation in mechanical assemblies using state transition models. IEEE Trans Robot
Autom 1999;15(1):12440.
[55] Cai W, Hu SJ, Yuan JX. A variational method of robust fixture configuration
design for 3D workpieces. Trans ASME, J Manuf Sci Eng 1997;119(4A):
593602.
[56] Desrochers A. A CAD/CAM representation model applied to tolerance transfer
methods. ASME J Mech Des 2003;125:1422.
[57] ASME Y14.5-2009 dimensioning and tolerancing. New York: ASME; 2009.
[58] ISO 1101: Geometric Product Specifications (GPS)-Geometrical tolerancingtolerancing of form, orientation, location and run-out. 2004.
[59] Hong YS, Chang TC. A comprehensive review of tolerancing research. Int J Prod
Res 2002;40(11):242559.
[60] Zhu D, Cheng HS, Arai T, Hamai K. A numerical analysis for piston skirts in
mixed lubrication-part I: basic modeling. ASME J Tribol 1992;114:55362.
[61] Zhu D, Hu YZ, Cheng HS, Arai T, Hamai K. A numerical analysis for piston skirts
in mixed lubrication-part II: deformation considerations. ASME J Tribol 1993;
115:12533.
[62] Desrochers A, Ghie W, Laperrire L. Application of a unified JacobianTorsor
model for tolerance analysis. ASME J Comput Inf Sci Eng 2003;3:113.
[63] Marziale M, Polini W. A review of two models for tolerance analysis of an
assembly: vector loop and matrix. Int J Adv Manuf Technol 2009;43:110623.
[64] Laperrire L, Desrochers A. Modeling assembly quality requirements using
Jacobian or screw transforms: a comparison. In: Proceedings of the 4th IEEE
international symposium on assembly and task planning, Soft Research Park.
May 2011. p. 3306.
[65] Marziale M, Polini W. A review of two models for tolerance analysis of an
assembly: Jacobian and Torsor. Int J Comput Integr Manuf 2011;24(1):7486.
[66] Ameta G, Serge S, Giordano M. Comparison of spatial math models for
tolerance analysis: tolerance-maps, deviation, domain, and TTRS. ASME J
Comput Inf Sci Eng 2011;11: 012004-18.

H. Chen et al. / Computer-Aided Design 53 (2014) 113

[67] Shah JJ, Ameta G, Shen Z, Davidson J. Navigating the tolerance analysis maze.
Comput-Aided Des Appl 2007;4(5):70518.
[68] Salomons OW, Poerink HJJ, Haalboom FJ, Slooten FV, et al. A computer aided
tolerancing tool I: tolerance specification. Comput Ind 1996;31:16174.
[69] Salomons OW, Haalboom FJ, Poerink HJJ, Slooten FV, et al. A computer aided
tolerancing tool II: tolerance analysis. Comput Ind 1996;31:17586.
[70] Shen Z, Ameta G, Shah JJ, Davidson JK. A comparative study of tolerance
analysis method. ASME J Comput Inf Sci Eng 2005;5:24756.
[71] Davidson JK, Shah JJ. Geometric tolerances: a new application for line
geometry and screws. In: Proceedings of A symposium commemorating the
legacy, works, and life of Sir Robert Stawell ball upon the 100th anniversary of
a treatise on the theory of screws. University of Cambridge; 2000.
[72] Bhide S, Davidson JK, Shah JJ. A new mathematical model for geometric
tolerances as applied to axes. In: Proceedings of DETC03, ASME 2003
design engineering technical conference and computers and information in
engineering conference. September 2003. p. 19.
[73] Bhide S, Ameta G, Davidson JK, Shah JJ. Tolerance-maps applied to the
straightness and orientation of an axis. In: Models for computer aided
tolerancing in design and manufacturing. 2007. p. 4554.
[74] Ameta G, Davidson JK, Shah JJ. The effects of different specifications on
the tolerance-maps for an angled face. In: Proceedings of ASME/DETC-2004.
September, 57199.
[75] Ameta G, Davidson JK, Shah JJ. Tolerance-maps applied to a point-line cluster
of features. ASME J Mech Des 2007;129:78292.
[76] Ameta G, Davidson JK, Shah JJ. Generate frequency distributions of clearance
and allocate tolerances for pin-hole assemblies. ASME J Comput Inf Sci Eng
2007;7:34759.
[77] Ameta G, Davidson JK, Shah JJ. Influence of form on tolerance-map-generated
frequency distributions for 1D clearance in design. Precis Eng 2010;34:227.
[78] Ameta G, Davidson JK, Shah JJ. Form tolerance on the frequency distributions
of clearance between two planar faces. ASME J Comput Inf Sci Eng 2011;11:
011002-110.
[79] Herv JM. Analysis structurelle des mcanismes par groupe des dplacements.
Mech Mach Theory 1978;13:43750.
[80] Desrochers A, Delbart O. Determination of part position uncertainty within
mechanical assembly using screw parameters. In: EIMaraghy HA, editor.
Geometric design tolerancing: theories, standards and applications. 1998.
p. 18796.

