三大汽车公司冲压工艺分析报告（英文）

fengwei-08

Automotive Sheet Steel
Stamping Process Variation
An analysis of stamping
process capability and
implications for design,
die tryout and process
control.
Auto/Steel Partnership
Automotive Sheet Steel
Stamping Process Variation:
An Analysis of Stamping Process Capability and
Implications for Design, Die Tryout and Process Control
Auto/Steel Partnership Program
Body Systems Analysis Project Team
2000 Town Center - Suite 320
Southfield, MI 48075-1123
2000
Auto/Steel Partnership
AK Steel Corporation
Bethlehem Steel Corporation
DaimlerChrysler Corporation
Dofasco Inc.
Ford Motor Company
General Motors Corporation
Ispat Inland Inc.
LTV Steel Company
National Steel Corporation
Rouge Steel Company
Stelco Inc.
U. S. Steel Group, a Unit of USX Corporation
WCI Steel, Inc.
Weirton Steel Corporation
This publication is for general information only. The material contained herein should not be used
without first securing competent advice with respect to its suitability for any given application. This
publication is not intended as a representation or warranty on the part of The Auto/Steel Partnership – or
any other person named herein – that the information is suitable for any general or particular use,
or free from infringement of any patent or patents. Anyone making use of the information assumes
all liability arising from such use.
This publication is intended for use by Auto/Steel Partnership members only. For more information or
additional copies of this publication, please contact the Auto/Steel Partnership, 2000 Town Center, Suite
320, Southfield, MI 48075-1123 or phone: 248-945-7777, fax: 248-356-8511, web site: www.a-sp.org
Copyright 2000 Auto/Steel Partnership. All Rights Reserved.
ii
Table of Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
Executive Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.0 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1 Motivation for Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Study Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.0 Stamping Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Components of Variation Explained . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Calculating Components of Variation Using ANOVA . . . . . . . . . . . . . . . . . . . . . 9
2.3 Description of the Sources of Stamping Variation . . . . . . . . . . . . . . . . . . . . . . . 13
3.0 Analysis of Stamping Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1 Mean Conformance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.1 Benchmark Comparison - Body Side Outer and Inner Panels . . . . . 14
3.1.2 Mean Bias and Part Tolerances . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.3 Benchmark Comparison - Tryout versus Production . . . . . . . . . . . . . 18
3.1.4 Mean Bias Stability over Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.5 Impact of Shipping on Mean Bias . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Stamping Process Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.1 Benchmark Comparison - Part-to-Part Variation . . . . . . . . . . . . . . . . 21
3.2.2 Variation Over Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.3 Impact of Shipping on Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.4 Components of Variation: Part-to-Part, Run-to-Run,
and Begin-End of Run . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.5 Steel Properties and Press Setup Control and Stamping Variation . . 27
3.2.6 Effect of Mean Shifts on Statistical Process Control Techniques . . . . 29
4.0 Tolerance Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1 Tolerances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2 Cp and Cpk (Pp and Ppk) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Recommended Tolerances for Sheet Metal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.4 Part Tolerances and Functional Build . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.0 Conclusions and Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Appendix A - Part Sketches by Company . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
iii
iv
List of Figures
Figure 1. Body Side Components Chosen for Company C . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Figure 2. Components of Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Figure 3. Potential Sources of Stamping Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Figure 4. Total Variation Partitioned into Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Figure 5. Body Side Outer for Company A: 12 Measurement Locations . . . . . . . . . . . . . . . . . 12
Figure 6. Histogram of Mean Values across 5 Parts for Company C . . . . . . . . . . . . . . . . . . . . 15
Figure 7. Mean Conformance: Rigid vs. Non-Rigid Panels . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Figure 8. Mean Conformance: Two-Piece Body Side Panel vs. One-Piece . . . . . . . . . . . . . . . 16
