You are on page 1of 31

The International Journal of Advanced Manufacturing Technology

https://doi.org/10.1007/s00170-018-1920-2

ORIGINAL ARTICLE

A comprehensive review of tolerance analysis models


Yanlong Cao 1,2 & Ting Liu 2 & Jiangxin Yang 2

Received: 6 November 2017 / Accepted: 15 March 2018


# Springer-Verlag London Ltd., part of Springer Nature 2018

Abstract
During the design and manufacturing process, the dimensional and geometric tolerances assigned to parts will affect the required
functionality through a stack-up of deviations. Choosing an appropriate tolerance analysis model is important to calculate the
influence that every tolerance has on key characteristics, and these models are fundamental tools for shortening the product
development cycle with improved quality at a lower cost. There are many tolerance analysis models proposed in the literature.
This paper, relying on the sizable literature, briefly presents eight of the most widely used models for tolerance analysis. Through
a study of the research status of the analysis methods and applications, a comparison is proposed to show each method’s
advantages and disadvantages, similarities, and differences. Such a comparison furnishes criteria which are helpful in choosing
the most appropriate model under various circumstances, as well as improving the accuracy of analysis.

Keywords Tolerance analysis . Tolerance models . Key characteristics . Research status . Comparison

1 Introduction different targets are considered. An assembly with the re-


striction of specified dimensional and geometric tolerances
The tolerances allocated for individual part dimensions, form, conforming to quality requirements or inversely allocation of
position, and direction in the design stage will transfer and individually optimal tolerances is implemented in design
accumulate in the process of mechanical assembly to severely while tolerance transfer, that is, taking into process planning
impact the final product performance and cost. Thus, tolerance and the complex variables including natural variability and
analysis is urgently needed to predict the accumulated effects unavoidable random or systematic variability is conducted in
of these individual tolerances on the functional requirement of manufacturing, as well; error calculation or evaluation is
the assembly. The focus of tolerance analysis is conditional proceeded based on data measured from coordinate measur-
upon the adopted mathematical or physical models that de- ing machines (CMMs) in inspection stage through algo-
scribe the variations of the tolerance zone and the accepted rithms, i.e., least squares and minimum zone method, to
analysis methods that assess the effect of part deviations on estimate the quality and thereby adjust the parameters in
the product functional requirements. design phase to achieve optimization. Testing or experiment
A study of tolerance analysis performed for mechanical is conducted to verify the effectiveness of the proposed
assembly can be divided into five aspects for the most part, methods in the functional analysis. It should be noted that
as shown in Fig. 1: the analysis stage, analysis principle, a visualization technology is important to show the results in
analysis object, analysis method, and final analysis state. inspection and testing stages intuitively, and then to conform
First, in the different stages of the product life cycle, to the design requirements. Second, the process of tolerance
analysis must meet certain principle; the division of indepen-
dent principle (IP) and the relevant requirements balances
* Jiangxin Yang the correlation of dimensional tolerances and geometric tol-
yangjx@zju.edu.cn erances, i.e., the maximum material requirement (MMR)
1
controls the dimensional precision, and the actual contour
State Key Laboratory of Fluid Power and Mechatronic Systems,
College of Mechanical Engineering, Zhejiang University,
satisfies the maximum material virtual boundary (MMVB).
Hangzhou 310027, China The choice for the tolerance zones’ interactions and the da-
2
Key Laboratory of Advanced Manufacturing Technology of
tum procedure deals with the tolerance zones and datums
Zhejiang Province, College of Mechanical Engineering, Zhejiang separately: the former schematizes more tolerances applied
University, Hangzhou 310027, China to the same features; the latter covers an assigned
Int J Adv Manuf Technol

Analysis
Object

Part Rigid

Assembly Flexible
Analysis Point-
Stage based
Feature-
Tolerance based
Design
Manufac
Conformance
turing
3
Analysis State
Tesng 1 Tolerance 5
Analysis Stac Sequence
Inspecon Time
Visualizaon
discrete
2 Dynamic Kinemac
Relave
4 Deviaons
posion

Analysis
Principle
Analysis
method
Material Linear/nonlin Explicit/im
Envelope/in condion 1D/2D/3D
ear plicit
modifier(MMC
dependence
/LMC)
Worst case/stascal Fuzzy/non-probabilisc

Tolerance
Datum
zone
precedure Assemblability Funconal
Sensivity analysis
interacon analysis
Interference
Sasfying the Sensivity/contrib
detecon/assemb
funconal uon/distributed
ly yield
requirements visualizaon

Fig. 1 The various aspects of tolerance analysis

precedence among the datums. Third, a single part or assem- dimensionality and objective function characteristics, i.e.,
blies are two analytical objects. At the part level, the design- linear or non-linear, explicit or implicit. Two main ap-
er specifies tolerances and expresses the corresponding tol- proaches, worst case (deterministic case) and statistical case,
erance zones. Conversely, at the assembly level, clearances are applied. The worst-case method relies on the complete
and fit are specified. The functional requirement is pursued interchangeability of the parts. Statistical tolerance methods
for analyzing the range of the admissible play. The differ- are mainly referred as the root sum squares (RSS) method
ence between rigidity and flexibility decides the model to be and Monte Carlo method, of which variations are probability
selected because some models are based on a rigid body distribution assumed. Fuzzy and non-probabilistic are de-
hypothesis, i.e., T-Map model, which fails to represent flex- scribed as the bounds of the parameters without distribution
ibility, whereas other models, such as the skin model shape requirement. By these means, tolerance analysis, i.e., assem-
and deviation domain (DD), are not limited to a rigid body bling analysis, functional analysis, and sensitivity verifica-
that available to model a flexible part or assembly as well tion, can be achieved. Finally, compared to conventional
[1]. These models can either be point based or feature based. static analysis, tolerance analysis of system in motion is
Compared to a feature that characterizes position and direc- accomplished by performing the kinematic analysis at differ-
tion information at the same time, the pose of a point in ent discrete positions considering several kinds of devia-
space is described by position rather than orientation, with tions, i.e., manufacturing deviation, deformation [3], joint
variations that vary along with the selection of different clearance [4], and deviation interactions [5], as well as the
points [2]. Fourth, according to the requirements for func- assembly sequences to obtain different resulting part posi-
tion, performance, and appearance, combined with tions and time-dependent function requirements [6].
Int J Adv Manuf Technol

Many tolerance analysis models have been presented in the geometric variations along the product lifecycle [51]. As a
published literature, as shown in Table 1. Boyer et al. [18] kernel concept of GeoSpelling, the skin model was developed
initially introduced a mathematical definition of variational for interfacing between a workpiece and its environment
space, followed by a variational solid modeling with paramet- [52–54]. To avoid the shortcoming of infinite points, the skin
ric modeling of the range of the tolerance zone [19]. A virtual model shape was developed for its convenience in computer
boundary [20–22] describes the part within the deviation of expression [55, 56]. On the basis of the models addressed
the extreme boundary. Desrochers et al. [23, 24] use elemen- above, a series of extended models were derived. For exam-
tary surfaces (spheres, cylinders, planes, etc.) to model the six ple, the combination of Jacobian and torsor models brought
lower kinematic pairs, that is the so-called topologically and about the unified Jacobian–torsor model with the advantages
technologically related surfaces (TTRS). The matrix uses a of tolerance representation and error propagation simulta-
displacement matrix to describe the roto-translational varia- neously [57, 58]. The modal interval and small degrees of
tion of a feature [25–28]. Three main sources of variations in freedom (MI-SDOF) model the size and tolerance with a small
a mechanical assembly are included in the vector loop-based degree of freedom and its induced uncertainty is denoted by a
model, in which each vector represents a corresponding com- modal interval [59]. As an extension of the torsor model, the
ponent dimension [29–32]. A semantic model is established proportioned assembly clearance volume (PCAV) weighs the
with logical quantifiers. The relationship between tolerance relative position of the assembly and the ideal surface with a
specifications and tolerance stacking is embedded in this small displacement torsor [13]. A comprehensive model com-
mathematic model [33–37]. A gap-based approach, called bining deviation propagation and small displacement torsor
GapSpace [38, 39], and the further Revised GapSpace [9] (DP-SDT) theory is efficient to realize tolerance analysis and
use a gap to describe the relationship between features belong- optimization [14]. The well-known tolerance models and
ing to different components. Davidson et al. [40–43] proposed some derived models are summarized in Table 1.
the T-Map model with the one-to-one mapping of points in a The tolerance analysis methods are mainly worst case or
hypothetical Euclidean volume, which can represent all the statistical case. A deterministic approach assumes that each
variational possibilities of a feature. Similar to the T-Map component dimension is at its maximum or minimum limit
model, the deviation domain considers the maximum dis- and the statistical methods are based on the hypothesis that
placement of all border points of the toleranced features which the tolerances follow a certain distribution. For a statistical
can represent all 3D variations that are allowed for certain approach, Chase et al. [60] maintained a root sum squared
features. The torsor model characterizes tolerance zones by a (RSS), assuming a normal or Gaussian distribution for the
set of screw parameters [44, 45]. The Jacobian model regards component variations. Taguchi aimed to choose the group
a substitute surface as functional element, and the kinematic with the most suitable experimental factors [61, 62].
chains formulate the displacement matrices [46, 47]. A modal Ilumoka et al. [63] adopted a Taylor series expansion meth-
model, which generates a basis for characteristic defects to od. Convolution can acquire any requisite convolution
relate the measured defects with the origin surface [48–50], density function [64]. Evans employed a numerical inte-
has an advantage in form error representation. GeoSpelling is gration method which is correct for an implicit assembly
proposed as a coherent and complete model to manage the function [65]. In particular, the Monte Carlo (MC) method

Table 1 Tolerance models


Models References Derived models

Variation model Boyer et al. (1991), Gupta et al. (1993) Combine with skin model [7]
TTRS Charles et al. (1989), Clément et al. (1991) FTTRS [8]
Matrix Clément et al. (1994), Desrochers (1994, 1997) –
Semantic Wang (2006, 2007, 2008) –
GapSpace Morse et al. (2000), Zou et al. (2003, 2004) Revised GapSpace [9]
T-Map Davidson et al. (2002, 2003), Bhide et al. (2003) M-Map [10]
Deviation domain Giordano et al. (1993, 1999, 2001) Combine with Modal [11]
Polytopes Teissandier et al. (1999) Combine with screws [12]
Torsor Bourdet et al. (1976, 1988, 1996), PCAV [13]; DP-SDT [14];
Clément et al. (1991) combine with skin model [15]
Jacobian Salomons et al. (1996), Laperriére et al. (1999) Combine with skin model [15]
Jacobian–torsor Laperriére et al. (2002, 2003), Ghie et al. (2003) Combine with CLIC [16]
Vector loop Chase et al. (1995, 1996), Gao et al. (1993, 1996) Combine with skin model [7]
Skin model Ballu et al. (1993), Mathieu et al. (2007) Skin model shapes [17]
Int J Adv Manuf Technol

is the most widely used method in statistical tolerance anal- Case study is provided in Sect. 3. Section 4 compares these
ysis with an ability to solve both linear and non-linear eight models with respect to several criteria. Section 5 pro-
assemblies, but there is a shortage in large computation vides a prospective review of the potential research in the
[66, 67]. To avoid this, a number theoretical net (NT-net) field of tolerance analysis.
was developed to reduce the computing effort and the var-
iability of the MC’s results for generating a good lattice
point (GLP) has a better convergent rate [68, 69]. The
abovementioned methods prevail in multi-dimensional 2 Tolerance analysis models
cases with general (i.e., Gaussian) distributions. In some
cases, it might be unknown or difficult to obtain the design Tolerance analysis models can be divided into a two-level
parameter information; therefore, fuzzy methods [70, 71] topological relationship, that is, a part level and an assembly
and non-probabilistic analysis are suggested because they level. The established models necessarily satisfy the compat-
do not need the presumed probabilistic information [72, ibility and computability criteria, which require not only meet-
73]. ing ISO/ASME standards but also possibly representing all
There is a significant amount of review in the literature tolerances including size, form, location, direction, and run-
that makes a lot contribution to the tolerance analysis out. They are also simultaneously and ably integrated into
research. Some pay particular emphasis on one elaborate CAT systems to realize the extraction of dimension chains or
model, i.e., skin model shape in [74, 75] and T-Map in feature information, therefore, automatically implementing
[76], or do a more detailed contrast of two or three famil- tolerance analysis and optimization.
iar models resembling in modeling theory and calculation
method [1, 77–81], analysis on several common-used
models [2, 82–84], some well-defined solution technolo- 2.1 TTRS model
gies of analysis methods [85, 86], and overview of current
CAT software in different nested tolerance models [87, The technologically and topologically related surfaces
88]. Hong et al. [89] reviewed on various tolerancing (TTRS) model is a typical representative of the degree of
issues in design and manufacturing. Polini described the freedom model that utilizes several different concepts from
most common criteria used to categorize existing models constraints and rigid body motions. Desrochers et al. [24,
for tolerance analysis [90]. However, there still lacks a 91] initially formulated the idea of using dimensioning and
body of conclusive literature in which the hot tolerance tolerancing. In this model, the tolerance is decided by the
analysis approaches are compared systematically and ob- correlation between 7 unit surfaces and 28 surface-related sit-
jectively regarding the status quo and in applications that uations. The two surfaces of the same part are called TTRS,
show their advantages, disadvantages, similarities, and while surfaces in contact belonging to different parts are called
differences. This paper aims to conduct a comprehensive Pseudo-TTRS. The tolerance expression process starts from
summarization and complete comparison of eight toler- the extraction of the necessary information from the CAD
ance analysis models, which cover a wide variety of system and arranges each surface in the form of a binary tree,
existing tolerance analysis approaches, i.e., screw that is, a TTRS binary tree. Then, it constructs its minimum
theory-based tolerance analysis models such as TTRS geometric datum element (MGDE) with 13 kinds of relative
model, matrix model, and Jacobian–torsor model (which constraints between a point, line, and plane [92]. Finally, the
combines the Jacobian model and torsor model); deviation different MGDE and their mutual relationships determine the
space-based models such as deviation domain and T-Map; type of tolerances. The TTRS model can effectively display
kinematic-based model, gap-based model; and point design specifications and process plan specifications [93]. The
cloud-based model. These models have been most widely available application of this model in industrial cooling water
used, extensively investigated, and consistently pursued pumps has shown the global consistency of tolerances in a
by the researchers in recent decades. A comprehensive practical industrial environment [94].
review of these models in the state-of-the-art research will The TTRS model adopted for tolerance analysis for assem-
facilitate a generalized acknowledgement of the tolerance bly can be performed using the following steps:
analysis domain and offer a convenient selection of the
most appropriate models to satisfy specific tolerance anal- 1. Create TTRS graph for the assembly with identified
ysis goals. MGDE surfaces.
The rest of the paper is organized as follows: Sect. 2 2. Create geometric tolerance TTRS graph combined with
introduces eight widely applicable tolerance analysis TTRS datum, tolerance type, and tolerance zone.
models and briefly presents each with respect to analytical 3. The toleranced TTRS builds tolerances based on torsors/
procedure, research status, and strengths and weaknesses. screws or matrices and several internal parameters.
Int J Adv Manuf Technol

