See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/302869211

Comparison of Optimization Techniques in a Tolerance Analysis Approach

Considering Form Defects

Article  in  Procedia CIRP · December 2016

DOI: 10.1016/j.procir.2016.02.022

CITATIONS READS

5 215

3 authors:

Lazhar Homri Jean-Yves Dantan

Ecole Nationale Supérieure d'Arts et Métiers Ecole Nationale Supérieure d'Arts et Métiers
21 PUBLICATIONS   124 CITATIONS    126 PUBLICATIONS   1,748 CITATIONS   

SEE PROFILE SEE PROFILE

Guillaume Levasseur
Université Libre de Bruxelles
4 PUBLICATIONS   40 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

AHTOLAND (Advanced Hybrid TOLerance ANalysis) View project

Integrated Product and Process Design View project

All content following this page was uploaded by Lazhar Homri on 11 May 2016.

The user has requested enhancement of the downloaded file.

 Available online at www.sciencedirect.com

ScienceDirect
Procedia CIRP 43 (2016) 184 – 189

14th CIRP Conference on Computer Aided Tolerancing

Comparison of optimization techniques in a tolerance analysis

approach considering form defects
Lazhar Homri*, Jean-Yves Dantan, Guillaume Levasseur
LCFC, Arts et Métiers ParisTech, ENIM, HESAM, 4 rue Augustin Fresnel, 57078 METZ CEDEX 3, France

*Corresponding author. Tel.: +33387375430; fax: +33 3 87375470. E-mail address : [email protected]

Abstract

In tolerancing analysis area, the most various existing approaches do not take form defects of parts into consideration.
As high precisions assemblies cannot be analyzed with the assumption that form defects are negligible, the paper
focuses in particular on the study of the form defects impacts on the assembly simulation and that by comparing two
optimization algorithms (iHLRF and Quapro). The study is limited firstly to the cylinders. For the optimization, two
main types of surfaces modelling are considered: difference surface-based method and real model. The compared
models allow assessing the non-interferences between cylinders with form defects, potentially in contact. This is in the
main issue to validate a tolerance analysis approach.
Authors. Published by Elsevier B.V.
© 2016 The Authors. B.V. This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of the organizing committee of 14th CIRP conference on Computer Aided Tolerancing
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of the 14th CIRP Conference on Computer Aided Tolerancing
Keywords: Tolerance analysis; form defects, optimization, iHLRF, difference surface

1. Introduction To simulate the geometric deviations, various

existing approaches (traditionally, statistical or by
It is impossible to manufacture perfectly formed domains) consider mostly that form defects of
parts. There will always be inaccurate dimensions and individual parts in the assembly are neglected and are
imperfect surfaces. Although the inaccuracies and not taken into consideration. However, the non-
imperfections might be reduced till they are negligible consideration of form defects of parts could lead to
the production cost would be raised prohibitively. noncompliant assemblies though all parts respect the
However, the designer can allocate tolerances which geometrical specifications [3]. More generally, this
state how much variation from the nominal shape can consideration in the tolerancing activities, allows to
be tolerated if the part is to function satisfactorily. highlight the form defects impacts when computing
Tolerance analysis of mechanical assemblies, the probability of assembly and when assessing the
essential part of design and manufacturing, is used to quality and the cost of products [4,5].
highlight their effects on assembly quality. The This paper aims to highlight the impact of form
appropriate allocation of tolerances among the defects consideration on the assembly simulation via
separate parts in an assembly can result in lower costs mathematical optimization. In the following, we limit
and higher probability of fit, reducing the number of ourselves to cylinders example and cylindrical joints.
rejects. Analyzing the cumulative effects of these Many different modes of form defects can be
tolerances on critical clearances or fits in the assembly generated in these surfaces regarding to their two main
is necessary to ensure the compliance of a product in features (axis and radius). In the first part of the
terms of functional conditions [1,2]. article, an overview of the tolerance analysis issues
and form defects consideration is given; the second

