Professional Documents
Culture Documents
com
ScienceDirect
Procedia CIRP 43 (2016) 76 81
Abstract
When talking about a large batch production of a certain mechanical product it generally comes along with specic requirements in terms of di-
mensional management to guarantee a consistently high product quality ensuring that customer requirements are met. Therefore, this contribution
is about the eect that the elasticity of the observed assembly has on the formation of the limiting positions of the considered key product charac-
teristics arising from tolerance analysis. On the example of an automotive door assembly a developed approach based on existing methodologies
is explained in detail containing CAT and FEM simulations.
c 2016
2016 The
The Authors.
Authors. Published
Published by
by Elsevier
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.
Peer-review under responsibility of the organizing committee of the 14th CIRP Conference on Computer Aided Tolerancing
Keywords:
tolerancing, computer aided tolerancing, geometric variation management, robust design, simulation, FEM
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.2016.02.150
Matthias Ehlert et al. / Procedia CIRP 43 (2016) 76 81 77
to verify the use of the method of inuence coecients in this CAT Simulation
context, see [12]. component tolerances
number of simulation runs n
approx. 10000 simulations
assembly tolerances limiting
K
joint tolerances positions
mounting tolerances
3. Simulation approach for tolerance calculations taking
1 PPP
into account non-linear eects
Statistical Design of Experiments, DoE
3.1. Presentation of existing and developed approaches limiting positions minimisation of the number of simulation
stiffness parameter combinatorics of limiting positions specifications
rived that dierent basic strategies exist, able to couple the re- FEM Simulation
3D System Simulation (side frame, sealing
sults from several simulations. There is a distinction to be made system, door, mounting of the sealing system
whether if this coupling is performed using statistical moments, number of simulation runs m statistical
simulation approx. 100 simulations basis data
such as estimated mean and standard deviation, or superposition specifications concerning
of simulation data or through direct integration of the simula- the system
K
tion methods. Due to the fact that direct source code adjust- behavior
ments are not an option and the simulation eort will increase
1 PPP
dramaticly using the Monte-Carlo-method as a state-of-the-art
tool in tolerance management, the latter mentioned method will Fig. 3. Basic process taking into consideration interactions between geometri-
not be used. Hence, the basic strategies shown in Fig. 2 can be cal deviations and system stiness
derived. It may be added that a MIC (see [21]) is not applicable
as a result of geometrical as well as material-related and contact
conditions of the observed automotive door assembly.
In order to minimize the number of required FEM simulation
runs, a statistical Design of Experiments (DoE) is performed,
A: B: C:
Estimated Mean Shift Statistical Moments Direct Source Superposition see [9]. Investigated and applicable methods are D-Optimal
process CAT DoE FEM CAT DoE FEM 1. CAT DoE FEM Designs as well as Latin Hypercube Sampling (LHS). Using
2. RSM CAT+ a DoE, simulation parameters and limiting position information
1 1 1
K
K
K
provide guidelines for proper simulation runs.
CAT Within the FEM simulations, the assemblies side frame,
method PPP PPP PPP door, sealing system as well as other stiness-relevant parts / as-
K
K
K
K
K
Fig. 2. Approaches for coupling of statistical simulations Approach A: Estimated Mean Shift
In comparison with a rigid CAT simulation, this approach
Overall, the presented approaches have one thing in common enables the user to calculate an overall distribution with a mean
- they are founded on a basic process that considers interac- shift for every KPC. This can be achieved through an arithmetic
tions between geometrical deviations and system stiness, see superposition of the averages in combination with the maxi-
Fig. 3. This means, the anticipated limiting positions are calcu- mum statistical variance coming from tolerance and FEM sim-
lated from a statistical consideration of the rigid overall system. ulation. The outcome is a substitue distribution that is described
As a result from these simulations, the limiting positions for all as a function of the parameters average and variance. This ap-
functionally relevant KPCs can be determined. proach is based on the hypothesis that the elastic deformation is
Due to a transformation, all information concerning the lim- not increasing or decreasing the variation in comparison to the
iting positions of the alignment points can be converted clearly result of the rigid system simulation.