4. Combine the torsor/screw parameters or matrix limits of Specifically, due to the fact that a point in space is depicted by
all the toleranced TTRS along the stack path/paths in the position without taking direction information into account,
assembly. therefore different matrix models may be obtained as different
5. Use the target features’ tolerance TTRS, identifying the points are selected.
limiting parameter directions. Every matrix model D for a feature can be described as the
6. The sum of all the parameters in the combined torsors/ product of six homogeneous transform matrices.
screws or matrices along the limiting parameter directions
will yield the final deviation for the target feature.
2 D ¼ T w  T v  T u  Rw  Rβ  Rα 3
CγCβ −SγCα þ CγSβSα SγSα þ CγSβCα u
The TTRS-based tolerance analysis has been demonstrated 6 SγCβ
6 CγCα þ SγSβSα −CγSα þ SγSβSα v77
applicable for series and parallel assemblies by means of worst 4 −Sβ SαCβ CαCβ w5
case (WC) or statistics methods along the specific SDT pa-
0 0 0 1
rameters’ axes. A genetic algorithm is employed to solve non-
linear objective functions and non-linear constraints [95]. A ð1Þ
modified TTRS model can automatically generate all the where C is the abbreviation of cos () and S is sin ().
GD&T in the manufacturing stage and is concurrently suitable This 4 × 4 homogeneous matrix is the general expression of
for tolerance analysis and allocation [96]. This model can deal the model. According to the division of seven elementary sur-
with the issue of datum precedence, and it may solve tolerance face types in the TTRS model and their different invariant de-
interactions as revealed in the ASME and ISO standards by grees, Eq. (1) can be correspondingly calculated. For each fea-
constraints of torsors or matrices. In using the TTRS model to ture, different tolerance types may result in the same matrix
reorganize 3D geometry information provided by a CAD sys- equation; however, the variation in each variable degree of free-
tem, only the topological surface is considered and the artistic dom (DOF) is always different. The focus of matrix model
face is ignored. The strengths and weaknesses of this model establishment is to confirm the local datum reference frames
are shown in Table 2. (DRFs) and its displacement matrix, which is defined with re-
spect to the local DRF for each feature to which a tolerance is
2.2 Matrix model applied. Taking a cylinder tolerance zone shown in Fig. 2 as an
example, line [A, B] is the axis element. The ideal cylinder
The matrix model introduced by Desrochers et al. [25] was without form defects and thickness has two invariance degrees
developed from the TTRS model, and it uses a homogeneous that cannot translate along the z-axis and rotate around the z-axis
transformation matrix to describe the small displacements of a in the tolerance zone, so w and γ in Eq. (1) have no effect on the
feature within the tolerance zone and the clearance between tolerance analysis of the cylinder and can be ignored. The other
two features. This model describes the nominal geometric four degrees, i.e., α, β, u, and v, decide the matrix D. Additional
relationship between parts and variations caused by geometric constraints are also necessary to ensure that the deviation varies
deviations. A displacement variation in the tolerance zone within the bounds of the cylindrical tolerance zone:
depends on six independent parameters, that is, three rotation 2 3
Cβ SαSβ SβCα u
parameters and three translation parameters. The rotation pa- 6 0 Cα −Sβ v 7
rameters can be expressed as α, β, γ, which, respectively, D¼6 4 −Sβ SαCβ CαCβ 0 5
7 ð2Þ
represent rotations around the x, y, and z axes while u, v, and 0 0 0 1
w denote, respectively, translations along the coordinate axes.
The six variable ranges are restricted by a series of inequal-
ities, which can effectively express the variation of features,
i.e., point, line, surface, and the fit between them in 3D space.

Table 2 Strengths and weaknesses of the TTRS model

Strengths [1, 97] Weaknesses [1]

• Successfully models datum • Partly conforms to tolerance


precedence standards
• Able to model clearances and • Tolerance zone types are not decided
spaces (by torsors or matrices) by tolerance type but by features
• Presents tolerance
specifications
Fig. 2 Cylindrical tolerance zone
Int J Adv Manuf Technol

0 ≤ α≤ π; 0 ≤8
β ≤2π 3 functions and various constraint inequalities which truly cause
>
> XA
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi >
> 4 YA 5 ¼ D  A a complex calculation and optimization process; thus, it is not
> t >
>
< X 2A þ Y 2A ≤ < conducive for parallel or complicated serial assembly. This
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 with 2 ZA 3 ð3Þ
model has been successfully integrated into CAD systems,
>
: XB þ YB≤
2 2 t >
> XB
>
> 4 5 i.e., CATIA.3D FDT and FROOM. The corresponding
2 >
> YB ¼ D  B
: strengths and weaknesses are listed in Table 3.
ZB

2.3 Vector loop model


It is obvious that a complete modeling of the toleranced
feature is realized through a matrix shown in Eq. (2) describ- Vector loop-based model uses vectors to represent the dimen-
ing the possible deviations of axis [A, B], along with a set of sions in an assembly. Each vector represents either a compo-
constraints shown in Eq. (3) representing the geometry and nent dimension or a kinematical variable dimension. Three
dimension of the tolerance zone. main sources of deviation in a mechanical assembly are fully
Tolerance analysis using a matrix model can be conducted considered, that is, respectively, (1) dimensional deviation, (2)
following the basic steps given below: geometric feature deviation, and (3) deviation due to small
kinematic adjustments which occur during the assembly pro-
1. Create an assembly graph where the local DRFs are asso- cess in response to the dimensional variations and geometric
ciated to the features, including functional features and feature variations of the components. In this model, dimen-
part features. sional tolerances are reflected as change in the length and
2. Determine the points for measuring the displacements and direction, which are considered by combining variations of
their corresponding paths that connect the points to the the vectors into vector loops. Geometric tolerances are taken
global DRF. into account by introducing some vectors of Zero-Length ±
3. Calculate the range of displacement of each point in the Tol into the vector loops whose orientations depend on the
global DRF and add additional constraints to restrict the type of kinematic joint and the corresponding geometric fea-
points of features inside the tolerance zone. ture tolerances associated with it. These two deviations are
4. Calculate the total displacement when multiple tolerances natural variations in manufacturing processes.
are specified. Kinematic deviation is embodied in the model by creating
5. The worst-case method is adopted to determine the max- kinematic joints at the interfaces of mating parts, and the type
imum and minimum deviations. of joint is determined by the conditions of the mating contact.
There are 6 common joints in 2D assemblies and 12 common
It should be mentioned that the statistical approach used to joints in 3D assemblies, and these joints are then connected by
be considered as not applicable to the matrix model [83, 90]; vectors and linked to form kinematic chains or vector loops.
however, new works in the literature show that the Monte Kinematic deviations normally serve as resultant assembly
Carlo method can be applied to plane and cylinder for solving dimensions which can only be measured by solving the vector
the non-linear objective function in this model [98]. equations after the parts are assembled. Chase [30] completed
The matrix model is a point-based model which is suitable a vector loop modeling for the 2D assembly, i.e., one-way
for assembly with joints that are in contact or gapped. All the clutch and geometric block assembly with cylinder sliders,
dimensional and geometric tolerances can be calculated ex- in which the direct linearization method (DLM) was applied
cept the form tolerance. It does not give a complete one-to-one to analyze the assembly tolerances. This method does not need
mapping between the model variable and assigned tolerance; an explicit function to describe the relationship between the
therefore, the conversion between tolerance and model vari- assembly dimensions and the manufactured part dimensions.
able hardly meet the ASME or ISO standards. The choice of Gao [32] further developed this model to 3D mechanical as-
different points for a feature may result in disparate objective semblies, for example, a serial crank slider mechanism, in

Table 3 Strengths and


weaknesses of the matrix model Strengths [25, 26] Weaknesses [2, 25]

• Successfully integrated in CAD • Without consideration of form tolerance


• Easily represents tolerance zones • Partly conforms to tolerance standards
• Easy for tolerance propagation calculation • Point-based model
(transformation matrix) • Complex non-linear constraint inequalities
• Worst case mostly, and the statistical method
has not been fully developed
Int J Adv Manuf Technol

which a closed vector loop defines a kinematic closure con- kinematic variation in worst case or statistical method
straint for the assembly and an open vector loop describes a can be performed:
gap or a stack dimension corresponding to a critical assembly
feature. Later on, Chase [99] provided a more generalized Worst case:
approach to include all the geometric feature variations into  
   
the tolerance analysis and took an example of Tape hub for ΔY ¼ ∑ni¼1 S Xi T Xi þ ∑nj¼1 S Zj T Zj ð7Þ
illustration.
The tolerance analysis process with the vector loop com- RSS:
bined with DLM can be performed as follows:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2ffi
1. Create the assembly graph, defining the number of vector  X X 2
ΔY ¼ ∑i¼1 S i T i þ ∑ j¼1 S j T j
n m Z Z
ð8Þ
chains and loops.
2. Draw a benchmark for the frame of reference, mark the
kinematic joints, and create the datum paths. where T Xi and T Zj are the tolerance vectors for the dimensional
3. Create vector loops, where the open loop describes the variables and geometric feature variables, respectively, and n
key characteristic and the closed loop indicates the pres- and m are the corresponding number of dimensional variables
ence of one or more adjustable elements in the assembly. and geometric feature variables.
4. Derive the stack-up equations. The assembly constraints 6. Estimation for variation of the assembly performance
within vector loop-based model may be expressed as a requirement can be obtained by substituting the kine-
concatenation of homogeneous rigid body transformation matic variations into the linearized equations for the
matrices, which result in a set of non-linear equations. open loop.
Apply a first-order Taylor series expansion of the
closed-loop equations, which can be written in matrix In addition to DLM, the homogeneous transform matrix
form. combined with the Newton–Raphson iterative procedure is
introduced to represent the vector loop-based assembly func-
tion and solve the assembly constraint equations. It has proved
to be useful when it is difficult or impossible to get the explicit
A½ΔX  þ B½ΔY  þ C ½ΔZ  ¼ 0 ð4Þ assembly function [100].
where A represents the partial derivatives with respect to the The DLM using a truncated Taylor’s series expansion is an
manufactured variables, B is the partial derivatives with re- approximation. In addition to this, Gao [101] developed a new
spect to the kinematic variables, C represents partial deriva- Monte Carlo simulation employing an iterative technique for
tives with respect to the geometric feature variables, ΔX closed-loop assemblies, breaking the limitation of required ex-
means the variations of the manufactured variables, ΔY is plicit functions and normal distribution in the traditional Monte
the variations of the kinematic variables, and ΔZ means the Carlo approach. A comparison of the DLM and modified
variations of the geometric feature variables. Monte Carlo simulation shows that the DLM is more accurate
Solving Eq. (4) for ΔY gives the following: under conditions of relatively small tolerances and not highly
For B as a full-ranked matrix assuming, the variations of non-linear assembly functions [102]. A new second-order tol-
the kinematic variables can be yielded: erance analysis (SOTA) method combining the advantages of
the linearized method and the Monte Carlo simulation achieves
½ΔY  ¼ −B−1 A½ΔX −B−1 C ½ΔZ  ¼ S X ½ΔX  þ S Z ½ΔZ  ð5Þ the strength in speed, tolerance allocation, closed-loop con-
straints, non-linear geometric effects, and non-normal input
For B as a singular matrix assuming, the variations of the and output distribution [103]. A good point set method is pro-
kinematic variables can be yielded: posed to generate the points for the toleranced dimensions be-
 −1  −1 cause of its more uniform point clouds [104].
½ΔY  ¼ − BT B BT A½ΔX − BT B BT C ½ΔZ  A vector loop-based equation evaluates the sensitivity ma-
¼ S X ½ΔX  þ S Z ½ΔZ  ð6Þ trix with respect to both manufactured dimensions and geo-
metric variables which can be convenient for sensitivity anal-
where SX and SZ are the tolerance sensitivity matrices for the ysis and assembly rejects calculations [105]. As can be seen,
dimensional variables and geometric feature variables, this approach considers only those parts in contact. The rigid
respectively. body motion in 3D embarks from the starting point, runs
through each node, and finally returns to the original point;
5. The kinematic variations can be obtained from Eq. (5) or it is thus a truly point-based model. Neither the datum proce-
(6) for the closed-loop constraint. A estimation of dure nor the tolerance principles can be considered in this
Int J Adv Manuf Technol

model. The tolerance interaction relationship cannot be solved


either. The manufacturing signature related to the points on the
manufactured surface is considered in vector loop model with
a skin model insertion [7]. Tolerance analysis and allocation
can be successfully handled by 3D software CETOL 6 Sigma
[106]. Table 4 compares the merits and demerits of the vector
loop model.

2.4 T-Map model

The T-Map® (Patent No. US6963824) model proposed by


Davidson is a one-to-one mapping from all the variational Fig. 3 Basis tetrahedron
possibilities of a feature within its tolerance zone to the hypo-
thetical Euclidean volume point space. The size, position, (2) choose the coordinate system of the tolerance zone, usually
form, and orientation tolerance allocated for a target feature in the geometrical center; (3) calculate the maximum param-
decide the shape, size, and internal subsets of the points. Each eters along the DOFs limited by each tolerance zone; (4) lo-
T-Map is a convex hull set on the metric space that captures all cate the n + 1 feature positions which represent the maximum
the variation possibilities of which the shape or shape varia- parameters when n DOFs are constrained; and (5) map these
tion are thought of as a subset. Based on this, some new n + 1 feature positions to the hypothetical Euclidean point
mathematical models for features are very compatible with space. After these procedures have been done, the T-Map
the ASME/ANSI/ISO standards for geometric tolerances model for round face [40], cylindrical face [41], axes [42],
[42]. The T-Map for a feature with combined tolerances is polygonal faces [43] (rectangular and triangular faces), angled
modeled by a basis simplex and described with areal coordi- face [107], point-line clusters [108], line profile [109], circular
nates. A T-Map created in n dimensions will be constructed run-out [110], and total run-out have been created and applied
from the n dimensions basis simplex made from n + 1 basis for tolerance analysis and allocation for assembly, such as pin
points. Taking a 3D variation as an example, the correspond- holes [111], tabs, and slots [112].
ing four defined basis points form the tetrahedron, as shown in The T-Map-based tolerance analysis for assembly using the
Fig. 3, and it is described with the centroid coordinates. Four worst-case method can be performed as follows:
masses γ1, γ2, γ3, and γ4, respectively, are located at σ1, σ2, σ3,
and σ4. As long as γ1 + γ2 + γ3 + γ4 ≠ 0, the position of the 1. Generate an individual T-Map for the toleranced feature.
centroid of these masses, that is, σ can be derived from Eq. 2. Make each T-Map conform to model the deviations at the
(9); obviously, it can be any point in the space of the basis target feature of the assembly by matrix transformation.
tetrahedron by changing the four γ. 3. Calculate the accumulated T-Map for the target feature by
ðγ 1 þ γ 2 þ γ 3 þ γ 4 Þσ ¼ γ 1 σ1 þ γ 2 σ2 þ γ 3 σ3 þ γ 4 σ4 ð9Þ Minkowski sum.
4. Create a functional T-Map to represent the functional
requirement.
The single T-Map generation steps are as follows: (1) con- 5. Keep the accumulated T-Map within the functional T-
firm the tolerance type and the DOF of the toleranced feature; Map.

