Professional Documents
Culture Documents
Simulation-Based Tolerance Stackup Analysis in PDF
Simulation-Based Tolerance Stackup Analysis in PDF
~'ACHINING
Y. Kevin Rong
Computer-aided Manufacturing Laboratory
Department of Mechanical Engineering
Worcester Polytec;hnic Institute, Worcester, MA
KEYWORDS INTRODUCTION
Monte Carlo Simulation, Tolerance Chain, Tolerance stackup can be defined as the
Tolerance Chart, Tolerance Allocation accumulation of errors when machining a
feature using different operational datums than
the ones specified in the blueprints. Analysis of
ABSTRACT tolerance stackup is critical to ensure accuracy
of the machined component. The two traditional
Tolerance stackup in machining results from methods used to analyze tolerance stackup in
using operational datums that are different from machining are worst-case analysis and
design datums. It is inevitable due to economic statistical analysis. These methods are based on
considerations of the machining process. assumptions that are too restrictive and have
Conventional methods used for tolerance several drawbacks:
stackup analysis include worst-case and
statistical analysis. These methods are based (1) Worst-case analysis assumes that all
on strong assumptions and have certain tolerances simultaneously occur at their
drawbacks, the most critical one being the worst limit; and thus, is exaggeratedly
inability to analyze geometric tolerances. This pessimistic in calculating tolerance
paper presents a novel method based on feature stackup.
discretization, manufacturing error analysis, (2) Statistical analysis assumes individual
Monte Carlo simulation, and virtual inspection. It tolerances to be independent and have
is generally applicable to stackup analysis of a normal distribution, which allows the
various types of tolerances and produces more use of root sum squares for stackup
accurate and less conservative results. The calculation. This will lead to
trade off is longer computational time. conservative results since individual
tolerances are more or less correlated in
machining.
(3) The analysis is restricted to dimensional
tolerances. In other words, tolerance
stack between features is preformed in
I,
If
f
The concluding link for the dimension chain This problem can be resolved using Monte Carlo
shown in Figure 1 can be found as: simulation. In fact, Monte Carlo simulation is a
popular tool for assembly tolerance stackup
c=:(i1+i2)-(d1+d2)
analysis [Chase and Parkinson (1991)]. Note
that Monte Carlo simulation is a general tool that
Apparently, the tolerance of concluding link is
determined by the tolerances of the increasing can be applied to many engineering or non-
and decreasing links. There are different engineering problems. The key to its successful
methods to determine this resultant tolerance. In application is an accurate model to describe the
worst-case method, the concluding link's underlying problem. For example, the vector-
loop-based model [Gao et al. (1998)] is used to
tolerance L\c is determined as follows:
describe assembly in 2 and 3-D, where
I oc ..oc
L1c = LI-:-I L1ij + LI-I L1dk tolerance stackup analysis can be achieved
j=1 oz j k=1 od k through Monte Carlo simulation or direct
linearization. The only model that is currently
available for machining tolerance stackup
.~ analysis is the dimension chain model, which
cannot deal with geometric tolerances and does
not take manufacturing errors into account.
t:... d2 ',~
; C i.. . SIMULATION-BASED TOLERANCE
STACKUP ANALYSIS
Virtu~1 YES
Inspection
g ~
type of applications, manufacturing errors are that are numbered from 1 to 6. Milling is the
classified based on different factors. These process used to machine the surfaces. Hole
factors are: (1) ~, which accounts for error drilling is not included in the simulation since it
variation with time; (2) randomness, errors are does not have an effect on the tolerance chain
categorized as deterministic and random; (3) of concern. Figure 7 shows a tolerance chart of
sources of errors, including geometric errors that the part in order to predict tolerance stackup
represent the inaccuracy of surfaces moving using worst-case and statistical methods. We
relative to each others, thermal errors that are interested in the dimension shown in line 8
account for thermal deformation in the tool as a concluding link. The contributing links are
because of heat provided by cutting process, shown in lines 7, 4 and 1.
machine, people. thermal memory (from
previous environments) and cooling effect from
coolant, and cutting-force induced errors that
come from the dynamic stiffness of all
components of the machine tool; (4) ~
influence on aeometric accurac~, which includes
locating error that accounts for the variation
between the ideal datum and the actual one
after locating and clamping, and machining error
that accounts for the variation between the ideal
position of the machine tool and the actual one
[Un and Zhang (2001). Un et al. (1997)].
FIGURE 4. ABS PART (ANTI-BLOCK SYSTEM
Since we are dealing with tolerance stackup HOUSING).
analysis that focuses on the geometric accuracy c 4~,1~,. 100 .,
of features, it is logical to adopt the last
classification method. All manufacturing errors
are lumped into two categories: (1) setup error
due to locating and clamping an imperfect
workpiece using an imperfect fixture, and (2)
cutting tool deviation that accounts for all other
error sources. Setup error can be determined by
first measuring the accuracy of the underlying .Ih-
fixture (3 rotation and 3 translation components) "I I II III
"I r II I"
and then synthesizing it with workpiece accuracy
(measured using flatness). Cutting tool deviation
can be determined by machining a surface and
then measure surface point variations in z-
-ISJ
coordinate. More details can be found in [Musa
(2003)]. FIGURE 5. SIMPLIFIED PART DRAWINGS.