2212-8271 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of the 14th CIRP Conference on Computer Aided Tolerancing
doi:10.1016/j.procir.2016.02.022
 Lazhar Homri et al. / Procedia CIRP 43 (2016) 184 – 189 185

part deals with the consideration of form defects when [12]. These relations resulted from the topological
assembly simulation. A comparison between two loops of the graph. The mathematical representations
optimization algorithms applied on a cylindrical joint include some specific dimensional parameters, gaps
is proposed. An application of tolerance analysis and tolerances. The behavior modeling of a planar
approach taking form defects on an over-constrained joint taking form defects into consideration can be
mechanism is described at the end of the article. We modeled by the concept of gap hull and the
put forward the following physical hypotheses: introduction of the difference surface concept [13].
• surfaces are considered with form defects and Other matting solutions for circular or prismatic joints
can be found in [4].
are identified by their meshes,
A tolerance analysis approach should define the
• no local strain in surfaces in contact, mathematical formulation that can be able to take into
• no deformable parts. account all the characteristics of the behavior model.
Different methods have been proposed and are
2. Mechanical tolerance analysis issues and form generally classified into two main categories:
defects consideration statistical approaches and deterministic approaches.
• Statistical approach: the goal is to compute the
Geometrical deviation or variation is defined by a probability that all requirements can be
displacement between two any surfaces in a satisfied under given individual tolerances
mechanical system. Each real surface can be modeled [14,15].
by a substitution surface. The model of substitution • Deterministic approach: the goal is to
surfaces is based on the assumption that form defects determine the dimensions and the tolerances
are generally negligible. Generally, the relative such that any possible combination produces a
position (position, orientation) between any two functional assembly. We can for example
surfaces in the mechanical system is determined by the mention the models based on geometrical
relative position between two substituted surfaces. The constraints handling. Associated mathematical
model will not be selected for surfaces modeling when representations can thus be defined by
dealing with form defects. domains [16], polytopes [17] or T-
Skin model provides a global representation of the Maps®[18,19].
real surfaces including all deviations [6]. The
representation of the skin model was investigated 3. Form defects consideration in a cylindrical joint
recently. A framework for skin model simulation was analysis
developed by Schleich [7]. Some examples of skin
model representation can be found in [8]. In this part, elementary form defects of cylinders
For tolerancing activities, form defects are dealt in are generated by translation application on the meshes-
different existing models and are generated according nodes of the surfaces. Different modes are thus built
to different techniques. Huang proposed a based and a global form defect is expressed as a combination
decomposition method called Discrete-Cosine these elementary modes. The amplitudes of the modes
Transformation (DCT) to represent various are randomly defined. Fig. 1 illustrates a cylindrical
manufacturing defect patterns with different modes pair joint including form defects. Cylinders are
[9]. Raja and Radhakrishnan introduced the Fourier assigned by numbers 1 and 2 for the shaft (red) and the
series decomposition method based on Discrete bore (blue) respectively.
Fourier Transform (DFT) for discrete objects [10] in
For the assembly analysis, the same mesh size
order to describe the forms of a cylinder and identify
(same number of nodes) is considered. The nodes are
specific types of defect shapes. Samper et al. [5,11]
here assigned by a one-to-one application between the
developed the Discrete Modal Decomposition (DMD),
based on the eigen-shapes of natural vibrations of two cylinders and are expressed in a given direction.
surfaces. In all reviewed techniques, a global Therefore, the nominal cylinders have different
generated form defect in a surface is defined by a lengths so the longest cylinder is truncated and re-
random combination of the elementary modes. meshed in order to take into account only the
The behavior modeling, an element key of functional surface in the contact.
tolerance analysis approaches, consists at the the concept of clearance torsor (SDT) [20] aims
definition of the mathematical models which generally to characterize the relative displacements
characterize the system behavior with deviations. between surfaces potentially in contact. The clearance
Building a behavior model allows to know how torsor associated to this analysis is defined by (1):
features of mechanisms interact. Here, mathematical
formulation (set of constraints) can be deduced from a
simplified characterization of the behavior by a graph
186 Lazhar Homri et al. / Procedia CIRP 43 (2016) 184 – 189

­α u ½ one point in the associated difference surface.

{T}O = °® β v °¾ (1)
°0 0°
¯ ¿O
Where u, v and α , β define respectively the
translations at a point O and the rotations along x and
y -axis.