Table 4 Strengths and


weaknesses of the vector loop Strengths [29, 101] Weaknesses [78, 99]
model
• Successfully integrated in CAD • Without consideration of tolerance interactions,
• Models three kinds of variations, especially datum procedures, and tolerancing principles
kinematic deviations • DLM is an approximation
• Convenient calculation of small displacements • Point-based model
and functionality
• DLM strengths in computational accuracy and
efficiency
• Easily obtains analysis parameters, such as
sensitivity, contributor, and assembly rejects
• Suitable for over-determined systems
Int J Adv Manuf Technol

The normal worst-case approach requires the accumu- 2.5 Deviation domain model
lation of the T-Map completely remaining inside the
functional T-Map. Statistical analysis with this model The deviation domain represents all the variations possible for
produces a geometric relationship between the statistical a feature or a target feature inside its tolerance zone. It is
model and its functional requirement, which follows proposed on the basis of the small displacement torsor where
these steps: the geometric variation (tolerance) of a feature is relatively
small compared to its nominal dimension and can be de-
1. Obtain the T-Map of the target feature. scribed with two vectors in the 6D configuration space, that
2. Calculate the intersection of the dimensions of interest. is, three kinds of rotations α, β, and γ representing rotations
3. Get the frequency distribution from which the variable of around the axes x, y, and z, respectively, and three kinds of
the dimensions of interest can be derived [113]. Obtain displacements indicating the translations along these three ax-
each point of the distribution as the area obtained by es. A complete assembly is composed of parts and joints sep-
intersecting the map with the surface of the functional arately corresponding to deviation domains and clearance
map that has been translated inward as a calculated domains.
amount. For every functional and toleranced feature of the parts, the
4. Normalize the frequency distribution to obtain the relative deviation space can be defined by inequalities related to the
frequency distribution of the dimensions of interest. components with limited DOFs of the deviation torsor. Its
5. Convolute the relative frequency distributions of the di- construction considers tolerance type, feature type, and the
mensions of interest for the target features. parameterized representation of SDT. Aimed at the non-
invariant components of the SDT for a toleranced feature, it
The statistical analysis and allocation that yield the uses a set of linear inequalities to indicate the boundaries of
geometric relationship of n-D variations to 1D clearance the tolerance zone. For example, a planar feature with a spec-
are displayed in ref. [112], where the effects of the ified position tolerance relative to reference A, as depicted in
implemented MMC and LMC are discussed. In addition, Fig. 4a, allows three DOFs (one for translations along the z--
the higher dimension probability T-Maps have been pro- axis and two rotations around x and y), and the maximum
posed such that the point clouds are constructed for the deviations along these three axes inside the tolerance zone
features within a size range to extract the pass percent- determine the deviation domain. For each border point p of
age [114]. the feature, the displacement is expressed by inequalities
The T-Map is a rigid assumption model that is suitable for shown in Eq. (10). The convex hull built with these inequal-
tolerance accumulation calculation in serial or parallel assem- ities is the deviation domain (see Fig. 4b).
blies. However, clamping-induced deformations resulting
t ! t
from imperfect geometry can be taken into account in this − ≤ δ P :!
z≤ ð10Þ
2 2
model to gain the functional T-Map [115]. Combining T-
Maps with established manufacturing maps (M-Maps) can
represent the analytical relationship among all relevant design The clearance space is the small relative displacement of
and machining tolerances for the transfer of cylindrical [116] the joint between two distinct parts. The bounds of the clear-
or planar datum [10]. The main characteristics are generalized ance space are decided by the contact conditions and the de-
in Table 5. gree of freedom of the joint. In the 6D configuration space, the

Table 5 Strengths and


weaknesses of the T-Map model Strengths [1, 42, 43, 76] Weaknesses [76]

• Models all types of 3D deviations, either size or geometric • The model has not been fully developed
tolerances, and their interactions • The calculation of points to areal coordinates
• Modeling of some features are consistent with and functional T-Maps may be difficult
ASME/ANSI/ISO standards for geometric tolerances • Covers only 1D clearance frequency
• Incorporates rule#1 (the trade-off between form and size in distributions
ASME standard), floating tolerance, bonus/shift tolerance, • Visualization of higher dimensional T-Maps is
and datum precedence troublesome
• Provides multiple stack-up equations and metric measures • Tolerance representations for joints need
to aid a designer in selecting optimal tolerances further research
• Accumulation of multiple tolerances for worst case and • Poorly integrated into CAD software
statistical analysis
Int J Adv Manuf Technol

Fig. 4 a Plane feature with position specification. b Association deviation domain

clearance space is along the direction of the non-invariant feature and the accumulated clearance domain. But if
degree, as shown in Fig. 5a. The domain corresponds to the the joints are assembled in parallel, the intersection oper-
DOF of the cylindrical joint, that is, the translations and rota- ation is performed on all the corresponding domains.
tions around the y and z axes, respectively [117]. The clear- 4. Check whether the accumulated deviation domain re-
ance domain obtained from a linear set of inequalities can be mains within the accumulated clearance domain.
shown in Fig. 5b. It can be obviously derived that the maxi-
mum displacement cannot be reached when a rotation exists. The assembly is feasible if step 4 is satisfied. Therefore,
The tolerance analysis process with the deviation domain this method is easily used for assemblability verification.
model can be carried out using the following steps: The obtained deviation domain or clearance domain en-
ables a visual maximum and minimum deviation of each
1. Build the dimension chain from the datum to the target of feature and its functional requirement. Optimal tolerancing
the assembly and identify all the toleranced features and in mechanical design is achieved by the relationship be-
all the joints. tween the accumulated deviation and the clearance do-
2. Model the deviation domains for all toleranced features, main. If no clearance exists in the assembly, a functional
and the clearance domains for all the joints in the chain. domain is needed to describe the functional requirement of
3. Calculate the sum of the deviation domains and clearance the assembly. The alignment of the accumulated and func-
domains. For a serial assembly, the Minkowski sum is tional domains results in the stack-up equations for the
used to get the accumulated deviations for the target assembly.

Fig. 5 a Cylindrical joint. b


Clearance space
Int J Adv Manuf Technol

A new worst-case method for the deviation domain model enables the transformation of local parts deviations to globally
was proposed by Mathieu, which uses the influence of coef- functional requirements [129]. The torsor model describes the
ficients to analyze the relation between functional tolerances small displacement variation of features with constrained
and the specified geometric tolerances. This method mainly torsors which define the extreme limits in the 3D tolerance
focuses on serial assembly without clearance at present, and zone.
the result for mechanisms with parallel linkages still needs The Jacobian model expresses the effect of functional ele-
further research [118]. Statistical analysis in this model em- ments (FEs) on functional requirements (FRs) in the tolerance
ploys the Monte Carlo simulation to represent the possible chain. An FE can be a point, line, or surface of the part. Two
variations resulting from the deviations and clearances [119] FEs separately belong to different parts constituting a kine-
with the shortage of absent parameters contributing to toler- matic pair, while two FEs coming from one part make up an
ance identification. To solve this, sensitivity analysis methods internal part. The Jacobian matrix proposed maps all SDT
are adopted with an interface module to simulate the impact of from their local origin to the point of interest. It is a matrix
deviating component surfaces on the assembly clearances whose columns are extracted from the various homogeneous
[120, 121]. transform matrices relating to the reference frames of the FEs
Form tolerances, i.e., straightness and flatness, can be to that of the FR. It can be formulated as
modeled by vibration modes which in turn modify the devia- 2  h i 3

tion domain. The variation of the clearance domain due to Ri0 … W ni Ri0 : RiPT i
6 33 33 33 33 7
form deviation is measured by modal parameterization [11]. ½ J FEi ¼ ½ J i0 ¼ 6
4 ⋮ ⋱  ⋮h i 7
5
A concept of use rate (UR) is presented to make a tolerance ½033 … Ri0 33 : RiPT i
33 66
analysis of the mechanism with flexible joints, which shows
ð11Þ
the potential of this model in flexible part representation [122].
The deviation domain with its main benefits and drawbacks is 
where Ri0 ¼ ½C 1i C 2i C 3i  is the local orientation change
shown in Table 6. h i
Similar to T-Map and deviation domain, the polytope meth- of frame i with respect to original frame 0; RiPT i ¼
od is proposed based on sets of constraints dealing with devia- ½C 1 C 2 C 3 PT i represents the direction change of the tolerance
tion volumes defined by the admissible displacements of with respect to the three coordinate axes of frame i. W ni is a
toleranced features inside its tolerance zone [125, 126]. skew-symmetric matrix indicating the position change of
Through operations on polytopes, namely Minkowski sums frame n with respect to frame i, during which
and intersections, different applications have been developed 2 3
for tolerance analysis of overconstrained and flexible assem-  n 0 −dzni dyni
blies [127] and assembly interference diagnosis [128]. These W i ¼ 4 dzni 0 −dxni 5; dxni ¼ dxn −dxi ; dyni
algorithms have been implemented in the open source tool −dyi dxi
n n
0
politopix, which together with the open source tool PolitoCAT ¼ dyn −dyi ; dzni ¼ dzn −dzi
allows the creation, visualization, and computation of polytopes
for tolerance analysis: http://i2m.u-bordeaux.fr/politopix.html. Torsor theory is presented to represent the surface or fea-
ture variation with respect to its nominal position in six de-
2.6 Jacobian–torsor model grees of freedom. Taking into account the TTRS foundation
and positional tolerancing principles, the components of var-
The Jacobian model is a kinematic-based method which uses iation can be shown as the screw parameters, which indicate
transformation matrices of open chains in robotics, and the invariant degrees and non-invariant degrees.

Table 6 Strengths and


weaknesses of the deviation Strengths [117, 121–124] Weaknesses [77, 118]
domain model
• Models all types of 3D deviations, either size or • Deviation domain is defined by linear inequalities,
geometric tolerances, and their interactions and complex surfaces need to be discretized
• Could handle flexible joints • No sufficient polyhedral computation code
• Incorporates bonus/shift tolerances • Set of inequalities enables complex computation,
• Optimal tolerance in mechanical design especially for parallel assembly
• Serial and parallel assembly • Not yet integrated into CAD software
• Easily performs an assemblability analysis
• Suitable for contact or non-contact joints
• Sensitivity or contribution analysis
Int J Adv Manuf Technol

The unified Jacobian–torsor model introduced by The Monte Carlo approach is commonly used for random
Desrochers et al. [130] is an innovative tolerance analysis meth- number generation in the statistical analysis of Jacobian–
od, which uses the small displacement torsor (SDT) for toler- torsor model [131]. Ghie [132] combined the Jacobian–torsor
ance representation and the Jacobian for tolerance propagation. model and uncertainty measurement to carry out
Its construction combines the advantages of both two methods. assemblability evaluation. A study on the influence of
The relationship between FEs and FR is shown in Eq. (12). cylindricity error to actual fitting clearance and final required
function makes a breakthrough for the Jacobian–torsor model,
½FR ¼ ½ J½FE ð12Þ which is normally considered unable to incorporate form error
[133].
where [FR] is the functional requirement (play, gap, clearance) In addition, the Jacobian–torsor model has successfully
represented by the translational and rotational vectors of general applied for geometrical variations and uncertainty manage-
assembly. The [FE] is related mathematically to the functional ment in a multi-disciplinary environment [134], error propa-
elements (tolerance, kinematic joint, etc.,) for the assembly gation analysis through fixture assembly [135], serial and par-
parts. [J] indicates the Jacobian matrix. tial parallel connections in assembly [136], modeling of con-
Expanding Eq. (12) into ical structures and their joints [137], the deformation effect
22h i 3 3 connected to geometrical error caused by the temperature field
6 uu 7
66h i 7 7 and force field [138], and model modification resulting from
66 v 7 7
66 v 7
66h i77 7
the interactions between tolerances. It should be noted that the
66 7 7
66 ww 7 7 constraints between translational vectors and rotational vec-
2h i3 66h i 7 7
66 7
u u 66 αα 7 7 7 tors and the relation between dimensional tolerances and geo-
6h i 7 66h i7 7
6 7 66 7
6 vv 7 66 ββ 7 7 metric tolerances can be used for other 3D tolerance analysis
6h 64h i 7 7
6 i77 6 5 7 models, such as the matrix model [139]. A summary of these
6 ww 7  6 7
6h 7 6 δδ 7
6 i 7 ¼ ½ J 1 J 2 J 3 J 4 J 5 J 6 FE1 …½ J 1 J 2 J 3 J 4 J 5 J 6 FEn :6 2 h i 3 7
6 … FE1
7 characteristics is displayed in Table 7.
6 αα 7 6 7
6h i7 6 6 h u ui 7 7
6 7 66 7
6
6 ββ 7
7 66 vv 7 7 7 2.7 GapSpace model
4h i 5 66h i7 7
66 7 7
δδ 66 w 7 7
66 7
i7
w
66h 7 7
66 7 The 1D GapSpace model was first proposed by Morse [140,
66 αα 7 7 7
66h i7 7 141] to derive the necessary and sufficient conditions for a
66 7 7
66 β β 7
44h i 7 5 5 satisfactory assembly. In this model, the GapSpace is divided
δδ FEn into regions in which the assembly is feasible or not feasible.
ð13Þ Two basic concepts, the directed dimension tree and the liai-
son, are presented. The directed dimension tree is a data struc-
where i represents the number of functional elements on the ture that describes the size scheme of the parts in a specified
assembly chain. direction, and each directed dimension tree corresponds to an
u; v; w; α; β; γ : Lower limit of dispersions (translation exclusive set of feature characteristics. The liaison is consid-
and rotation) defining the tolerance zone. ered to be the adjacent specification for pairs of features,
u; v; w; α; β; γ : Upper limit of dispersions (translation which belong to different parts, that is, the signed distance
and rotation) defining the tolerance zone. of two features. Considering the liaisons as independent var-
The deterministic (worst case) and statistical case analysis iables, a linear vector space with basis vectors is constructed,
based on the Jacobian–torsor model can be performed by the which forms the GapSpace model. Taking a hinge assembly in
following procedures [57, 58]: Fig. 6 as an example to illustrate the modeling, we can see that
four liaisons obviously exist separately g1, g2, g3, and g4.
1. Build the expression of the Jacobian–torsor model. The Therefore, four basis vectors ½ 1 0 0 0 , ½ 0 1 0 0 
work includes FEs and FR identification, reference frame ; ½ 0 0 1 0 ; and ½ 0 0 0 1  are chosen to make up
foundation, and the Jacobian matrix establishment. the GapSpace.
2. For the worst-case method, use the multiplication operator The statistical analysis based on the 1D GapSpace model
in arithmetic interval algorithms to obtain the maximum follows [38]
and minimum values of FR.
3. For statistical analysis, random number generators are used 1. Identification of all liaisons and dimension trees.
to generate a sample of random variables (tolerances) with- 2. Assembly graph construction.
in each SDT component, and then compute the correspond- 3. Fit condition (FC) representation.
ing average and standard deviation of the torsor to obtain 4. Assembly space capture.
the objective FR in the specified direction. 5. Tolerance regions within the assembly space generation.
Int J Adv Manuf Technol

Table 7 Strengths and


weaknesses of the Jacobian– Strengths [136, 138, 139] Weaknesses [133, 136]
torsor model
• Easily handles tolerance representation and propagation • Incorporation of form error is not fully developed
• Easily obtains functional requirements • Insufficient handling of bonus tolerances
• Serial and partial parallel assembly • The set of inequalities engenders complex
• Deals with tolerance interactions computation, especially for parallel assembly
• Deals with deformations under working conditions • The final functional requirement is an interval
instead of a determined value
• Easily handles sensitivity or contribution analysis
• Not yet integrated into CAD software