=
~
moments when having less sample size. Note A comparison of the results of worst-case,
that 10,000 iterations are considered large statistical, and simulation methods in finding
enough by most Monte Carlo simulation concluding link tolerance stackup is shown in
researchers [Cvetko (1998)]. The first two
moments (mean and variance) at 100,000
iterations are shown in Table 1 for dimensions in
lines 1, 4, 7 and 8.
Table 2. Monte Carlo simulation results were
found to be less than both the traditional
methods. The ratio of simulation tolerance
stackup to the worst case tolerance stackup was
",' I "
,
found to be 0.34 and the ratio of simulation
';
:
tolerance stackup to the statistical method ;
tolerance stackup was found to be 0.59.
I; ,
TABLE 2. TOLERANCE EVALUATION USING THE ,;;,'
~I THREE APPROACHES. ;',1
i i
"
.l , Standard Deviation Tolerance=6a
-~
DO [DIJ
! [JIIJ
, ~
~ "-"-"---b\; ,t;c~
l-'~-,"
: Ii
~ L1
L4 0.010698664 0.064191982
! ! Ii L7 0.010872206 0.065233235
FIGURE ,,
6. SETUP PLAN. .. La 0.010890703 0.065344216
Worst
.., Case
Statistical
0.193434944
0.111683618
).. I Simulation
Simulation
'4r:-o
/
0.065344216
, i ITTI _III
-
II1~161
I 1.S I
-~+1
10
(iterations.)
I 1110 "'Iii II 4+~
ii:1:11111
1
8
mlli 1:1
'---""'- 1J
'+'
~ Jj---~
Mean Variance
~
84.9983246659704 0.000113812363943454
59.9943255754447 0.000114461403303197
49.9926265647646 0.000118204861229129
0.000118607406206314 FIGURE 8. PROGRESS OF RESULTS WITH
SAMPLE SIZE INCREASE.
I
~
~'"
-~~
~""
~
~
0.010668288
I0.064009727
~
Suppose that we need to maintain 0.066 mm TABLE 3. PART PER MILLION (PPM) REJECT
for the dimension shown in line 8 01' the COMPARISON WHEN ALLOCATING PRODUCTION
tolerance chart (Figure 7). Then, we will need to TOLERANCE USING WORST CASE, STATISTICAL
allocate proper tolerances for the concluding AND SIMULATION METHODS.
links (dimensions shown in lines 1, 4 and 7). Worst .
According to our simulation, assigning 0.060 mm Case Statistical Simulation
for each of the three contributing links will be Tolerance 2.02a 3.5a 6a
good enough. However, if we make the C-p-- 0.37 0.64 1
allocation decision based on the worst case and PPM rejects 312,500 80,100 2,700
statistical methods, 0.022 mm and 0.038 mm will
be needed for each contributing link,
respectively. Apparently, traditional methods CONCLUSIONS AND FUTURE WORK
overestimate the severity of tolerance stackup RECOMMENDATIONS
and hence allocate tighter than necessary
production tolerances. A method for evaluating tolerance stackup
using Monte Carlo simulation driven by feature
Process capability indicates how capable discretization, manufacturing error analysis, and
manufacturing process is to maintain design virtual inspection is proposed in this work.
requirements in terms of number of rejects Cutting tool deviation and setup error were
expected out of a million. Usually. it is simulated along with machining and inspecting
represented by the Cpindex, which is the ratio of processes for virtual parts. This method is
design specifications (tolerance T) to the generally applicable to geometric as well as
dimensional tolerances. It also gives less
process variability (6a) , i.e., C = ~. If we
p 60- conservative results compared to the traditional
choose processes that are capable to achieve ones (worst case and statistical methods).
f our simulation requirements, then we will be Overestimating tolerance stackup could result in
> satisfied with 2,700 parts per million (PPM) precluding good process plans that should be
)
rejects when having the process capability index accepted. Therefore, accurate evaluation of
1 equals to 1. However, if we use the same tolerance stackup can lead to cost-effective
process considering the traditional methods, process plans.
much more rejects per million will be
erroneously expected (as shown in Table 3 and Validation is used to ensure that the simulation
5 model matches accurately enough with the real
Figure 9) since the allocated production
) world behavior. Our future work will focus on
tolerances are tighter than necessary. In other
~ conducting experiments to validate the
words, simulation results suggest that the
s simulation results. In our illustrative example, it
chosen processes are capable to do the job,
whereas the traditional methods do not. This was assumed that the probability distribution
shows the importance of having a less functions (pdt) of cutting tool deviation and setup
conservative method for tolerance allocation. error are available and follow normal distribution.