Fig. 2. Distance between surfaces with form defects consideration

The distances between surfaces considering the

difference surface-based method and the real model
are respectively denoted by δ df = N1Mdf and
δ df = M1M2 . Analytically, the equivalence between
the two models can be expressed by (2):
δ r = M1 M 2 = M1N1 + N1 N 2 + N 2 M 2
Fig. 1. (a) Nominal surfaces, (b) surfaces with form defects (2)
= N1 N 2 + u 2 − u1 = N1 N 2 + N 2 M df = δ df
The contact between surfaces is assessed by
The optimization problem introduced in this part
mathematical optimization which consists to verify, by
allows determining the clearance torsor components
determining the clearance torsor components, if the
(1) and requires the definition of the signed distance
two cylinders interfere with each other. Two types of
between cylinders, see fig. 3. Negative values of the
surfaces modelling are considered for optimization: a
distance represent the interferences between surfaces.
real model defined by the real surfaces with their form
Nevertheless, the occurrence of interferences in the
defects and a difference surface-based model
initial configuration does not imply the non-assembly,
associated to the building of a “difference surface”
but an analysis should be carried out. The signed
which can takes into consideration all stacked form
distance denoted d is defined by the scalar product:
defects of surfaces in contact.
d = M1 M 2 .n (3)
3.1. Surfaces modelling in a tolerance analysis Where n the outward pointing vector normal to the
approach surface C1 in the point M1 .

The main objective of the proposed difference

surface-base method is to simplify the tolerance
analysis approach by considering only superimposed
form defects of the two cylinders when assessing the
contact position. The distance can be thus computed
between the points of the difference surface and one of
the perfect cylinders I accordance with the following
rule:
Rule: the nominal surface assigned by an even index
is considered to define the difference surface with
superimposed form defects; the nominal surface
Fig. 3. Interference between cylinders in contact
assigned by an odd index remains perfect.
In a second way, the real model can be also 3.2. Optimization problem : simple joint
considered in a tolerance analysis approach. The two
models are generally equivalents. The equivalence was The aim of this section is to deal with two
recently proved for a 2D planar joint by Samper et al. optimization algorithms: iHLRF and Quapro. The
[21]. Here, let C1 and C 2 two nominal cylinders in algorithms will be compared in terms of computing
contact and let N 1 ∈ C1 and N 2 ∈ C 2 . M1 and M 2 time and of results accuracy.
define the correspondent points in the cylinders with Zhang and Kiureghian [22] developed the iHLRF
form defects as shown in fig. 2. We denote by M df algorithm by introducing a non-differentiable merit
 Lazhar Homri et al. / Procedia CIRP 43 (2016) 184 – 189 187

function and by using the Armijo rule to select the step The fig. 5 shows that iHLRF algorithm has a linear
size in the linear search. Quapro algorithm is a linear computing time representation however it is difficult
quadratic programming solver proposed in the to predict the computing time of Quapro. The
ATOMS module. It is based on the QR-decomposition probability of assembly can be thereafter evaluated via
method and the Chloesky factorization of the Hessian a Monte Carlo method with an appropriate numbers of
matrix of constraints for optimization [23]. simulations equal to 1000 points randomly generated
Using these two algorithms, the above signed and that for two cases of ISO fits, see Table 1.
distance (3) is to be minimized or to made zero. Fig. 4 Table 1. Assembly probabilities for iHLRF and Quapro algorithms
illustrates the application of the iHLRF algorithm on a
simple cylindrical joint and considering difference iHLRF Quapro ΔP
surface-based model. The algorithm converges just Case 1 56,4% 57,6% 1,2%
after 8 iterations and the clearance torsor components Case 2 51,6% 54,0% 2,4%
are determined and controlled according to the
numerical value of the signed distance. The table 1 shows two different levels of convergence
of the two algorithms. The difference remains no
significant ( ΔP < 3% ) . This returns to the ability of
Quapro to control and to manage the degenerated
points. Quapro can be therefore used to determine the
torsor components (1) when a cylindrical joint
analysis is asked.
In order to evaluate the form defects impacts on the
optimization problem, a numerical analysis is
performed. For small form defects, the distance is
lightly positive then it becomes increasingly negative.
If the form defect exceed 0.00525 mm, the distance
can takes values below −10−4 . The numerical study
takes into consideration the two models of surfaces.