6. Joint probability density function (pdf) is built for the (LMCS), then obtain the maximum and minimal values
manufactured dimensions. of the FC. However, in the statistical case, the distribution
7. Integration of the pdf through the tolerance region and the of each tolerance is assumed first and then arbitrary FC is
assembly region. generated with the pdf.
8. Functional requirement verification. 4. Tolerance analysis such as assemblability analysis re-
quires that all the FCs are non-negative.
Zou later extended the GapSpace model to the multi-
dimensional assembly. In contrast to its 1D condition, there You [9] modified this model by integrating various dimen-
is no need for the dimension tree. The gap is no longer repre- sions and tolerances, strictly distinguishing component and
sented by directed arcs, and the node does not correspond to a assembly D&T specifications, bringing in more effective
feature, but rather to a constraining simplex (CS). There are mathematical analysis methods, i.e., the Monte Carlo and sim-
planar gap, cylinder-slider gap, edge-slider gap, parallel cylin- plex algorithms for complex linear or non-linear multivariate
der gap, and eccentric gap included according to the different distribution stack-up functions. Compared to the original
node characteristics and serve as basic analysis unit. The CS is model, the revised GapSpace assembly analysis (RGSA)
proposed as the basic functional unit in a part, which is asso- covers a more comprehensive analysis, such as tolerance syn-
ciated with constraints on the dimensions, while fitting condi- thesis, checking of clearances, sensitivity analysis,
tion is a key element for performing various GapSpace assem- assemblability analysis, and tolerance synthesis. It considers
bly analyses. Each fitting condition has two equivalent repre- more D&T schemes including surface deviations, shape and
sentations: gap representation and geometric representation, direction deviations, and the different dimensional schemes
which form the basis of tolerance analysis. A more developed like manufacturing dimensioning, parametric dimensioning,
assembly graph is used to represent the relationship among and parallelism dimensioning. Gap is one of the basic units
these constraints for every component. The GapSpace-based of the RGSA model; dimensions and tolerances instead of CS
analysis incorporates several steps as shown below [39, 142]: are treated as basic functional units in fitting condition. This
model have merits and demerits in Table 8.
1. Find all the gaps and compute all the CSs for each
component.
2.8 Skin model/skin model shape model
2. Build the assembly graph and find all FCs based on the
assembly graph.
A part will deviate from its initial nominal geometry due to
3. Representation of FCs by means of a linear sum of gaps or
the existence of manufacturing imprecision and measure-
geometric parameters.
ment uncertainty. Various models either ignore the form or
1. For the worst-case method, it is necessary to consider the
shape defects of features or translate geometric deviations of
extreme values of the tolerances and calculate the corre-
features into simple translation and rotation deviations which
sponding maximum material constraining simplex
have tremendous effects on the part representation as well as
(MMCS) and least material constraining simplex
on tolerance analysis results at the assembly level. In re-
sponse to this, the skin model is proposed with a consider-
ation of form error and it certainly accords with the geomet-
rical product specification standard (GPS). As a physical
interface between the workpiece and the environment, the
skin model itself is an infinite model, which is not con-
cerned with geometry information. The discrete shape ap-
proach for skin model simulation was presented by Zhang
Fig. 6 A hinge assembly [54, 143] in which modeling with geometric error was
Int J Adv Manuf Technol

Table 8 Strengths and


weaknesses of the GapSpace Strengths [9, 38] Weaknesses [9]
model
• Suitable for certain GD&T tolerancing methods • Extreme values of fitting conditions are obtained by,
• Tests the quality of an assembly and calculates the respectively, substituting corresponding MMCSs
sensitivities and LMCSs into fitting conditions
• RGSA can deal with floating assembly and bonus • GSA is not suitable for all types of tolerance
tolerances representation and RGSA does not easily handle 3D
constraining simplex calculations
• Suitable for tolerance representation and analysis in
the mechanical design, manufacturing, and • Integration into the Solidworks software has not yet
inspection stages been fully realized
• Easily performs analysis for over-constrained • The effect of tolerances on motion has not been
assemblability discussed
• Suitable for contact or non-contact joints
• Addresses the relative motion of the parts of a
floating assembly

divided into two phases. Systematic and random deviations 1. Modeling of deviations
generated by different methods are added to the nominal
model in the prediction stage and simulate the possible out- The generation of the skin model shape can be divided into
comes with feasible sample methods in the later design stage two stages depending on the available information in the prod-
[144]. A comprehensive framework for skin model simula- uct design.
tions was constructed by Schleich [53]. The conceptual and In the prediction stage, the point clouds which make up the
computational process for geometric representation is toleranced features build up on the nominal model and then
highlighted in ref. [145]. However, the intrinsically infinite enhanced by random and systematical deviations [53, 54].
representation hinders the opportunity for the identification Random deviations can be derived from Markov Chain
and simulation of the skin model in computer-aided Monte Carlo (MCMC) [144] or Gaussian random fields
software. (GRF) [143], while system deviations can be obtained by
The skin model shapes are finite representation of the second-order shapes [144] or modal method [147]. To gener-
skin model that use point clouds or surface meshes to allow ate consistent skin model shape for deviated part, FEA method
a finite workpiece representative. This method has success- is adopted [148].
fully incorporated form, orientation, location, and profile In the observation stage, limited observations are obtained
tolerances with well-defined computation and scaling from measurements on part prototypes or results of
[146]. A skin model shape that satisfies the specified tol- manufacturing process simulations. Computer visualization
erance specifications can be generated in a five-stage sim- tools or regression analysis are used to generate geometry
ulation [17], as shown in Fig. 7. The tolerance analysis is sampling close to real world situations [149].
performed to demonstrate the effects of geometric part de-
viations and different assembly sequences on multiple 2. Relative positioning
function key characteristics [74], which comprises three
steps as shown in Fig. 8, respectively, (1) skin model shape Tolerance analysis requires the simulation of the relative
creation, (2) assembly simulations, and (3) key character- positioning considering different kinds of geometric part de-
istic calculations as well as visualization. viations. The approaches for the relative positioning of dis-
As can be seen, the main technologies that need to be crete geometry are mainly constrained registration [150] and
addressed focus on two aspects as follows: difference surface [50]. The former falls short in the

Fig. 7 Illustration of the skin model shape generation process


Int J Adv Manuf Technol

Fig. 8 Skin model shape-based


tolerance analysis

consideration of the assembly process, i.e., assembly direc- The strengths and weaknesses of the skin model shape in a
tion, assembly force, workpiece gravity, or assembly se- geometric tolerance presentation are summarized in Table 9.
quence; the latter employs difference surface to compute the
deterministic contact points based on point-wise feature dis-
tances [151]; further, an optimized approach is proposed com- 3 Case study
bining both the difference surface and the constrained regis-
tration method [152]. To compare the eight models previously described, a simple
The skin model shape-based simulations fulfill the follow- assembly shown in Fig. 9 has been applied. It features two
ing criteria: parts in perfect contact. The aim of the tolerance analysis is to
measure the gap g on the leftmost of the assembly when the
1. Contact quality evaluation. The model uses the weight of tolerances applied to the components. The relevant parameters
point-onto-point sets under different methods and objec- are listed in Table 10:
tive functions to measure the accuracy of relative
positioning. Tolerance analysis by TTRS The tolerance analysis based on
2. Motion tolerancing [6]. It conducts tolerance analysis for TTRS model starts with the detection of the kinematic loops
discrete time steps over a motion cycle. from the assembly graph. Figure 10 shows the assembly graph
3. Deformation calculation [151]. It presents a calculation of for the case, then the loop (O1, O2, O3, O4) can be obtained.
deformation effects caused by the part’s thermal and The TTRS trees resulting from the assembly with
working environment. MGDE/MGRS surfaces are shown in Fig. 11.

Table 9 Strengths and


weaknesses of the skin model Strengths [17, 74, 150, 151, 153] Weaknesses [143, 146]

• Represents all types of tolerances • No sufficient means for tolerance principles


• Conforms to GD&T standards • The discrete accuracy depends on processing
• Point cloud or mesh surface can be directly obtained in FEA software
from FEA and CFD • Has not been integrated into CAD software
• Deals with tolerance interactions • Difficult to analysis on complex assembly
• Deals with deformations in the working conditions
• Suitable for rigid and flexible parts
• Easily handles geometric tolerance management in
the design, manufacturing, and inspection stages
• Allows a more realistic prediction of assembly
characteristics in product development
• Deals with over-constrained assembly
• Allows contact and mobility simulations
Int J Adv Manuf Technol

Fig. 9 Illustration of the case. a Part 1. b Part 2. c The assembly and function requirement

The torsor representation for the toleranced TTRS, that is 2 3


separately TTRS.1 and TTRS.2, by considering that the ex- ½0; 0 2 3
6 7
treme points A and B or C and D remain into the tolerance 6 h ½0; 0 i 7 ½0; 0
6 t1 t1 7 6 ½0; 0 7
zone are 6 − ;þ 7 6 7
6h 2 2 7
i 7 ¼ 6 ½−0:2; þ0:2 7
6
8 9 T1 ¼ T2 ¼ 6 t
6 − ;þ
1:2 t 1:2 7
7 6 ½−0:005; þ0:005 7 ð16Þ
>
<0 − >
= 6 20 7 6 7
6h t t 1:2 i 7
20 4 ½−0:005; þ0:005 5
T1 ¼ − h t þt − i ; T2 6 1:2 7
> 1:2 t 1 þ t 1:2 > 4 − ;þ 5 ½0; 0
:− − − 1 ; ; 20 20
2 2 ½0; 0
8 9
>
<0 − >
= where the T1 is the torsor of the toleranced TTRS.1 and T2 is
¼ − h t þt 0 t þt i ; ð14Þ the torsor of the TTRS.2.
>
:− − − 2 2:2
;
2 2:2 >
;
2 2 The cumulative torsor G is expressed along the stack loop
in the assembly:
where t1 and t2 are the tolerances on the dimensions and t1.2
and t2.2 are the tolerances of geometry. The functional require- G ¼ T1 þ T2 ð17Þ
ment Δg can be expressed by the translation along the global
And, therefore, the functional requirement in the worst-case
z-axis.
approach performed along the z-direction is
ðt 1 þ t 2 þ t 1:2 þ t 2:2 Þ
Δ g¼ g ¼ ð20−18Þ  ð0:2 þ 0:2Þ ¼ 2  0:4 ð18Þ
2
ð0:4 þ 0:4 þ 0:1 þ 0:1Þ The solution of TTRS model can also be screw [154] and
¼ ¼ 0:5 ð15Þ
2 matrix approaches.

However, considering tolerance interactions, the torsor of


plane specified with an orientation tolerance besides a dimen-
sional tolerance becomes [139]

Table 10 Tolerance
values proposed ta tb tc td

− 0.2/0.2 0.1 − 0.2/0.2 0.1


Fig. 10 The assembly graph of the case
Int J Adv Manuf Technol

displacements of point E is decided by two sections that


describes in Eq. (20).

Δg ¼ ΔZ E −ΔZ F ð19Þ
ΔZ E ¼ ΔZ E;O1 þ ΔZ E;O3 ð20Þ

where the ΔZ E;O1 and ΔZ E;O3 mean the variations resulting


Fig. 11 The TTRS trees for each part of the case
from the local DRF O1 of part 1 and local DRF O3 of part 2,
respectively.
Considering the planar tolerance zone for the case. The
Tolerance analysis by matrix model The initial step for the tolerance matrix is then defined as shown in Eq. (21). If
matrix model in tolerance analysis is to transform the POi→ O j is the transformation matrix of coordinate frame O0
tolerances of the assembly according to TTRS method; to frame Oj,then the vector of displacement MM′ of a point
the corresponding assembly graph can be derived as M in the tolerance zone to point M′ expressed in O0 is now
shown in Fig. 10. The related links between the two described in Eq. (22). Therefore, the displacements of point E
parts form a global DRF that been called O0 and three are given as Eqs. (23) and (24).
local DRFs namely O1, O2, and O3 associated with the 2 3
Cβ 1 −Sα1 Sβ1 Sβ1 Cα1 0
key features, as displayed in Fig. 9. Moreover, the 6 0 Cα1 −Sα1 0 7
graph shows the required measure (g) between the point Dðα; β; wÞ ¼ 6 4 −Sβ 1 Sα1 Cβ 1 Cα1 Cβ 1 w1 5
7 ð21Þ
E and point F. 0 0 0 1
The clearance (g) between E and F can be approxi- h 0
i  h 0
i
mately measured in the z-axis of the global DRF that MM ¼ P−1O0→ O j Δ D j −I Δ M ð22Þ
O0 Oj
can be expressed in Eq. (19), among which the value of
ZF is zero for the behalf of origin of coordinates. The

2 32 32 3
1 0 0 0 Cβ 1 −1 −Sα1 Sβ1 Sβ1 Cα1 0 0
60 1 0 0 7 6 0 Cα1 −1 −Sα1 0 7 6 7
ΔZ E;O1 ¼ ½p1 ⋅½D1 −I ⋅½E O1 ¼6 76 76 0 7
40 0 1 20 54 −Sβ 1 Sα1 Cβ 1 Cα1 Cβ 1 −1 w1 54 −18 5
2 3 0 0 0 1 0 0 0 0 1
ð23Þ
−18Sβ 1 Cα1
6 18Sα1 7
¼6 7
4 −18ðCα1 Cβ 1 −1Þ þ w1 5
0

2 32 32 3 2 3
1 0 0 0 Cβ2 −1 −Sα2 Sβ2 Sβ2 Cα2 0 0 0
60 1 0 07 6 0 Cα2 −1 −Sα2 0 7 607 6 0 7
ΔZ E;O3 ¼ ½p2 ⋅½D2 −I ⋅½E O3 ¼6
40
76 76 7 ¼ 6 7 ð24Þ
0 1 2 54 −Sβ2 Sα2 Cβ 2 Cα2 Cβ 2 −1 w2 54 0 5 4 w2 5
0 0 0 1 0 0 0 0 1 0

To make sure all the points of planar surface within the 8 9


>
> −10Sβ 1 þ w1 > >
tolerance zone, the constraints have to be fulfilled at the ex- < =
10Sα1 Cβ 1 þ w1 10Sβ 1
treme four points of part 1 and part 2. The constraints of the −0:2 ≤ ≤ 0:2 −0:1 ≤ ≤0:1
>
> 10Sβ1 þ w1 > > 10Sα1 Cβ 1
: ;
components under the dimensional and geometric tolerances −10Sα1 Cβ 1 þ w1
of part 1 are ð25Þ
Int J Adv Manuf Technol

where −0.2 ≤ w1 ≤ 0.2, − tb/20 ≤ α1 ≤ tb/20, − tb/20 ≤ β1 ≤ Table 11 Elements of R


and T matrices of the Loop
tb/20.
loop
The constraints of the components under the specified di- R T
mensional and geometric tolerances for four vertices of part 2
are 0 x1
8 9 0 α1 = 0 ± 0.05
>
> −10Sβ 2 þ w2 > > 90° x3
< =
10Sα2 Cβ 2 þ w2 10Sβ 2
−0:2 ≤ ≤ 0:2 −0:1 ≤ ≤0:1 90° x2
>
> 10Sβ2 þ w2 > > 10Sα2 Cβ 2
: ; 0 α2 = 0 ± 0.05
−10Sα2 Cβ 2 þ w2
0 g
ð26Þ

where −0.2 ≤ w2 ≤ 0.2, − td/20 ≤ α2 ≤ td/20, − td/20 ≤ β2 ≤


td/20. The constraint with vector loop-based assembly model can
The sum of two contributors due to the DRFs O1 and O3 be expressed as a concatenation of homogeneous transforma-
makes up the displacement of ΔZE, which can be shown in tion matrices and gives
Eq. (27). For worst-case method, the variation of g is calcu-
Δg ¼ Δx1 þ Δα1 −Δx2 −Δα2 ð30Þ
lated by Eq. (28); this constrained non-linear multivariable
function can be solved by built-in non-linear programming where Δx = {Δx1, Δx2 }T = {0.2, 0.2 }T; Δα = {Δα1, Δα2 }T
solvers in MATLAB. Therefore, the required measurement g T
= ; tb2 ;
tb
2
may be maintained in Eq. (29).
As concerning the sensitivity analysis, we have
ΔZ E ¼ −18ðcosα1 ⋅cosβ 1 −1Þ þ w1 þ w2 ð27Þ    
Δg ¼  ∑jS i jΔxi þ ∑S j Δα j
Δg ¼ −36ðcosα1 ⋅cosβ 1 −1Þ þ 2ðw1 þ w2 Þ ð28Þ
¼ ð0:2 þ 0:2 þ 0:05 þ 0:05Þ ¼ 0:5 ð31Þ
g ¼ ð20−18Þ
g ¼ ð20−18Þ þ Δg ¼ 2  0:5 ð32Þ
 ½−36ðcosð−1:1385e−8Þ⋅cosð−1:1385e−8Þ−1Þ þ 0:4≈2
 0:4 Tolerance analysis by T-Map The tolerance analysis by
ð29Þ means of T-Map method begins with the modeling of T-
Maps for all the toleranced features in the assembly chain
from the datum to the target. The T-Map for the tolerance
Tolerance analysis by vector loop The vector loop has been zone of the planar surface with definition of dimensional
created for the case assembly using the datum paths as a guide, tolerance t can be shown in Fig. 13a. Figure 13b displays
as shown in Fig. 12. There exists only one open loop that an across section of it; the maximum orientation varia-
represents the gap whose variation is the result of the accumu- tions in one direction for the tolerance zone is a very
lation of components’ tolerances. As concerning for the open small angle that equals t/dy. However, the orientation tol-
loop, the corresponding elements of R and T matrices of the erance t1 applied allows a smaller rotation angle t1/dy, as
loop are concluded in Table 11. illustrated in Fig. 13c; the directional deviation is replaced
by t1 for any part on which an orientational tolerance is
specified.
The functional feature for the assembly is the left face
of part 2. The T-Map models for toleranced features of
two parts can be demonstrated in Fig. 14a, b. The
toleranced surfaces of the two parts share the same lateral
dimensions and tolerances; therefore, the non-vertical
lines of the two T-Maps have the same slope. The dimen-
sions of the accumulation map are in terms of ta and tc
using Minkowski Sum, as illustrated in Fig. 14c. To
avoid specifying excessively tight tolerances on the indi-
vidual parts, the functional T-Map for the target face
should allow the accumulation T-Map to inscribe in it,
as expressed in Fig. 14d. The equations for tolerances
Fig. 12 The vector loop for the case stack-up are got for the normal and x or y directions of
Int J Adv Manuf Technol