to displacement boundary conditions (translation and rotation)
needed for FEM simulations. Derived specications for the Approach B: Statistical Moments
limiting positions occur. These specications are varied in ac- Performing a statistical superposition of the mean values as
cordance with typical FEM input parameters, such as Shore well as of the variances, using the Root-Sum-Square method
hardness of the sealing system, elastic modulus of the involved (see [8]), an overall distribution with a mean shift can be cal-
parts and friction coecient to model the process of door clos- culated within this approch. Analogue to Approach A the out-
ing. This means, a set of parameters appears containing derived come is a substitue distribution that is described as a function of
limiting positions and stiness aecting parameters. the parameters average and variance. This approach is based on
Matthias Ehlert et al. / Procedia CIRP 43 (2016) 76 81 79
the hypothesis that the elastic deformation causes an increase Table 1. Qualitative (dis-)advantages of the approaches
of the variance compared to the rigid system. Using the RSS
Advantages Disadvantages
method, this increase becomes a contributor of the statistical
calculation of the KPC if it is ensured that it acts as an indepen- Approach A: + numerically - correlations will
dend contributor (law of error propagation, see [15]). Estimated Mean Shift less expensive not be preserved
+ rapid superposing - no predications
of the results about statistical
Approach C: Direct Source Superposition + low FEM eort variations from
If the outcome both from CAT and FEM simulations is avail- (only a few simula- FEM simulation
able as directly allocatable data series, a direct numerical super- tions necessary;
position of the source data can be performed. As a rst result Condence interval
for the mean)
of the direct allocation of the linked data series, a distribution is
Approach B: + numerically - correlations will
generated (e.g. visualized by a histogram). It is characterized Statistical Moments less expensive not be preserved
by a mean shift and a variance that can increase or decrease + rapid superposing - form of the result
compared to the rigid system. Second, this overall distribution of the results distribution
can be described generally as a function of the parameters av- o medium FEM eort
(Condence interval
erage, variance, skewness and kurtosis. Compared to the rigid
for the standard
system, this approach is based on the hypothesis that the elastic deviation)
deformation is not interacting too strong with the deviation sit- Approach C: + correlations and - numerically
uation, which in turn would mean that the outcomes of the CAT Direct Source result distributions expensive
simulations (geometrical states) are wrong. Hence, this would Superposition will be preserved - high consistency
+ simple, direct requirements
be a problem due to the fact that the input for the FEM sim-
numerical super- on DoE
ulations are generated on the basis of the corresponding CAT position of the - RSM model has a
simulations. source data signicant eect
Proceeding in this way, a few hundred input data sets are on the result
calculated leading to the same amount of output data sets. Con-
sequently, a relatively small and statistically less reliable output
scope results. To increase the statistical signicance of the re-
sults, the following procedure can now be used: Since a DoE 4. Case Study
is performed the system behavior can be described at discrete
points inside an n-dimensional space of input and output param- 4.1. System description and input variables
eters. The experimental design comes along with a positve fea-
ture: A homogeneous space of grid point occurs, which allows The environment of an automotive door assembly is chosen
to generate an interpolation between these grid point using sub- to be the demonstrator. In particular, the main components that
stitute functions. This method is known as Response Surface belong to this assembly are the side frame, the multi-layer seal-
Method (RSM). Therefore, the outcome is a descriptive substi- ing system and the door as well as some other attachment and
tute function that represents the overall system behavior. This trim parts, see Fig. 4.
function enables the user to extract the requested result distri-
butions depending on the respective input data sets. If the CAT multi-layer
simulations are repeated accordingly, the results at the KPCs sealing system
can be corrected online. It yields an overall result distribution
which can be described as a function of the parameters average,
variance, skewness and kurtosis. Therefore it has a potentially side frame
higher siginicance than the rst two approaches. door
Fig. 4. Schematic representation of the demonstrator
positions, [7]. Within corresponding FEM simulations these present case. The procedure is as follows: Based on the tol-
eects have to be included to improve the statements about the erance simulation outcome a covariance matrix can be derived
limiting positions. To reduce the number of time-consuming that includes all observed KPCs. From this symmetrical ma-
and expensive simulation runs, a statistical experimental de- trix the 20 biggest interactions (likely or contrary) between all
sign, based on rigid tolerance calculations, is set up. Among KPCs will be chosen. These interactions as well as those six
other things, this experimental design serves as an input for the KPCs that are supposed to reduce the six degrees of freedom
FEM simulations. (with reference to the six alignment points) are then entered
in the preparation of the D-Optimal experimental design. This
4.2. Reference simulation: tolerance simulation without con- resulting experimental design will be assigned to every of the
sideration for elastic eects three approaches / coupling concepts A, B and C. The ndings
will be shown in the following subsection 4.4
As said in subsection 4.1, the tolerance simulation takes into
consideration deviations and displacements as well as related
4.4. Quantitative comparison of the approaches
adjustment concepts and mounting sequences in order to deter-
mine statistic distributions for all KPCs, [7].