Fig. 4. Perfect cylinder C1 and displacement of the difference surface 4. Comparison between the two algorithms:
Cdf after optimization application to an over-constrained mechanical
Quapro allows determining all the components of system
the clearance torsor with a precision around 10 −4 mm .
It includes a stopping criterion and converges in 80 The considered optimization algorithms are in this
iterations, which means 10 times more than iHLRF section applied on a simplified example assembly of
algorithm. The Quapro programming has the an industrial electrical connector. The mechanical
advantage that it not depends on a gradient calculation system is made up of two parts on which 4 cylindrical
as iHLRF. It is here clear that the two algorithms pair joints (4 pins must get into their pin holes) are
converge similarly, a comparison between the defined. By considering only functional surfaces, each
computing time of the two algorithms in accordance to surface can be characterized by a cylinder.
the meshes sizes is give in the fig. 5. The comparison between the optimization
algorithms proposed for a simple cylindrical joint in
the previous part of the article will be reapplied here.
The clearance torsor is expressed in a global base of
the system and it characterizes the relative
displacements between the “holes” part and the “pins”
part.
The four cylinders are parallels. That can remove a
degree of freedom defined by the rotation along z −
axis. The clearance torsor, defined in a global base,
can therefore be written as (4):
­α u ½
° °
{T}O = ® β v ¾ (4)
°γ 0 °
Fig. 5. Computing time comparison ¯ ¿O
188 Lazhar Homri et al. / Procedia CIRP 43 (2016) 184 – 189

Where O is the origin of the base. distance value appears higher particularly for the
To deal with the assembly issue, only one iHLRF optimization. So although iHLRF is faster than
functional surface is considered. In order to ensure that Quapro algorithm, it has a much lower accuracy. The
the calculated clearance torsor by optimization is fig. 7 illustrates the final optimization result of the
efficient, the accuracy of the contact after assembly including form defects and giving an
displacement is assessed and the interferences between optimized contact position.
surfaces are then identified. Our proposed method can
be summarized in five steps: (1) generation of surfaces
of the connector, 4 cylinders with form defects are
only considered for each part, (2) linear optimization
application using Quapro, the objective is to determine
the clearance torsor components, (3) displacement of
the “pins” part (interior surface colored blue) using the
optimized torsor, (4) determination of the contact
position between the two parts of the connector, (5)
computing of the distance simulating the contact.
A simplified representation of surfaces with form
defects can be illustrated in the fig 6.

Fig. 7. Assembly optimization of the connector

Fig. 7. Assembly optimization of the connector

A Monte Carlo simulation is finally applied for the

assembly analysis. 5000 test examples of the
mechanism with random form defects are considered.
Among the various simulations, 4912 assemblies are
been successful. Otherwise, the probability of
assembly is equal to 98.24%.

5. Conclusion and future research

Form defects of parts are considered in all issues of

mechanical tolerance analysis approach. The paper
Fig. 6. Form defects consideration in optimization issue
focuses on the study of their impact on the assembly
simulation by proposing a comparison between two
The optimization is in this part performed via the main algorithms (iHLRF and Quapro).
two algorithms that are compared in 20 test examples In the first part of the paper, an investigation about
the introduction of form defects of parts in the
where form defects are randomly generated. The table
assembly simulation is proposed. Only cylindrical
2 shows the comparison in terms of computing time
joint analysis was considered. The analysis concerns
and accuracy of the optimized contact position.
the determination of the contact position via
Table 2. Accuracy and computing time of the algorithms mathematical optimization. Two algorithms are so
compared in terms of accuracy and computing time.
iHLRF Quapro The objective of the algorithms is merely to determine
the clearance torsor components when the signed
Order of magnitude 10-4 to 10-3 10-2 to 10-1
distance between parts potentially in contact is
of the distance mm mm
minimized or reduced to zero.
Computing time (30 to 150)s (20 to 40)s Finally, the comparison between the two algorithms
is applied to an over-constrained mechanism in which
The analysis highlights another challenge when surfaces are considered with form defects.
considering several joints in a mechanism: precision A prototype of tolerance analysis considering form
degradation. Indeed, the definition of the two defects of parts is currently being developed. Any
algorithms is based on a fixed stopping criterion. The class of surface will be thus considered. With this tool
 Lazhar Homri et al. / Procedia CIRP 43 (2016) 184 – 189 189