Fig. 13 a The T-Map for the toleranced planar feature with when size and orientation tolerances t and t1 are specified. d A half-
dimensions dx × dy. b A point map that represents half the planes in one section of the dicone when size tolerance t, orientation tolerance t1 , and
planar section of the tolerance zone of a. c A half-section of the dicone form tolerances t0 are specified

the superimposed T-Maps as Eq. (33) for dimension and the substitution surface with respect to the nominal surface can
direction control. be modeled by SDT method. For a planar surface, the SDT
representation can be shown in Eq. (35).
t f ¼ ta þ tc ¼ 0:4
ð33Þ
t fb ¼ tb þ td ¼ 0:2 E ¼ f0; 0; w; α; β; 0gT ð35Þ

As considering the worst-case range of gap in this case, we As concerning the tolerance zone of tolerance, which
focus on z-direction only, the result is bounds a two-parallel plane centered on the nominal surface.
The constraints have to be fulfilled in three components for
g ¼ ð20−18Þ þ t f ¼ 2  0:4 ð34Þ limiting the displacement of the plane for part 1; we obtain the
following system of inequalities where εi = ± 1.

Tolerance analysis by deviation domain The case conducted tb tb


for analysis in this paper is a general serial assembly without − ≤ ðε2 ⋅10β 1 þ ε3 ⋅10α1 Þ≤
2 2 ð36Þ
clearance. For each part, the variation of position and the −0:2 ≤ ðε1 ⋅w1 þ ε2 ⋅10β1 þ ε3 ⋅10α1 Þ≤ 0:2
orientation of toleranced features are completely defined by
the tolerancing of one surface with respect to the other. On the Each of these inequalities represents a boundary hyper-
hypothesis of the small displacements, the small variations of plane; the intersection of these hyperplanes forms the

Fig. 14 a Half-section of the T-


Map for part 1. b Half-section of
the T-Map for part 2. c A half-
section of the accumulation map
for the assembly. d Half-section
of the functional T-Map and,
inscribed in it, the accumulation
T-Map
Int J Adv Manuf Technol

deviation domain, which is a convex polytope as shown in Tolerance analysis by Jacobian–torsor The key work for
Fig. 15a. The limit variation of each component becomes Jacobian–torsor model is to obtain the Jacobian matrix for
−0:2 ≤ w1 ≤ 0:2; − 20
tb
≤ α1 ≤ 20
tb
; − 20
tb
≤ β1 ≤ 20
tb
. In the same way, error propagation and the torsors for all consecutive pairs.
the deviation domain can be built for part 2 with boundary con- The definition of pairs, functional surface, and coordinate
straints satisfying −0:2 ≤ w2 ≤ 0:2; − 20 td
≤ α2 ≤ 20td
; − 20
td
≤ β2 ≤ 20
td
. frames can be presented in Fig. 16.
Combining the two deviation domains using Minkowski sum, Based on the relationship of this assembly in Fig. 16, we
which is formed by adding each vector in domain A to each vector can identify the FEs in chain that make influence on the func-
in domain B, as expressed in Eq. (37), in which A and B are two tional requirement g. They are internal pair (0, 1), (2, 3) and
polytopes. And then, we obtain the tolerances stack-up for the kinematic pair (1, 2). Then, the Jacobian matrix can be calcu-
target feature (see Fig. 15b); the result is presented in Eq. (38). lated for each FE; the kinematic torsor between FE1 and FE2
is considered null because the form or shape tolerances are not
A þ B ¼ fa þ bja∈A; b∈Bg ð37Þ being considered here; we find the Jacobian matrix:
8 22 3 2 3 3
>
> t z ¼ 
ð0:2 þ 0:2
Þ 1 0 0 0 18 0 1 0 0 0 0 0
>
> 66 0 −18 0 0 7 6 07 7
<r ¼  tb
þ
td 66 1 0 7 60 1 0 0 0 7 7
x
ð38Þ 66 0 0 1 0 0 07 6
7 60 0 1 0 0 077 7
7
>  20 20  J ¼6 6
66 0 0 07 6 07 7 ð40Þ
>
> tb td 66 0 0 1 7 60 0 0 1 0 7 7
>
: ry ¼  þ 44 0 0 0 0 1 05 40 0 0 0 1 05 5
20 20 0 0 0 0 0 1 FE1 0 0 0 0 0 1 FE2

Concerning the gap g along z-direction in worst-case meth-


The dispersions of the virtual joints, due to the applied
od, we get
tolerances, can be established based on the torsor theory.
g ¼ ð20−18Þ þ t z ¼ 2  0:4 ð39Þ Then, the expression of Jacobian–torsor model is

22 3 3
½0; 0
2h i3 66 ½0; 0 7 7
66 7 7
u; u
6 h i 7 22 6 6 ½−0:2; þ0:2 7 7
6 7 3 2 3 36 6 7 7 2 3
6 v; v 7 1 0 0 0 18 0 1 0 0 0 0 0 6 6 ½−0:005; þ0:005 7 7 ½−0:09; 0:09
66 7 7
6h i7
7 6 60 0 −18 0 0 7 6 07 76 4 5 7 6 ½−0:09; 0:09 7
6
6 w; w 7 66 66 1 7 60 1 0 0 0 7 76 ½−0:005; þ0:005 7 6 7
0 07 6 07 76 ½  7 6 7
6h
6 i7
7 ¼ 66
0 0 1 0 7 60 0 1 0 0 7 76 2 0; 0 3FE1 7 ¼ 6 ½−0:4; 0:4 7 ð41Þ
6 α; α 7 6 60 0 0 1 0 07 6 07 76 ½  7 6 ½−0:01; 0:01 7
66 7 60 0 0 1 0 7 76 0; 0 7 6 7
6h
6 i7
7 44 0 0 0 0 1 0 5 40 0 0 0 1 0 5 5666
6 ½0; 0 7 7 4 ½−0:01; 0:01 5
7 7
6 β; β 7 7 7
6 7 0 0 0 0 0 1 FE1 0 0 0 0 0 1 FE2 6 6
6 6 ½−0:2; þ0:2 7 7 ½0; 0
4 h i 5 6 6 ½−0:005; þ0:005 7 7
δ; δ 66 7 7
4 4 ½−0:005; þ0:005 5 5
½0; 0 FE2

(a) (b)
Fig. 15 a Deviation domain for part 1. b Minkowski sum of two deviation domains
Int J Adv Manuf Technol

because gap g2 is zero. Therefore, the maximum and min-


imum gap can be maintained by Eq. (44).

FC ¼ g1 þ g 2 ¼ CSLengthðpart 1Þ−CSLengthðpart 2Þ ð43Þ

g max ¼ LMCS part1 −LMCS part2 ¼ 2 þ 0:6


ð44Þ
g min ¼ MMCS part1 −MMCS part2 ¼ 2−0:6

Tolerance analysis by skin model shape This analysis process


starts from the generation of the skin model shapes for all
toleranced features; the constraints are applied to all the dis-
crete points to ensure the geometric tolerance zones. The
Fig. 16 Definition of the parts and the associated coordinate frames modeling of key features of part 1 and part 2 can be seen in
Fig. 18. These two skin model shapes are assembled in the
rightmost side employing the relative positioning approach
Paying attention to the “z” direction, the translational var- based on the constrained registration method. It should be
iation of the functional requirement is [−0.4, +0.4]; therefore, mentioned that the skin model shape considering actual sur-
the objective gap can be represented as face with form defect distinguishes it from other models. In
this regard, the displacement between feature C and feature D
g ¼ ð20−18Þ  0:4 ¼ 2  0:4 ð42Þ cannot be ignored anymore. To obtain this small displace-
ment, an optimization problem is defined with a constraint
that requires the signed normal distance between any of the
Tolerance analysis by GapSpace There are three gaps asso- reference points in feature C and their correspondences in
ciated with the part 2: g 1 , g 2 , and g 3 . However, no feature D to take a non-negative value, as expressed in Eq.
constraining simplex can be found for the assembly be- (45). The numerical solution of this constrained optimization
cause the assembly is movable along one direction of x. problem can be performed employing an active set, quapro or
To make the model fixed, an extra fixture is imagined to iHLRF algorithms [152].
form the gap g 4 which is exactly opposite to g 3 . The
allowed range of g1 is the objective of interest. All of the min f ðα; β; wÞ
four gaps are planar gaps, among which g2, g3, and g4 are ð45Þ
Subject to DistðC; DÞ ≥ 0 ∀C; ∀D
theoretical zero. The assembly graph for the example is
depicted in Fig. 17b; obviously, two FCs are found. We
focus on FC in z-direction; the corresponding parameters The functional requirement of this case is the determination
are listed in Table 12. The gap-based and the geometric of gap between feature A of part 1 and feature B of part 2. The
parameter-based representation of the FC is listed in Eq. distances can be measured by means of ray tracing ap-
(43). Table 13 reveals the dimensions of the constraining proaches. The results of the gap distribution can be seen in
simplex. The range of gap g 1 equals the range of FC Fig. 19. Orientation variations can be obtained by applying

Fig. 17 a Gaps of assembly. b


Assembly graph
Int J Adv Manuf Technol

Table 12 Parameters in FC = g1 + g2

CT Sine law Contribution coefficient Sign

Part 1 g1 + g2 1 +
Part 2 g1 + g2 1 –

relative positioning between the final position of feature B and


its nominal position.

Results of the case Table 14 makes a summary of the results


when the eight selected models are applied to analyze the
effect of tolerances on the final key function requirement. Fig. 18 Illustration of the assembly
The GapSpace-based worst-case analysis concerns the
dimensions MMCS and LMCS which are borrowed from 4 Comparisons
virtual sizes of MMC or LMC. Extreme values of fitting
conditions are obtained by, respectively, substituting cor- So far, we have listed eight tolerance analysis models for parts
responding MMCSs and LMCSs into fitting conditions, or assemblies, and each of them has their own advantages and
thus resulting in larger variations. Vector loop considers disadvantages. In this section, we propose comparisons to
geometric tolerances as variation sources that are indepen- benchmark the tolerance analysis models, in order to identify
dent from the dimensional variations in the assembly; their common features as well as those that set them apart. The
therefore, there exists no tolerance interaction in this mod- points of comparison are selected from those proposed in the
el. The calculation of TTRS, matrix, T-Map, deviation literature, which are commonly involved in a tolerance anal-
domain, and Jacobian–torsor involves the evaluation of ysis problem, such as tolerance principle, stack-up functions
the six small displacement variations of features specified built, and solved methods. This kind of comparison will pro-
with an orientation tolerance besides a size tolerance, vide a guideline for selecting a model most appropriate for
where tolerance interactions can be considered. specific goals. The differences and similarities between the
Orientation tolerances mainly restrict the rotational dis- eight models with the selected indicators of comparison are
placements, and size tolerances limit the translational dis- listed in Table 15, in which the symbol “×” means the indica-
placements. Analysis performed by skin model shape tor is currently unknown or unable to calculate based on the
yields smaller results; several reasons may account for published literature to date, “√” denotes the indicator as appli-
this. First, the part feature deviations are not simplified cable or already put into use, and “Δ” means the indicator is
to rotational and translational imperfections as in other incompletely developed based on the latest publications, but
models. Second, the calculation of small displacements showing potential to be solved. The following section inter-
considering dependency between function element pair, prets the results in detail based on the respective items.
i.e., the actual feature D not only determines the small
displacement between feature A and D, but restricts the
allowed variation between feature D and C, which allows
a more realistic prediction of assembly characteristics.
However, this method requires the generation of skin
model shapes for all surfaces, assembling of parts bit by
bit and measure of distances for all points between func-
tional features which restricts its use on complex assem-
blies containing a large number of parts and joints.