On the right hand side of Fig. 6, the comparison of the re-
In the present example, the manufacturing process for the
sults (estimated mean, standard deviation, resulting distribu-
considered automotive door assembly is modeled within a tol-
tion) between the three approaches can be seen using the ex-
erance simulation using the state-of-the-art way, [6]. In order to
ample of one selected KPC. The two subgures shown on the
build the overall assembly according to specication, it needs
left hand side show the probability distribution functions (CAT
to be ensured that the door is aligned accurately with respect
& FEM) based on their respective simulations. These are the
to the sideframe. For this purpose the door is positioned on a
variables to be coupled. The results for estimated mean (solid
workpiece carrier with regard to its specic alignment concept.
line) and standard deviation (dashed line) according to the in-
The workpiece carrier itself is positioned along the side frame
dividual approach can be seen on the right hand side of Fig. 6.
using adjustment points on the side frame afterwards. Due to
the fact that these mounting process steps come along with cor-
responding mounting tolerances, it is now the aim to minimize h (-)
them before connecting side frame and door via hinges. There-
a) CAT Approach A
fore a so-called BestFit-Move is realised. This move corrects
the relative position between side frame and door in the best
possible way using laser measurement technology when build-
m
ing the real assembly in the plant. Subsequently, the relevant
KPCs as well as their corresponding measurements are going to h (-)
be assigned. The manufacturing process is carried out 10 000 Approach B
times leading to an equal number of measured values for every
vectorial KPC. These values result in probability distribution coupling concept
functions from which the Lower / Upper Specication Limit m
(LSL / USL) can be derived. Inside this simulation all dened
geometric deviations and displacements will be varied within h (-)
their respective tolerance specications using a MCS, see [18] b) FEM Approach C
and [23].
Fig. 5 shows possible probability distribution functions on
10 chosen KPCs as a result of the rigid tolerance simulation m
that in turn represents the manufacturing process.
5. Summary of results and conclusion [8] Fischer BR. Mechanical Tolerance Stackup and Analysis. Marcel Dekker,
Inc.. New York. USA; 2009.
[9] Gundlach C. Entwicklung eines ganzheitlichen Vorgehensmodells zur
This paper deals with the question of interface issues
problemorientieren Anwendung der statistischen Versuchsplanung. Disser-
between a tolerance simulation (rigid system behavior) and tation. Universitaet Kassel. Germany; 2004.
a corresponding numeric simulation (elastic system behav- [10] Kapici S, Ehlert M, Stockinger A, Coska C, Sammer K. Virtual validation
ior). Within this publication the goal is twofold; one goal of limiting positions in the nal assembly of the new BMW 3 series. Au-
is to introduce existing coupling approaches including their tomotive Circle International: Montagesysteme 2013. Bad Nauheim. Ger-
many; 2013.
(dis-)advantages that enable the user to perform overall system
[11] Lindau B, Andersson A, Lindkvist L, Soederberg R. Using Forming Simu-
simulations. The second goal is to point out a complemen- lation In Virtual Assembly Analysis. Proceedings of ASME 2012 Inter-
tary approach. Furthermore, within the scope of a case study national Mechanical Engineering Congress & Exposition, IMECE2012;
(demonstrator: automotive door assembly), the procedure de- 2012.
scribed is applied. As initially expected, the results of this [12] Lindau B, Andersson A, Lindkvist L, Soederberg R. Body in White Ge-
ometry Measurements of Non-Rigid Components: a Virtual Perspective.
case study show dierent characteristics with reference to the
Proceedings of the ASME 2012 International Design Engineering Techni-
attributes examined, which are estimated mean, standard devi- cal Conferences & Computers and Information in Engineering Conference,
ation, skewness and kurtosis. IDETC/CIE. Vol. 5; 2012. p. 497-505.
Approach C has been identied to be the most promising ap- [13] Liu SC, Hu SJ. Variation Simulation for Deformable Sheet Metal Assem-
proach due to two abilities. On the one hand, this approach is blies Using Finite Element Methods. Journal of Manufacturing Science and
Engineering. Vol. 119; 1997. p. 368-374.
able to perform a direct numerical superposition of the source
[14] Lustig R. Integration und Verarbeitung von Verformungsinformationen
data. On the other hand, this one keeps the form and charac- im Umfeld rechnergestuetzter Toleranzanalysen. VDI Fortschritt-Berichte.
teristics of the distribution for each and every KPC across the Reihe 20. Dissertation. FAU Erlangen-Nuernberg. Germany; 2008.
simulation chain. At the same time, however, this is the most [15] Mannewitz F. Komplexe Toleranzanalysen einfach durchfuehren. Kon-
cost-intensive and time consuming approach. A comparison struktion; 2005.