In reality, this may not be true, especially when
dealing with new products. We have developed
r ~t,,~.J.l~)~PI'-i
~TCI.sE experimental procedures to estimate error pdt's
and preliminary results showed that these errors
RojOCW.J,4)OO-
r STA11511:AI. do not follow normal distribution. Therefore,
,/ ~ Ro~.hI3~WM
(SlMJLAna4 general probability density estimation will be an
important future research topic.
~ 2.02a
, ~ Finally, we would like to mention that Monte
1:-. -3.Sa Carlo simulation is computationally expensive,
while analytical methods are more efficient.
~---~ J- =6a Therefore, when dealing with dimensional
tolerances, it should be beneficial to develop
statistical methods that consider error
FIGURE 9. REJECTION AREAS COMPARISON
correlations and different error pdt's. This will
H WHEN ALLOCATING PRODUCTION TOLERANCES
USING WORST CASE, STATISTICAL AND solve the overestimation problem in tolerance
SIMULATION METHODS. stackup analysis while preserving computational
--
~
efficiency. When dealing with geometric Huang, S. H., and H.-C. Zhang, 1996, "Use of
tolerances, analytical methods may not be Tolerance Chart for NC Machining," Journal of
feasible because of the need for feature Engineering Design and Automation, Vol. 2, No.
discretization. To improve computational 1,pp.91-104.
efficiency, modified Monte Carlo simulation
maybe the only way. Kelton, W. D., R. P. Sadowski, and D. A.
Sadowski,
McGraw-Hili, 2002,
Second Simulation
Edition. with Arena'
ACKNOWLEDGEMENT
Un, E. and H.-C. Zhang, 2001, "Theoretical
This material is based upon work supported by Tolerance Stackup Analysis Based on Tolerance
the National Science Foundation under Grant Zone Analysis," International Journal of
No. 0099735. Any opinions, findings, and Advanced Manufacturing Technology, Vol. 17,
conclusions or recommendations expressed in pp.257-262.
this material are those of the authors and do not
necessarily reflect the views of the National Un, S., H. P. Wang, and C. Zhang, 1997,
Science Foundation. "Statistical Tolerance Analysis Based on Beta
Distribution," Journal of Manufacturing Systems,
Vol. 16, No.2, pp. 150-158.
REFERENCES
Musa, R., 2003, Simulation-based Tolerance
ANSI Y14.5-1994, 1995, Dimensioning and Stackup Analysis in Machining, M.S. Thesis,
Tolerancing, New York, NY: American Society of Department of Mechanical, Industrial, and
Mechanical Engineers. Nuclear Engineering, University of Cincinnati.
Chase, K. W., and A. R. Parkinson, 1991, "A Ngoi, B. K. A., L. E. N. Urn, A. S. Ong, and B. H.
Survey of Research in the Application of Urn, 1999, "Applying the Coordinate Tolerance
Tolerance Analysis to the Design of Mechanical System to Tolerance Stack Analysis Involving
Assemblies," Research in Engineering Design, Position Tolerance," International Journal of
Vol. 3, pp. 23-37. Advanced Manufacturing Technology, Vol. 15,
No.6, pp. 404-408.
Cvetko, R., K. W. Chase, and S. P. Magleby,
1998, "New Metrics for Evaluating Monte Carlo Shan, A., R. N. Roth, and R. J. Wilson, "A New
Tolerance Analysis of Assemblies," Proceedings Approach to Statistical Geometrical Tolerance
of the ASME International Mechanical Analysis," International Journal of Advanced
Engineering Conference and Exposition, Manufacturing Technology, Vol. 15, pp. 222-
Anaheim, CA, Nov. 15-20. 230.
Gao, J., K. W. Chase, and S. P. Magleby, 1998, Whybrew, K. and G. A. Britton, 1997, "Tolerance
"Generalized 3-D Tolerance Analysis of Analysis in Manufacturing and Tolerance
Mechanical Assemblies with Small Kinematic Charting," Advanced Tolerancing Techniques
Adjustments," liE Transactions, Vol. 30, pp. 367- (edited by H.-C. Zhang), Chapter 2, John Wiley
377. & Sons.
Gladman, C. A., 1980, "Applying Probability in Woo, T. C., R. Liang, ' C. Hsieh, and N. K.
Tolerance Technology," Transactions of the Lee, 1995, "Efficient Sampling for Surfaceal
Institution of Engineers Australia Mechanical Measurement," Journ of Manufacturingpp.
Engineering, Vol. 5, No.2, pp. 82-85. Systems, Vo. 14, No.5, 345-354.
Huang, Q.; S. Zhou, and J. Shi, 2002, Yao, Y., J. Li, W. B. Lee, C. F. Cheung, and Z.
"Diagnosis of Multi-Operational Machining Yuan, 2002, "VMMC: a Test-Bed for Machining,"
Processes by Using Virtual Machining," Robotics Computers in Industry, Vol. 47, pp. 255-268.
and Computer-Integrated Manufacturing, Vol.
18, No. 3-4, pp. 233-239.
C.