Table 13 Dimensions of CS

CS Nominal CS length MMCS LMCS

Part 1 (g1, g2) 20 19.7 20.3


Part 2 (g1, g2) 20 18.3 17.7
Fig. 19 Gap distribution of the case
Int J Adv Manuf Technol

Table 14 Case study results DLM, SOTA, and the estimated mean shift are adopted
Model Results in the vector loop. The simplex algorithm and numerical
integration are taken in the GapSpace model. The com-
TTRS [1.6 − 2.4] bination of a Monte Carlo simulation and a simplex al-
Matrix [1.6 − 2.4] gorithm approaches a solution method for linearization
Vector loop [1.5 − 2.5] and optimization [156]. Table 16 reviews the various
T-Map [1.6 − 2.4] approaches proposed for tolerance analysis in the litera-
Deviation domain [1.6 − 2.4] ture with respect to the best choice for certain purposes.
Jacobian–torsor [1.6 − 2.4] 4. Sensitivity/contribution indicates the influence of each
GapSpace [1.4 − 2.6] composition loop change on the variation range of a
Skin model/skin model shape [1.8401 − 2.1805] closed ring in an assembly chain. It is important to obtain
a recommendation from the sensitivity or contribution
results to optimize tolerance values or the tolerance
1. Dimensionality means the model allows the analysis of schemes. The TTRS and the matrix models still lack a
parts or assembly in one dimension or two dimensions or suitable algorithm to calculate them while the vector
three dimensions. All of the models listed show the abil- loop, GapSpace, and Jacobian–torsor [157] models eas-
ity for error analysis in the three spaces, which is neces- ily settle this issue. The T-Map model and deviation
sary for actual mechanical structure characterization. domain model adopt global sensitivity-based tolerancing
2. Tolerance type refers to the type of tolerance consisting techniques for convex hulls [120]. For the skin model
of form, orientation, position, and run-out. It is obvious shape, sensitivity analysis is applied to evaluate the part
that nearly all of the models are suitable for the repre- tolerances to key characteristics (KCs) and then iden-
sentation of dimensional, orientation, position, and form tifies their main contributors.
tolerance except for the TTRS and matrix models, which 5. Objective means the functional requirement is represent-
are not able to deal with form tolerance. The unified ed through the surface or through points belonging to the
Jacobian–torsor model used to be considered as having features or through gaps between the features. The
no ability to handle form tolerance as a Jacobian model TTRS, T-Map, deviation domain, and Jacobian–torsor
or a torsor model [2], but a revised model integrates are surface-based approaches. More specifically,
cylindricity error into the tolerance analysis of precision GapSpace model has gap representation and geometric
rotary assemblies by modifying the expressions of the representation to form the target objective determined by
SDTs of the cylinder pairs [133] and shows a potential to the dimensions and tolerances. Therefore, it can be a
represent other form tolerances. Deviation domain using gap-based or a surface-based model as well. The
SDT components to represent each geometric tolerance surface-based model has more stable precision for selec-
zone, except form error, in the form of convex polytope tion of different points which in a point-based model
referred to as the deviation domain. To solve this demer- may lead to different results. A gap-based model is con-
it, model parameterization is adopted to modify the de- venient for the quality test of an assembly where clear-
viation domain so as to represent form variations [155]. ances are the key elements.
The representation and propagation for run-out toler- 6. Envelope/independent describes a related requirement or
ance, which includes circular run-out and total run-out, unrelated requirement between size tolerance and geo-
can barely be solved, except for the T-Map and skin metric tolerance. The envelope principle demands that
model shape method. the actual size does not exceed the maximum material
3. Analysis method refers to the type of approach that can dimensions. However, this has only been realized in the
be adopted to solve the assembly stack-up functions, i.e., T-Map models at present.
worst case or statistical method. Table 15 displays the 7. Datum precedence represents a sequence of datums
analysis method applied to each model in the literature. when multiple datums are applied to the same feature,
The root sum square solution for T-Map model and ma- and a low benchmark accommodates a high priority. The
trix model still needs further development while worst- matrix model and the vector loop are two models that are
case method is commonly applied to linear function for not able to distinguish the datum precedence among the
seven models. Due to the fact that assembly and mobility eight models.
simulation of skin model shapes are performed based on 8. Joint type refers to the joint types between the parts,
relative positioning of point clouds, in which no assem- either with contact between the features or with clear-
bly function is defined, therefore WC and RSS are not ance. Apart from vector loop model, the others may
applicable. The Monte Carlo simulation has been accept- consider both contact and clearance among the mating
ed for non-linear functions. Some other methods such as parts. GapSpace can be gap-based where the gap can be
Table 15 Comparisons of the eight models

Order Models TTRS Matrix Vector loop T-Map Deviation Unified GapSpace Skin model
domain Jacobian torsor shape

1 Dimensionality 3D 3D 3D 3D 3D 3D 3D 3D
2 Tolerance type Dimensional √ √ √ √ √ √ √ √
Geometrical Form × × √ √ Δ Δ √ √
Orientation √ √ √ √ √ √ √ √
Position √ √ √ √ √ √ √ √
3 Analysis WC √ √ √ √ √ √ √ ×
method Statistical RSS √ × √ × √ √ √ ×
Others GA GA/Monte Carlo DLM/Monte Carlo/s Convolution of Monte Carlo Monte Carlo Monte Carlo/simplex Statistical shape
(Δ) order (SOTA)/ probability algorithm/numerical analysis
estimated mean T-Maps integration
shift
4 Sensitivity/contribution × × √ Δ Δ √ √ √
5 Objective Suface Point Point Suface Suface Suface Gap/surface Point
6 Envelope/independence × × × √ × × × ×
7 Datum precedence √ × × √ √ √ √ √
8 Joint type Contact/clearance Contact/clearance Contact Contact/clearance Contact/clearance Contact/clearance Contact/clearance Contact/clearance
9 Material modifier × × × √ √ × √ ×
10 Stack-up function Linear/network Linear/network (δ) Linear/network Linear/network Linear/network Linear/network Linear/network ×
11 Tolerance interaction √ √ × √ √ √ × √
12 Rigid/flexible Rigid Rigid Rigid Rigid Rigid/flexible Rigid Rigid Rigid/flexible
13 Compatibility with specification Not fully ISO Not fully ISO Not fully ISO Not fully ISO Not fully Not fully ISO Not fully ISO Not fully ISO
and ASME and ASME and ASME and ASME ASME/ISO and ASME and ASME and ASME
14 Suitable for tolerance synthesis √ × √ √ √ √ √ Δ
15 Tolerance transfer √ × × √ √ √ √ √
16 Tolerance evaluation × × × × √ × √ √
17 Series/parallel assembly Series/parallel Series Series Series/paralle Series/parallel Series/parallel Series/parallel Series
18 Application CATIA.3D FDT/ CATIA.3D FDT/ CETOL/Sigmund. × × × GSAS1.0 ×
FROOM FROOM
Int J Adv Manuf Technol
Int J Adv Manuf Technol

Linear/non-linear
Explicit/implicit
Does not matter
zero; skin model shape may consider an unconstrained

Not required
objective function for assembly simulations.
9.

GLP/GP
Material modifier refers to bonus tolerance and contains

NT-net

High
MMC and LMC, which describes the maximum or min-
imum material condition in the size boundary. The T-
Map and deviation domain may take into account the
MMC principle and create the maximum hypothetical
Can use any distribution
Monte Carlo simulation
space and domain space. GapSpace calculates the
MMCS and LMCS in its tolerance stack-up function.

Linear/non-linear
Explicit/implicit
Does not matter
Other models need to be extended to consider material

Not required
Very High
modifiers.
10. Stack-up function can be divided into linear stack-up func-
tions or networks (non-linear) of stack-up functions.
Different objective functions can choose different analysis
methods based on Table 16. It is worth noting that the
Taguchi’s method
Normal assumed

matrix model is changed for non-linear function analysis,


Explicit/implicit

and therefore, all the models can deal with both linear
Not required
Non-linear

functions and networks, except the skin model shapes


due to the fact that no stack-up function exists at all.
High

11. Tolerance interaction indicates more than one tolerance


1

applied to the same feature. Models without the capabil-


ity of representing this interaction are the vector loop and
Estimated mean shift

GapSpace. The Jacobian–torsor model has been modi-


Normal assumed

fied to consider the variations and constraints for a fea-


ture specified by more than one tolerance [139]. The
Non-linear

Moderate
required
Explicit

modification of small displacement variations in six de-


grees of freedom can be applied to matrix and TTRS
1

models.
12. Rigid/flexible, most of the models established based on
rigid body assumptions, can only deal with small dis-
Root sum square
Normal assumed
Linearized 3D

placement deviations. Deviation domain has shown the


Moderate

ability to represent the rigid as well as elastic or flexible


Required
Explicit
Linear

parts for extending the models to flexible joints [122,


158]; a similar method called polytopes has been applied
to flexible assemblies and can be easily calculated using
available online software politopix. Skin model shape is
Not required

Not required

a discrete representation which suitable for both rigid


Worst case

Explicit

and flexible components.


Linear
Comparison of the tolerance analysis methods

Less

13. Conformance to standards, all the models currently are


1

not completely conform to ISO or ASME standards for


the limitation of some applications prescribed in the ISO
Comparison of the tolerance analysis methods

Linear/non-linear

and ASME standards, i.e., envelope and independent


Explicit/implicit

principles. Skin model shape is a core concept of


GeoSpelling and conforms to GPS standard.
Number of computation/simulation

14. Tolerance synthesis, regarded as tolerance optimization,


is an iterative process of tolerance analysis, tolerance
allocation, and adjustment of tolerance values. All the
eight models except matrix have achieved tolerance
Assembly function

synthesis.
Partial derivative
Dimensionality

15. Tolerance transfer, thought as tolerance analysis and


Distribution

synthesis in process planning. TTRS can be used as tol-


Table 16

Aspects

erance transfer technique to succeed in processing plane


specifications and manufacturing tolerancing synthesis
Int J Adv Manuf Technol

[23, 96]; T-Map was developed for transfer datum in model have been applied to CATIA.3D FDT in CATIA,
manufacturing process [10, 116]; Jacobian–torsor model CETOL uses vector loop model, and GSAS1.0 integrat-
realized the error propagation in manufacturing process- ed in Solidworks adopts GapSpace model. A conclusion
es [135]; manufacturing dispersions were taken into ac- on CAT systems can be seen in Table 17. Software pack-
count for assembly in deviation domain model [159]. ages, including CETOL 6 Sigma, eM-TolMate, VSA,
GapSpace considers manufacturing dimensioning 3DCS, MECAmaster, Mechanical Advantage, and
scheme imposed by the manufacturing process [38], Sigmund, normally predict the amount of variation with-
and skin model generation in observation stage con- in the assembly, perform sensitivity analysis to deter-
siders the stochastic manufacturing process [149]. mine the contributors or the “source” of variation, and
These models achieve tolerance transfer that bridges reduce costs with a well-optimized tolerance definition
the gap between design and manufacturing. keeping a full-controlled quality on geometrical require-
16. Tolerance evaluation refers to evaluate the deviations of a ments. Functional dimensioning and tolerancing
part using the measured data obtained from CMMs or (CATIA.3D FDT) and functional tolerancing and anno-
other systems. Skin model shapes built in the observation tation (FT&A) for different versions of CATIA provide a
stage realize the statistical analysis of observed geometric syntactic and semantic tolerance checking of the anno-
deviations and guide to optimize inspection planning [17], tations on the active product or part with respect to
deviation domain decides the relations between the devi- tolerancing standards. Analytix can be used for full ki-
ation parameters that measured to inspect the specification nematic, inverse dynamic, and force analysis. It should
[117], and GapSpace assesses manufacturing dimensions be noted that Mechanical Advantage and Analytix are
according to an inspection scheme [38]. dimension-driven models which still have difficulty in
17. Series/parallel assembly, some models are not limit to analysis of complex geometric tolerances, especially for
the connections in series like TTRS [1], T-Map [115], situations where 3D models are used.
deviation domain [124], and Jacobian–torsor model;
however, the research of parallel assembly is not suffi-
cient, for example, Jacobian–torsor model just take par- As for the user interface, since most of the abovementioned
tial parallel structure into account and concerns only on packages are built-in tools of mainstream CAD systems, they
plane and cylinder features [16, 136]. typically share the same user interface with the CAD system.
18. Application, some of the eight models have been devel- While Analytix and PolitoCAT are standalone packages with
oped and applied successfully in CAD systems for in- their own GUI. Currently, most of the CAD packages can deal
dustrial environment, such as TTRS model and matrix with assemblies of varied complexities, among which CETOL

Table 17 Conclusions for the current CAT systems

Tool Providers CAD systems Models Analysis methods References

CETOL 6 Sigma Sigmetrix Creo/Solidworks/CATIA V5/NX Vector Loop WC/RSS/Monte Carlo [87, 160–163]
(TI/TOL)
eM-TolMate Tecnomatix/UGS CATIA/NX/ Creo /UG/SDRC/ Variation model Monte Carlo [164]
CADDS5
VSA Tecnomatix Creo /I-deas /CATIA/ UG Variation model Monte Carlo [87, 160–163]
3DCS Dimensional Control CATIA V5/NX/Creo/CATIA V6 Variation model WC/RSS/Monte Carlo/ [87, 160–163]
System 3DEXPERIENCE High-Low-Medium/
GeoFactor
CATIA.3D FDT Dassault System CATIA V4 TTRS/Matrix WC [87, 160–163]
FT&A Dassault System CATIA V5/V6 TTRS Semantic Analysis [55]
MECAmaster MECAmaster SARL CATIA V5/V6/Euclids3 Torsor WC/RSS [165]
Mechanical Cognition Mechanical Advantage CAD Dimension-driven WC/RSS/Monte Carlo [166]
Advantage
Analytix Saltire Software Microsoft Windows Dimension-driven WC/RSS [166]
FROOM Salomons CATIA TTRS/Torsor WC/GA [26]
CLIC Anselmetti Excel Software Positioning Tables WC/RSS [163, 167]
Sigmund Varatech Solidworks/Solid Edge/Creo Vector loop Worst case, RSS, MRSS, [168]
PCRSS, and Monte Carlo
PolitoCAT Teissandier Microsoft Windows Polytope Worst case [126]
/Politopix
Int J Adv Manuf Technol

6 Sigma and Sigmund are able to consider part/assembly var- (non-linear) constraints or the stack-up functions, and
iation and kinematic motion for kinematic mechanisms due to they maintain good results. However, this method re-
the adoption of vector loop model. quires a large amount of calculation with low efficiency.
In response, a method called NT-net [68] can provide a
better convergence rate over the MC method with im-
5 Conclusions proved accuracy. The random values generated by the
MC and NT-net are based on the theory of probability
The paper focuses on eight tolerance analysis models. The and statistics, which may result presumably in irrational-
related research on these models as well as the comparisons ity in many practical engineering problems due to uncer-
between them is discussed in detail. Each of these tolerance tain factors or gradient. Therefore, a non-probability dis-
analysis models has its own strengths and weaknesses, and tribution or fuzzy method is needed to represent the fuzzy
guidelines are presented for selecting the most proper one or unknown-but-bounded variables.
based on the specified requirements. 5. Skin model shape paradigm that is based on non-ideal
None of the models proposed in the literature that is avail- geometries and includes the consideration of form defects
able handle all the requirements included in the tolerancing enables the simulation of geometric deviations that are
standards [84], and many problems still exist in the tolerance expected, predicted, or already observed in real
analysis process. These limitations are considered to be sig- manufacturing processes. However, the complexity of
nificant future research projects, which are challenging but simulation and measurement of functional requirement
promising. Some of these issues are referred to in the paper restricts its use to simple assembly. The shortage in toler-
[2]; in addition, several aspects should be extended or supple- ance propagation for this model can be possibly made up
mented further. by combining with models that beneficial to tolerance
propagation, i.e., Jacobian model and polytopes. As a
1. All of the following problems have been referred to in consequence, the combined model will substantially facil-
some models more or less, but they still lack a universal itate tolerance representation and tolerance propagation at
method, which can completely and perfectly solve them: the same time.
the problems of applications of related requirements or an 6. Along with the application of updated information tech-
independence principle to assemblies, the representation nologies in industry and manufacturing, the big data-
of all the possible types of part couplings that include driven manufacturing era is coming [169]. A more realis-
clearance, the definition of the interaction of geometric tic virtual model is needed to serve as a digital twin of the
deviations with dimension deviations, and the choice for physical product in life cycle. The application of skin
the selection of datum precedences when a feature con- model shapes has been proposed [170], but still requires
fronts various datums. ongoing research.
2. Tolerance analysis conducted for assembly mainly focus- 7. Addictive manufacturing (AM) is a promising technology
es on static analysis, but a rotating mechanism, i.e., a disk with a capability for complex customized part construc-
cam or a slider mechanism, keeps a continuous movement tion. Dimensional and geometric tolerance analysis has a
under working conditions. The simulation of kinematic great impact on AM in production and industry. A model
behavior and the ability to specify motion tolerances are is necessary to represent the geometric variability and to
growing necessities. The related research on assembly support statistical analysis and compensation in addictive
sequences and relative positioning for motion tolerancing manufacturing. The skin model shape has potential to deal
is still deficient. with its shape deviations, yet further study is still needed
3. Research on tolerance analysis combines the processes of [171].
tolerance allocation, tolerance cost, and tolerance optimiza-
tion design. A simple application of the statistical analysis Acknowledgments This research was supported by the National Natural
Science Foundation of China (Nos. 51575484 and U1501248) and
has some limitations. The interaction between tolerance
Science Fund for Creative Research Groups of National Natural
analysis, tolerance allocation, and tolerance cost aggravates Science Foundation of China (No. 51521064).
the difficulty of the analysis process. The sensitivity and
percentage contribution of tolerances are general objects
for tolerance analysis which are still difficult to calculate
in some models, such as the TTRS and matrix models. And References
even the T-Map and deviation domain models, the analyses
in these two aspects are not developed in depth. 1. Ameta G, Samper S, Giordano M (2011) Comparison of spatial
4. Most of the models at present adopt the Monte Carlo math models for tolerance analysis: tolerance-maps, deviation do-
method (except the TTRS model) to deal with network main, and TTRS. J Comp Inf Sci Eng 11(2):255–267
Int J Adv Manuf Technol