[16] Markvoort L. Mthodologie danalyse statistique de tolrances dans les as-
with the less-expensive approaches A and B shows that there
semblages impliquant des composants dformables. Mmoire de doctorat en
are signicant mean shifts due to elastic eects arising from Automatique et Informatique des Systmes Industriels et Humains. Univer-
their respective FEM simulations. However, they do not reach sit de Valenciennes; 2007.
the level of approach C. With regard to the variances it can be [17] Prakash N, Phoomboplab T, Izquierdo LE, Ceglarek D. Self-Resilient Pro-
stated that the hypothesis of a growing variance (approach B) duction Systems: Going Beyond Design Robustness. 11th CIRP Interna-
tional Conference on Computer Aided Tolerancing. Annecy; 2009.
can be rejected due to the results achieved in approch C. This
[18] Rubinstein R. Simulation and the Monte Carlo method. Technion. Israel
is in line with the assumption that approach C stands for the Institute of Technology. New York: John Wiley & Sons, Inc.; 1981.
highest precision. When comparing the approaches A and C, [19] Stockinger A, Lustig R, Meerkamm H. Computer-based and experimental
it is noticeable that the standard deviations dier signicantly, validation of an approach to combine tolerance zones with elastic defor-
whereby the smaller standard deviation in approach C opens up mations. International Conference on Engineering Design. Paris. France.
28-31 August 2007; 2007.
a huge potential for the sealing design / conguration.
[20] Stockinger A, Meerkamm H. Virtual dimensional product validation by in-
One substantial task in this context is the experimental vali- tegration of simulations. International Conference on Engineering Design.
dation. A comparison with experimental data as well as corre- Stanford, CA, USA; 2009, p. 181-192.
sponding explanations will be addressed in a future publication; [21] Stockinger A. Computer Aided Robust Design-Verknuepfung rechnerun-
to nally select the most suitable approach. tersttzter Entwicklung und virtueller Fertigung als Baustein des Toleranz-
managements. Dissertation. FAU Erlangen-Nrnberg. Germany; 2010.
[22] Thornton A C, Donnelly S, Ertan B. More than Just Robust Design: Why
Product Development Organizations Still Contend with Variation and its
References Impact on Quality. Research in Engineering Design. Vol. 12; 2000. p. 127-
143.
[1] Camelio JA, Hu SJ, Ceglarek D. Modeling Variation Propagation of Multi- [23] VDI. Monte-Carlo-Simulation. VDI 4008. Dsseldorf: Verein Deutscher In-
Station Assembly Systems with Compliant Parts. Journal of Mechanical genieure; 1999.
Design. Vol. 125; 2003. p. 673-681. [24] Waermefjord K, Lindkvist L, Soederberg R. Tolerance Simulation of Com-
[2] Camelio JA, Hu SJ, Marin S. Compliant Assembly Variation Analysis Us- pliant Sheet Metal Assemblies Using Automatic Node-Based Contact De-
ing Component Geometric Covariance. Journal of Manufacturing Science tection. Proceedings of IMECE2008. ASME International Mechanical En-
and Engineering. Vol. 126; 2004. p. 355-360. gineering Congress and Exposition. Boston. USA; 2008.
[3] Ceglarek D, Huang W, Zhou S, Ding Y, Kumar R, Zhou Y. Time-Based [25] Xie K, Wells L, Camelio JA, Youn B. Variation Propagation on Compli-
Competition in Multistage Manufacturing: Stream-of-Variation Analy- ant Assemblies Considering Contact Interaction. Journal of Manufacturing
sis (SOVA) Methodology-Review. The International Journal of Flexible Science and Engineering, Vol. 129; 2007. p. 934-942.
Manufacturing Systems. Vol. 16; 2004. p. 11-44.
[4] Chen Y, Ding Y, Jionghua J, Ceglarek D. Integration of Process-Oriented
Tolerancing and Maintenance Planning in Design of Multistation Manu-
facturing Processes. IEEE Transactions on Automation Science and Engi-
neering. Vol. 3. No. 4; 2006. p. 440-453.
[5] Day D, Raines M, Swift K. Setting proper tolerances early in the de-
sign phase helps avoid headaches in production. MachineDesign. Vol. 3.
www.machinedesign.com; 2005.
[6] Ehlert M, Stockinger A, Wartzack S. Validation of customer requirements
by system simulation taking tolerances and system variations into account.
ISoRD - International Symposium on Robust Design. Copenhagen. Den-
mark; 20114. p. 61-68.
[7] Ehlert M, Heling B, Wartzack S. Virtual validation of functional automo-
tive door assembly properties by means of superposed CAT and FEM. In-
ternational Conference on Engineering Design ICED. Milan. Italy; 2015.