it will be possible to take form defects into account in Surface Profiles using Fourier Series, Int. J. Mach. Tool
Des. Res. 17 (1977) 245–251.
a global tolerance analysis approach of complex [11] S. Samper, F. Formosa, Form Defects Tolerancing by
mechanical systems. Natural Modes Analysis, J. Comput. Inf. Sci. Eng. 7 (2006).
[12] A. Ballu, L. Mathieu, O. Legoff, Representation of
Acknowledgment Mechanical Assemblies and Specifications by Graphs, in:
Geom. Toler. Prod., ISBN 978-1-84821-118-6, ISTE-
WILEY, 2010: pp. 87–110.
The authors would like to acknowledge the support [13] H.N. Lê, Y. Ledoux, A. Ballu, Experimental and theoretical
of FUI “AHTOLAnd” project. investigations of mechanical joints with form defects, J.
Comput. Inf. Sci. Eng. 14 (2014).
[14] S.D. Nigam, J.U. Turner, Review of statistical approaches of
References tolerance analysis, Comput.-Aided Des. 27 (1995) 6–15.
[15] P. Beaucaire, N. Gayton, E. Duc, J.Y. Dantan, Statistical
tolerance analysis of overconstrained mechanisms with gaps
[1] K.W. Chase, W.H. Greenwood, Design issues in mechanical using system reliability methods, Comput. Aided Des. 45
Tolerance analysis, Manuf. Rev. 1 (1987) 50–59. (2013) 47–55.
[2] J. Guilford, J.U. Turner, Advanced tolerance analysis and [16] M. Giordano, S. Samper, E. Pairel, Tolerance Analysis and
synthesis for geometric tolerances, in: Proc. Int. Forum Synthesis, Method of Domains, in: Geom. Toler. Prod.,
Dimens. Toler. Metrol., 1993: pp. 187–198. ISBN 978-1-84821-118-6, Iste-Wiley, 2010: pp. 152–181.
[3] J. Grandjean, Y. Ledoux, S. Samper, On the role of form [17] L. Homri, D. Teissandier, A. Ballu, Tolerance analysis by
defects in assemblies subject to local deformations and polytopes: Taking into account degrees of freedom with cap
mechanical loads, Int. J. Adv. Manuf. Technol. 65 (2013) half-spaces, Comput. Aided Des. 62 (2015) 112–130.
1769–1778. [18] J.K. Davidson, A. Mujezinovic, J.J. Shah, A new
[4] R.S. Pierce, D. Rosen, A method for integrating form errors mathematical model for geometric tolerances as applied to
into tolerance analysis, J. Mech. Des. 130 (2007). round faces, Asme Trans. J. Mech. Des. 124 (2002) 609–
[5] P.A. Adragna, H. Faverlière, S. Samper, M. Pillet, Statiscal 622.
assemblies with form errors - a 2D example, in: Micro- [19] A.D. Jian, G. Ameta, J.K. Davidson, J.J. Shah, Tolerance
Assmbly Technol. Appl., (Boston: Springer), Ratcchev, S., analysis and allocation using Tolerance-Maps for a power
Koelmeijer, S., 2008: pp. 23–33. saw assembly, in: ISBN 978-1-4020-5437-2, Springer,
[6] A. Ballu, L. Mathieu, Analysis of Dimensional and Tempe (Arizona - USA), 2007: pp. 267–276.
Geometrical Specifications: Standards and Models, in: Proc. [20] P. Bourdet, L. Mathieu, C. Lartigue, A. Ballu, The concept
3rd Cirp Semin. Comput. Aided Toler., Cachan (France), of the small displacement torsor in metrology, Ser. Adv.
1993. Math. Appl. Sci. Adv. Math. Tools Metrol. Ii. 40 (1996)
[7] B. Schleich, N. Anwer, L. Mathieu, M. Walter, S. Wartzack, 110–122.
“A Comprehensive Framework for Skin Model Simulation, [21] S. Samper, P.A. Adragna, H. Favreliere, M. Pillet, Modeling
in: 11th Biennel Conf. Eng. Syst. Des. Anal., ASME (Ed.), of 2D and 3D assemblies taking into account form errors of
Nantes-France, 2012: pp. 567–576. plane surfaces, J. Comput. Inf. Sci. Eng. 9 (n.d.) 2009.
[8] J.Y. Dantan, Comparison of Skin Model Representations [22] Y. Zhang, A.D. Kiureghian, Finite element reliability
and Tooth Contact Analysis Techniques for Gear Tolerance methods for inelastic structures, Department of Civil and
Analysis, J. Comput. Inf. Sci. Eng. 15 (2015). Environmental Engineering, University of California,
[9] W. Huang, D. Ceglarek, Mode-based decomposition of part Berkeley, 1997.
form error by discrete-cosine-transform with [23] E. Casas, C. Pola, An algorithm for indefinite quadratic
implementation to assembly and stamping system with programming based on a partial Cholesky factorization,
compliant parts, Cirp Ann. - Manuf. Technol. 51 (2002) 21– Rev. Française Autom. Inf. Rehcerche Opérationnelle. 7
26. (1993) 401–426.
[10] J. Raja, V. Radhakrishnan, Analysis and Synthesis of

View publication stats