2. Chen H, Jin S, Li Z, Lai X (2014) A comprehensive study of three 21. Jayaraman R, Srinivasan V (1989) Geometric tolerancing: I.
dimensional tolerance analysis methods. Comput Aided Des 53: Virtual boundary requirements. IBM J Res Dev 33(2):90–104
1–13 22. Saiki EM, Biringen S (1996) Numerical simulation of a cylinder in
3. Walter M, Breitsprecher T, Gruber G, Wartzack S Simulation uniform flow: application of a virtual boundary method. J Comput
based generation of an initial design taking into account geometric Phys 123(2):450–465
deviations and deformations. In: DS 68-10: Proceedings of the 23. Desrochers A (2003) A CAD/CAM representation model applied
18th International Conference on Engineering Design (ICED to tolerance transfer methods. J Mech Des 125(1):14
11), Impacting Society through Engineering Design, Vol. 10: 24. Desrochers A, Clément A (1994) A dimensioning and tolerancing
Design Methods and Tools pt. 2, Lyngby/Copenhagen, assistance model for CAD/CAM systems. Int J Adv Manuf
Denmark, 15.-19.08. 2011, 2011 Technol 9(9):352–361
4. Stuppy J, Meerkamm H Tolerance analysis of mechanisms taking 25. Desrochers A, Rivière A (1997) A matrix approach to the repre-
into account joints with clearance and elastic deformations. In: DS sentation of tolerance zones and clearances. Int J Adv Manuf
58-5: Proceedings of ICED 09, the 17th International Conference Technol 13(9):630–636
on Engineering Design, Vol. 5, Design Methods and Tools (pt. 1), 26. Salomons OW, Haalboom FJ, Poerink HJJ, Slooten FV, Houten
Palo Alto, CA, USA, 24.-27.08. 2009, 2009 FJAMV, Kals HJJ (1996) A computer aided tolerancing tool II:
5. Walter M, Sprügel T, Wartzack S (2013) Tolerance analysis of tolerance analysis. Comput Ind 31(2):175–186
systems in motion taking into account interactions between devi- 27. Thimm G, Britton GA (2002) A matrix method for calculating
ations. Proc Inst Mech Eng B J Eng Manuf 227(5):709–719 working dimensions and offsets for tolerance charting. Int J Adv
6. Schleich B, Wartzack S (2015) Tolerance analysis of rotating Manuf Technol 20(6):448–453
mechanism based on skin model shapes in discrete geometry. 28. Whitney DE, Gilbert OL Representation of geometric variations
Procedia CIRP 27:10–15 using matrix transforms for statistical tolerance analysis in assem-
7. Corrado A, Polini W (2017) Manufacturing digi in variational and blies. In: Robotics and Automation, 1993. Proceedings., 1993 I.E.
vector-loop models for tolerance analysis of rigid parts. Int J Adv International Conference on, 1993. pp 314–321 vol.312
Manuf Technol 88(5–8):2153–2161 29. Chase KW Tolerance analysis of 2-D and 3-D assemblies (auto-
8. Liu Y, Gao S, Wu Z, Yang J (2003) Hierarchical representation mated method)
model and its realization of tolerance based on feature. Jixie 30. Chase KW, Gao J, Magleby SP (1995) General 2-D tolerance
Gongcheng Xuebao(Chinese Journal of Mechanical analysis of mechanical assemblies with small kinematic adjust-
Engineering)(China) 39(3):1–7 ments. J Design Manuf 5:263–274
9. You X (2008) GapSpace multi-dimensional assembly analysis.
31. Gao J (1993) Nonlinear tolerance analysis of mechanical
Dissertations & Theses - Gradworks
assemblies
10. Jiang K, Davidson JK, Shah JJ, Liu J Using tolerance-maps to
32. Gao J, Chase KW, Magleby SP (1998) Generalized 3-D tolerance
transfer datum plane from design tolerancing to machining
analysis of mechanical assemblies with small kinematic adjust-
tolerancing. In: ASME 2013 International Design Engineering
ments. IIE Trans 30(4):367–377
Technical Conferences and Computers and Information in
33. Wang Y (2006) Semantic tolerance modeling—an overview. Proc
Engineering Conference, 2013. American Society of Mechanical
of Ierc’
Engineers, pp V004T005A023-V004T005A023
11. Adragna PA, Samper S, Favreliere H (2010) How form errors 34. Wang Y Semantic tolerance modeling. In: ASME 2006 interna-
impact on 2D precision assembly with clearance? Ifip Adv Inf tional design engineering technical conferences and computers
Comm Technol 315:50–59 and information in engineering conference, 2006. pp 261–273
12. Arroyave-Tobón S, Teissandier D, Delos V (2017) Applying 35. Wang Y (2008) Semantic tolerance modeling with generalized
screw theory for summing sets of constraints in geometric intervals. J Mech Des 130(81701):1–7
tolerancing. Mech Mach Theory 112:255–271. https://doi.org/ 36. Wang Y (2008) Closed-loop analysis in semantic tolerance model-
10.1016/j.mechmachtheory.2017.02.004 ing. J Mech Des 130(6):876–877
13. Teissandier D, Couétard Y, Gérard A (1999) A computer aided 37. Wang Y (2008) Semantic tolerance modeling based on modal
tolerancing model: proportioned assembly clearance volume. interval analysis. Proceedings of the International Workshop on
Comput Aided Des 31(13):805–817 Reliable Engineering Computing Rec’:46–59
14. Li H, Zhu H, Zhou X, Li P, Yu Z (2016) A new computer-aided 38. Zou Z, Morse E (2003) Statistical tolerance analysis using
tolerance analysis and optimization framework for assembling GapSpace. In: Geometric product specification and verification:
processes using DP-SDT theory. International Journal of integration of functionality. Springer, pp 105–114
Advanced Manufacturing Technology:1–12 39. Zou Z, Morse EP (2004) A gap-based approach to capture fitting
15. Corrado A, Polini W (2017) Manufacturing signature in Jacobian conditions for mechanical assembly. Comput Aided Des 36(8):
and torsor models for tolerance analysis of rigid parts. Robot 691–700
Comput Integr Manuf 46:15–24 40. Davidson JK, Mujezinović A, Shah JJ (2002) A new mathematical
16. Zeng W, Rao Y, Wang P, Yi W (2017) A solution of worst-case model for geometric tolerances as applied to round faces. J Mech
tolerance analysis for partial parallel chains based on the unified Des 124(4):609–622
Jacobian-Torsor model. Precis Eng 47:276–291 41. Davidson JK, Shah JJ (2002) Geometric tolerances: a new appli-
17. Schleich B, Anwer N, Mathieu L, Wartzack S (2014) Skin model cation for line geometry and screws. ARCHIVE Proc Insti Mech
shapes: a new paradigm shift for geometric variations modelling in Eng Part C J Mech Eng Sci 1989–1996 (vols 203–210) 216(1):
mechanical engineering. Comput Aided Des 50(3):1–15 95–103
18. Boyer M, Stewart NF (2010) Modeling spaces for toleranced ob- 42. Bhide S, Davidson JK, Shah JJ A new mathematical model for
jects. Int J Robot Res 10(5):570–582 geometric tolerances as applied to axes. In: ASME 2003 interna-
19. Gupta S, Turner JU (1993) Variational solid modeling for toler- tional design engineering technical conferences and computers
ance analysis. Comp Graphics Appl, IEEE 13(3):64–74 and information in engineering conference, 2003
20. Dantan JY, Mathieu L, Ballu A, Martin P (2005) Tolerance syn- 43. Mujezinović A, Davidson JK, Shah JJ (2004) A new mathematical
thesis: quantifier notion and virtual boundary. Comput Aided Des model for geometric tolerances as applied to polygonal faces. J
37(2):231–240 Mech Des 126(3):504–518
Int J Adv Manuf Technol

44. Clement A Theory and pratice of 3D tolerancing for assembly. In: 64. Liu SG (2007) Convolution method in statistical tolerance analy-
Proc. Of 2nd CIRP seminar on computer aided Tolerancing, 1991 sis. Mechanical Research & Application
45. Bourdet P, Mathieu L, Lartigue C, Ballu A (1996) The concept of 65. Evans DH (1974) Statistical tolerancing: the state of the art.
the small displacement torsor in metrology. Series on advances in Journal of Quality Technology
mathematics for. Appl Sci 40:110–122 66. Lin EE, Zhang HC (2001) Theoretical tolerance stack-up analysis
46. Lafond P, Laperriere L Jacobian-based modeling of dispersions based on tolerance zone analysis. Int J Adv Manuf Technol
affecting pre-defined functional requirements of mechanical as- 17(17):257–262
semblies. In: Assembly and Task Planning, 1999 (ISATP '99) 67. Shan A, Roth RN, Wilson RJ (1999) A new approach to statistical
Proceedings of the 1999 I.E. International Symposium on, 1999. geometrical tolerance analysis. Int J Adv Manuf Technol 15(3):
pp 20–25 222–230
47. Laperrière L, Lafond P Modeling tolerances and dispersions of 68. Huang W (2013) Sample size determination in NT-net quasi-
mechanical assemblies using virtual joints. In: Proceedings of Monte Carlo simulation. J Comput Inf Sci Eng 13(3):034501
ASME design engineering technical conferences, 1999. pp 12–15 69. Huang W, Ceglarek D, Zhou Z (2003) Tolerance analysis for de-
48. Adragna PA, Samper S, Pillet M, Favreliere H (2006) Analysis of sign of multistage manufacturing processes using number-
shape deviations of measured geometries with a modal basis. theoretical net method (NT-net). Int J Flex Manuf Syst 16(1):65–
Journal of Machine Engineering : Manufacturing Accuracy 90
Increasing Problems - Optimization 70. Khodaygan S, Movahhedy M, Foumani MS (2011) Fuzzy-small
49. Formosa F, Samper S (2007) Modal expression of form defects. degrees of freedom representation of linear and angular variations
Springer Netherlands, Modal Expression of Form Defects, in mechanical assemblies for tolerance analysis and allocation.
50. Samper S, Adragna PA, Favreliere H, Pillet M (2010) Modeling of Mech Mach Theory 46(4):558–573
2D and 3D Assemblies Taking Into Account Form Errors of Plane 71. Wu W, Rao SS (2006) Fuzzy analysis of geometric tolerances
Surfaces. J Comp Inf Sci Eng 9(4):790–792 using interval method. ARCHIVE Proc Instit Mech Eng Part C J
51. Mathieu L, Ballu A (2007) A model for a coherent and complete Mech Eng Sci 1989–1996 (vols 203–210) 220(4):489–497
tolerancing process. Springer Netherlands, A Model for a 72. Sun W, Dong R, Xu H (2009) A novel non-probabilistic approach
Coherent and Complete Tolerancing Process, using interval analysis for robust design optimization. J Mech Sci
52. Mathieu L, Ballu A (2007) A model for a coherent and complete Technol 23(12):3199–3208
tolerancing process. In: Models for computer aided Tolerancing in 73. Zhu H, Zhou X, Li H (2015) A novel tolerance analysis for me-
design and manufacturing. Springer, pp 35–44 chanical assemblies based on convex method and non-
53. Schleich B, Walter M, Wartzack S, Anwer N, Mathieu L A com- probabilistic set theory. Int J Adv Manuf Technol:1–9
prehensive framework for skin model simulation. In: ASME 2012
74. Panagopoulos I, Gorunova L, Bjerkehagen B, Boye K, Heim S
11th biennial conference on engineering systems design and anal-
(2014) A comparative study on tolerance analysis approaches.
ysis, 2012. American Society of Mechanical Engineers, pp 567–
Oncol Rep 32(1):40–44
576
75. Schleich B, Wartzack S (2016) A quantitative comparison of tol-
54. Zhang M, Anwer N, Mathieu L, Zhao H A discrete geometry
erance analysis approaches for rigid mechanical assemblies.
framework for geometrical product specifications. In:
Procedia CIRP 43:172–177
Proceedings of the 21st CIRP Design Conference, Kaist, MK
76. Shah JJ, Ameta G, Shen Z, Davidson J (2007) Navigating the
Thompson, ed., Paper, 2011. vol 20
tolerance analysis maze. Comp-Aided Design Appl 4(5):705–718
55. Anwer N, Schleich B, Mathieu L, Wartzack S (2014) From solid
modelling to skin model shapes: shifting paradigms in computer- 77. Mansuy M, Giordano M, Davidson JK (2013) Comparison of two
aided tolerancing. CIRP Ann Manuf Technol 63(1):137–140 similar mathematical models for tolerance analysis: T-map and
56. Schleich B, Anwer N, Mathieu L, Wartzack S Skin model shapes: deviation domain. J Mech Des 135(10):101008
offering new potentials for modelling product shape variability. In: 78. Marziale M, Polini W (2009) A review of two models for toler-
ASME 2015 International Design Engineering Technical ance analysis of an assembly: vector loop and matrix. Int J Adv
Conferences & Computers and information in engineering confer- Manuf Technol 43(11–12):1106–1123
ence (IDETC/CIE 2015), 2015 79. Marziale M, Polini W (2011) A review of two models for toler-
57. Ghie W, Laperrière L, Desrochers A (2003) A unified Jacobian- ance analysis of an assembly: Jacobian and torsor. Int J Comput
torsor model for analysis in computer aided Tolerancing. Springer Integr Manuf 24(1):74–86
Netherlands, A Unified Jacobian-Torsor Model for Analysis in 80. Marziale M, Polini W (2011) Review of variational models for
Computer Aided Tolerancing, tolerance analysis of an assembly. Proc Instit Mech Eng Part B J
58. Laperrière L, Ghie W, Desrochers A (2002) Statistical and deter- Eng Manuf 225(B3):305–318
ministic tolerance analysis and synthesis using a unified Jacobian- 81. Bo C, Yang Z, Wang L, Chen H (2013) A comparison of tolerance
Torsor model. CIRP Ann Manuf Technol 51(1):417–420 analysis models for assembly. Int J Adv Manuf Technol 68(1–4):
59. Khodaygan S, Movahhedy MR, Fomani MS (2010) Tolerance 739–754
analysis of mechanical assemblies based on modal interval and 82. Pasupathy TMK, Morse EP, Wilhelm RG (2003) A survey of
small degrees of freedom (MI-SDOF) concepts. Int J Adv mathematical methods for the construction of geometric tolerance
Manuf Technol 50(9–12):1041–1061 zones. Journal of Computing & Information Science in
60. Chase KW, Greenwood WH, Chase KW, Greenwood WH (2000) Engineering 3(1):64–75
Design issues in mechanical tolerance analysis. Manufacturing 83. Polini W (2011) Geometric tolerance analysis. Springer London,
Review Geometric Tolerance Analysis,
61. Seo HS, Kwak BM (2002) Efficient statistical tolerance analysis 84. Shen Z, Ameta G, Shah JJ, Davidson JK (2005) A comparative
for general distribution using three-point information. Int J Prod study of tolerance analysis methods. J Comp Inf Sci Eng 5(3):
Res 40(4):931–944 471–482
62. Skowronski VJ (1998) Calculating derivatives in statistical toler- 85. Nigam SD, Turner JU (1995) Review of statistical approaches to
ance analysis. Comput Aided Des 30(5):367–375 tolerance analysis. Comput Aided Des 27(1):6–15
63. Ilumoka A, Spence R (1982) A sensitivity-based approach to tol- 86. Nejad MK, Vignat F, Villeneuve F (2009) Tolerance analysis in
erance assignment. Iee Proceedings G 129(4):139–149 machining using the model of manufactured part (MMP)—
Int J Adv Manuf Technol