13

[81] Clment A, Rivire A, Serr P, Valade C. The TTRSs: 13 constraints for

dimensioning and tolerancing. In: EIMaraghy HA, editor. Geometric design
tolerancing: theories, standards and applications. 1998. p. 12231.
[82] Clment A, Rivire A, Serr P. A declarative information model for functional
requirements. In: Computer-aided tolerancing. 1996. p. 316.
[83] Salomons OW. Computer support in the design of mechanical products. Ph.D.
Thesis. University of Twente, The Netherlands; 1995. p. 21142.
[84] Clment A, Bourdet P. A study of optimal-criteria identification based on the
small-displacement screw model. CIRP Ann 1988;37(1):5036.
[85] Bourdet P, Marhieu L, Lartigue C, Ballu A. The concept of the small displacements torsor in metrology. In: International euroconference, advanced mathematical tools in metrology. 1995.
[86] Desrochers A. Modeling three dimensional tolerance zones using screw
parameters. In: Proceedings DETC: 25th design automation conference.
September 1999. p. 895903.
[87] Laperrire L, Ghie W, Desrochers A. Statistical and deterministic tolerance
analysis and synthesis using a unified JacobianTorsor model. CIRP Ann
Manuf Technol 2002;51(1):41720.
[88] Ghie W. Statistical analysis tolerance using Jacobian Torsor model based on
uncertainty propagation method. Int J Multiphys 2009;3(1):1130.
[89] Ghie W, Laperrire L, Desrochers A. Statistical tolerance analysis using the
unified JacobianTorsor model. Int J Prod Res 2010;48(15):460930.
[90] Ghie W, Laperrire L, Desrochers A. Re-design of mechanical assembly using
the unified JacobianTorsor model for tolerance analysis. In: Models for
computer aided tolerancing in design and manufacturing. 2007. p. 95104.
[91] Desrochers A. Geometrical variations management in a multi-disciplinary
environment with the JacobianTorsor model. In: Models for computer aided
tolerancing in design and manufacturing. 2007. p. 7584.
[92] Chase KW, Gao J, Magleby SP, Sorensen CD. Including geometric feature
variations in tolerance analysis of mechanical assembly. IIE Trans 1996;28(10):
795807.
[93] Glancy CG, Chase KW. A second order method for assembly tolerance analysis.
In: Proceedings of 1999 ASME design engineering technical conferences. 1999.
DETC99/DAC-8707.
[94] Pierre L, Teissandier D, Nadeau JP. Integration of thermomechanical strains
into tolerancing analysis. Int J Interact Des Manuf 2009;3:24763.
[95] Zhang W, Chen C, Li P, et al. Tolerance modelling in actual working condition
based on JacobianTorsor theory. Comput-Integr Manuf Syst 2011;17(1):
7783.