comparison and evaluation of three different approaches. Int J 109. He Y (2013) Generation of tolerance maps for line profile by
Comput Integr Manuf 25(2):136–149 primitive T-map elements. Arizona State University,
87. Prisco U, Giorleo G (2002) Overview of current CAT systems. 110. Clasen PJ, Davidson JK, Shah JJ Modeling of geometric varia-
Integr Comp-Aided Eng 9(4):373–387 tions within a tolerance-zone for circular runout. In: ASME 2009
88. Salomons OW, van Houten FJ, Kals HJ (1998) Current status of Int Design Eng Tech Conf Comp Inf Eng Conf 2009:597–604
CAT systems. In. Theories, standards and applications. Springer, 111. Ameta G, Davidson JK, Shah JJ (2007) Using tolerance-maps to
Geometric design tolerancing, pp 438–452 generate frequency distributions of clearance and allocate toler-
89. Hong YS, Chang TC (2002) A comprehensive review of ances for pin-hole assemblies. J Comput Inf Sci Eng 7(4):347–359
tolerancing research. Int J Prod Res 40(11):2425–2459 112. Ameta G, Davidson JK, Shah JJ (2010) Statistical tolerance allo-
90. Polini W (2012) Taxonomy of models for tolerance analysis in cation for tab-slot assemblies utilizing tolerance-maps. J Comp Inf
assembling. Int J Prod Res 50(7):2014–2029 Sci Eng 10(1):237–256
91. Desrochers A, Maranzana R (1996) Constrained dimensioning 113. Ameta G (2006) Statistical tolerance analysis and allocation for
and tolerancing assistance for mechanisms. Springer, Netherlands assemblies using Tolerance-Maps, vol 67. vol 11
92. Clément A (1998) The TTRSs: 13 constraints for dimensioning 114. Khan NS, Shah JJ, Davidson JK probability tolerance maps: a new
and Tolerancing. Springer US, statistical model for non-linear tolerance analysis applied to rect-
93. Desrochers A, Verheul S (1999) A three dimensional tolerance angular faces. In: ASME 2010 international design engineering
transfer methodology. Springer, Netherlands technical conferences and computers and information in engineer-
94. Salomons OW, Begelinger RE, Post E, Houten FJAMV (1999) ing conference, 2010. pp 529–548
Application of TTRS method in industrial practice tolerance spec- 115. Jaishankar LN, Davidson JK, Shah JJ, Jaishankar LN, Davidson
ification for industrial cooling water pumps. Springer, Netherlands JK, Shah JJ (2013) Tolerance analysis of parallel assemblies using
95. Liu J, Wilhelm RG (2003) Genetic algorithms for TTRS tolerance tolerance-maps® and a functional map derived from induced de-
analysis. In. Integration of Functionality. Springer, Geometric formations. American Society of Mechanical Engineers:
Product Specification and Verification, pp 73–82 V03BT03A008
116. Jiang K, Davidson JK, Liu J, Shah JJ (2014) Using tolerance maps
96. Jaballi K, Bellacicco A, Louati J, Riviere A, Haddar M (2011)
to validate machining tolerances for transfer of cylindrical datum
Rational method for 3D manufacturing tolerancing synthesis
in manufacturing process. Int J Adv Manuf Technol 73(1–4):465–
based on the TTRS approach R3DMTSyn. Comput Ind 62(5):
478
541–554
117. Giordano M, Samper S, Petit JP (2007) Tolerance analysis and
97. Salomons O, Jonge Poerink H (1995) Slooten Fv, Houten F. A
synthesis by means of deviation domains, axi-symmetric cases.
computer aided tolerancing tool based on kinematic analogies,
Springer Netherlands
Kals H
118. Mansuy M, Giordano M, Hernandez P (2011) A new calculation
98. Hua C (2016) A new approach of constraints establishment and
method for the worst case tolerance analysis and synthesis in
optimization for matrix tolerance model. J Mech Eng 52
stack-type assemblies. Comput Aided Des 43(9):1118–1125
99. Chase KW, Gao J, Magleby SP, Sorensen CD (2000) Including 119. Germain F, Denimal D, Giordano M (2008) A method for three
geometric feature variations in tolerance analysis of mechanical dimensional tolerance analysis and synthesis applied to complex
assemblies. IIE Trans 28(10):795–807 and precise assemblies. Springer US
100. Yan H, Cao Y, Yang J (2016) Statistical tolerance analysis based 120. Ziegler P, Wartzack S (2015) A statistical method to identify main
on good point set and homogeneous transform matrix. Procedia contributing tolerances in assemblability studies based on convex
Cirp 43:178–183 hull techniques. J Zhejiang Univ - Sci A: Appl Physics Eng 16(5):
101. Gao J, Chase KW, Magleby SP (1996) A new Monte Carlo sim- 361–370
ulation method for tolerance analysis of kinematically constrained 121. Ziegler P, Wartzack S (2015) Sensitivity analysis of features in
assemblies. Mechanical Engineering Department, Brigham Young tolerancing based on constraint function level sets. Reliability
University Eng [?] System Safety 134:324–333
102. Gao J, Chase KW, Magleby SP (1999) Comparison of assembly 122. Samper S, Petit JP, Giordano M (2007) Elastic clearance domain
tolerance analysis by the direct linearization and modified Monte and use rate concept applications to ball bearing and gears.
Carlo simulation methods. Adcatsetbyuedu Springer Netherlands
103. Glancy CG, Chase KW, Glancy CG, Chase KW A Second-order 123. Fukuda K, Petit J-P, d’Annecy ESDI, Annecy F Optimal
method for assembly tolerance analysis. In: Asme design automa- tolerancing in mechanical design using polyhedral computation
tion conference, 1999. pp 891–899 tools. In: 19th European workshop of Comput Geom, 2003. pp
104. Cao Y, Yan H, Liu T, Yang J (2016) Application of quasi-Monte 24–26
Carlo method based on good point set in tolerance analysis. J 124. Mansuy M, Giordano M, Hernandez P (2013) A generic method
Comp Inf Sc Eng 16(2):021008 for the worst case and statistical tridimensional Tolerancing anal-
105. Gao J, Chase KW, Magleby SP Global coordinate method for ysis. Procedia Cirp 10:276–282
determining sensitivity in assembly tolerance analysis. In: 125. Homri L, Teissandier D, Ballu A (2013) Tolerancing analysis by
Proceedings of the ASME international mechanical engineering operations on polytopes.597–604. doi:https://doi.org/10.1007/
conference and exposition, Anaheim, California, 1998 978-3-642-37143-1_71
106. Chase KW (1999) Minimum cost tolerance allocation. 126. Homri L, Teissandier D, Ballu A (2015) Tolerance analysis by
Department of Mech Engg, Bringham Young University, polytopes: taking into account degrees of freedom with cap half-
ADCATS Report (99–5) spaces. Comput Aided Des 62:112–130. https://doi.org/10.1016/j.
107. Ameta G, Davidson JK, Shah JJ The effects of different specifi- cad.2014.11.005
cations on the tolerance-maps for an angled face. In: Int Design 127. Gouyou D, Ledoux Y, Teissandier D, Delos V (2017) Tolerance
Eng Tech Conf, 2004. pp 303–311 analysis of overconstrained and flexible assemblies by polytopes
108. Ameta G, Davidson JK, Shah JJ tolerance-maps applied to a point- and finite element computations: application to a flange. Res Eng
line cluster of features. In: ASME 2005 international design engi- Des 29:55–66. https://doi.org/10.1007/s00163-017-0256-5
neering technical conferences and computers and information in 128. Gouyou D, Teissandier D, Delos V (2016) Tolerance analysis by
engineering conference, 2005. pp 497–507 polytopes: application to assembly interferences diagnosis.
Int J Adv Manuf Technol

Procedia CIRP 43:52–57. https://doi.org/10.1016/j.procir.2016. 151. Garaizar OR, Qiao L, Anwer N, Mathieu L (2016) Integration of
02.019 thermal effects into Tolerancing using skin model shapes. Procedia
129. Laperrière L, Lafond P (1999) Tolerance analysis and synthesis CIRP 43:196–201
using virtual joints 152. Homri L, Goka E, Levasseur G, Dantan J-Y (2017) Tolerance
130. Desrochers A, Laperrière L, Ghie W (2003) Application of a uni- analysis—form defects modeling and simulation by modal de-
fied Jacobian—torsor model for tolerance analysis. J Compu Inf composition and optimization. Comput Aided Des 91:46–59.
Sci Eng 3(1):2–14 https://doi.org/10.1016/j.cad.2017.04.007
131. Ghie W (2009) Statistical analysis tolerance using jacobian torsor 153. Schleich B, Wartzack S (2015) Approaches for the assembly sim-
model based on uncertainty propagation method. Int J ulation of skin model shapes. Comput Aided Des 65:18–33
Multiphysics 3(1):11–30 154. Chiabert P, Orlando M (2004) About a CAT model consistent with
132. Ghie W, Laperrière L, Desrochers A (2009) Statistical tolerance ISO/TC 213 last issues. J Mater Process Technol 157–158(51):61–
analysis using the unified Jacobian–torsor model. Int J Prod Res 66
48(15):4609–4630 155. Adragna PA, Samper S, Favreliere H (2010) How form errors
133. Ni W, Yao Z (2013) Integrating cylindricity error into tolerance impact on 2D precision assembly with clearance? In: Int
analysis of precision rotary assemblies using Jacobian–Torsor Precision Assembly Seminar:50–59
model. Proc Instit Mech Engi Part C J Mech Eng Sci 227(11): 156. Dumas A, Dantan JY, Gayton N, Bles T, Loebl R (2015) An
2517–2530 iterative statistical tolerance analysis procedure to deal with line-
134. Desrochers A (2007) Geometrical variations management in a arized behavior models. J Zhejiang Univ - Sci A: Appl Physics
multi-disciplinary environment with the Jacobian-torsor model. Eng 16(5):353–360
Springer, Netherlands 157. Ghie W, Laperrière L, Desrochers A (1970) Re-design of mechan-
135. Zuo XY, Li B, Yang J, Jiang X (2013) Application of the ical assemblies using the unified Jacobian-torsor model for toler-
Jacobian–torsor theory into error propagation analysis for machin- ance analysis. Springer Netherlands,
ing processes. Int J Adv Manuf Technol 69(5–8):1557–1568 158. Jaishankar LN, Davidson JK, Shah JJ, Jaishankar LN, Davidson
136. Hua C, Sun J, Li Z, Lai X (2015) A solution of partial parallel JK, Shah JJ Tolerance analysis of parallel assemblies using toler-
connections for the unified Jacobian–torsor model. Mech Mach ance-maps® and a functional map derived from induced deforma-
Theory 91:39–49 tions. In: ASME 2013 International Design Engineering Technical
137. Jin S, Chen H, Li Z, Lai X (2015) A small displacement torsor Conferences and Computers and Information in Engineering
model for 3D tolerance analysis of conical structures. Proc Inst Conference, 2013. p V03BT03A008
Mech Eng C J Mech Eng Sci 229(14):2514–2523 159. Giordano M, Adragna PA, Germain F (2006) Taking manufactur-
138. Shen W, Pang K, Liu C, Ge M, Zhang Y, Wang X (2015) The ing dispersions into account for assembly: modelling and simula-
quality control method for remanufacturing assembly based on the tion. Journal of machine engineering : manufacturing accuracy
Jacobian-torsor model. Int J Adv Manuf Technol 81(1):1–9 increasing problems - Optimization 6
139. Chen H, Jin S, Li Z, Lai X (2015) A modified method of the 160. Schleich B, Anwer N, Zhu Z, Qiao L, Mathieu L, Wartzack S
unified Jacobian-Torsor model for tolerance analysis and alloca- (2014) A comparative study on tolerance analysis approaches.
tion. Int J Prec Eng Manuf 16(8):1789–1800 International Symposium on Robust Design – Isord
140. Morse EP (2000) Models, representations, and analyses of 161. Amajd W (2013) Tolerance and Stack Analysis Manual and
toleranced one-dimensional assemblies. Cornell University, Jan. Advance approach
141. Morse EP (2005) Implementation of GapSpace analysis. Am Soc 162. Salomons OW, Kappert JH, Slooten FV, Houten FJAMV Kals
Mech Eng:329–333 HJJ computer support in the (re) design of mechanical products.
142. Zou Z, Morse EP (2003) Applications of the gapspace model for In: Ifip Tc5/wg53/ifac Int Working Conf Knowledge Based
multidimensional mechanical assemblies. J Comput Inf Sci Eng Hybrid Syst Eng Manuf 1993:91–103
3(1):22–30 163. Anselmetti B, Chavanne R, Yang JX, Anwer N (2010) Quick
143. Zhang M (2011) Discrete shape modeling for geometrical product GPS: a new CAT system for single-part tolerancing. Comput
specification: contributions and applications to skin model simu- Aided Des 42(9):768–780
lation. Cachan Ecole Normale Supérieure 164. Chiesi F, Governi L (2003) Tolerance analysis with eM-TolMate. J
144. Zhang M, Anwer N, Stockinger A, Mathieu L, Wartzack S (2013) Comp Inf Sci Eng 3(1):100–105
Discrete shape modeling for skin model representation. Proc Inst 165. Corrado A, Polini W (2016) Computer-aided techniques for ge-
Mech Eng B J Eng Manuf 227(5):672–680 ometry assurance. In: Computer-aided technologies-applications
145. Anwer N, Ballu A, Mathieu L (2013) The skin model, a compre- in engineering and medicine. InTech
hensive geometric model for engineering design. CIRP Ann 166. Islam MN (2004) Functional dimensioning and tolerancing soft-
Manuf Technol 62(1):143–146 ware for concurrent engineering applications. Comput Ind 54(2):
146. Schleich B, Wartzack S (1991) Evaluation of geometric tolerances 169–190
and generation of variational part representatives for tolerance 167. Anselmetti B (2006) Generation of functional tolerancing based
analysis. Nagoya. Math J 123(9):1–12 on positioning features. Comput Aided Des 38(8):902–919
147. Samper S, Adragna PA, Favreliere H, Pillet M (2010) Modeling of 168. Barbero BR, Aragón AC, Pedrosa CM (2015) Validation of a
2D and 3D assemblies taking into account form errors of plane tolerance analysis simulation procedure in assemblies. Int J Adv
surfaces. Computer. Science 9(4):790–792 Manuf Technol 76(5–8):1297–1310
148. Yan X, Ballu A (2017) Generation of consistent skin model shape 169. Tao F, Cheng J, Qi Q, Zhang M, Zhang H, Sui F (2017) Digital
based on FEA method. Int J Adv Manuf Technol 92(1–4):789– twin-driven product design, manufacturing and service with big
802 data. Int J Adv Manuf Technol:1–14
149. Schleich B, Wartzack S (2012) Generation of deviated geometry 170. Schleich B, Anwer N, Mathieu L, Wartzack S (2017) Shaping the
based on manufacturing process simulations Weimarer Stochastik- digital twin for design and production engineering. CIRP Ann
und Optimierungstage WOST 9 66(1):141–144
150. Schleich B, Anwer N, Mathieu L, Wartzack S (2015) Contact and 171. Zhu Z, Anwer N, Mathieu L (2017) Deviation modeling and shape
mobility simulation for mechanical assemblies based on skin mod- transformation in design for additive manufacturing. Procedia
el shapes. J Comp Inf Sci Eng 15(2):979–985 CIRP 60:211–216

You might also like