Figure 9. Correlation of Mean at Part Approval vs. Production . . . . . . . . . . . . . . . . . . . . . . . . . 19
Figure 10. Effect of Stamping Mean Shift on Body Side Assembly . . . . . . . . . . . . . . . . . . . . . . 20
Figure 11. Average Variation (Standard Deviation) by Type of Part . . . . . . . . . . . . . . . . . . . . . . 23
Figure 12. Part-to-Part Variation: Home Line Tryout Approval vs. Production, by Dimension . . . 24
Figure 13. Components of Variation for Body Side Panel at Company C and Company D . . . . . 26
Figure 14. Relationship between Press Tonnage and Mean Shift Variation ( mean shift) . . . . . 29
Figure 15. X-Bar/Range Chart vs. Individuals/ Moving Range Charts . . . . . . . . . . . . . . . . . . . . 32
Figure 16. Illustration of Cp and Cpk calculations for three scenarios . . . . . . . . . . . . . . . . . . . . 35
Figure 17. Part Sketches at Company A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Figure 18. Part Sketches at Company B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Figure 19. Part Sketches at Company C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Figure 20. Part Sketches at Company D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Figure 21. Part Sketches at Company E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Figure 22. Part Sketches at Company F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Figure 23. Part Sketches at Company G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
v
List of Tables
Table 1. Participating Automotive Manufacturers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Table 2. Components Studied at Each Automotive Manufacturer . . . . . . . . . . . . . . . . . . . . . . 5
Table 3. Formulae for Calculating Components of Variation . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Table 4. 36-Data Samples for a Stamping Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Table 5. SPSS Output Calculations for Mean Squared Errors (all factors) . . . . . . . . . . . . . . . . 11
Table 6. SPSS Output Calculations for Mean Squared Errors without Begin-End Factor . . . . . 11
Table 7. Summary of Components of Variation Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Table 8. Variance Summary for twelve Body Side Dimensions . . . . . . . . . . . . . . . . . . . . . . . . 12
Table 9. Mean Conformance by Company . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Table 10. Mean Bias by Type of Part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Table 11. Mean Conformance and Tolerances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Table 12. Summary of Mean bias: Tryout vs. Production - Case Study Parts . . . . . . . . . . . . . . 18
Table 13. Comparisons of the Change in Mean Bias from Tryout to Home Line . . . . . . . . . . . . 19
Table 14. Change in Mean from Home Line to Long-term Production . . . . . . . . . . . . . . . . . . . 20
Table 15. Summary of Panels Measured Before and After Shipping . . . . . . . . . . . . . . . . . . . . . 21
Table 16. Part-to-Part Variation for the Body Side Outer Panels . . . . . . . . . . . . . . . . . . . . . . . . 22
Table 17. Effect of Dimension Location on Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Table 18. Part-to-Part Variation: Home Line Approval vs. Production, by Company . . . . . . . . . 24
Table 19. Summary of Remeasured Data Before and After Shipping via truck . . . . . . . . . . . . . 25
Table 20. Summary of Part-to-Part and Total Variation for the Body Side Outers . . . . . . . . . . . 25
Table 21. Sources of Variation by Part for Company A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Table 22. Sources of Variation by Part for Company C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Table 23. Summary of Product and Process Variation Compliance . . . . . . . . . . . . . . . . . . . . . 28
Table 24. Summary of Mean Shift Variation across Companies . . . . . . . . . . . . . . . . . . . . . . . . 30
Table 25. Process Control Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Table 26. Effect of Stamping Mean Shifts on Assembly Variation . . . . . . . . . . . . . . . . . . . . . . . 33
Table 27. General Recommended Tolerances for
Stamped Parts Based upon Process Capability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Preface
This report is one of a series published by the
Auto/Steel Partnership Body Systems Analysis
Project Team on stamping and assembly variation,
body measurement systems and process validation.
These reports provide a summary of the project
research and are not intended to be all inclusive
of the research effort. Numerous seminars
and workshops have been given to individual
automotive manufacturers throughout the project
to aid in implementation and provide direct technical
support. Proprietary observations and implementation
details are omitted from the reports.
This automotive body development report,
"Stamping Process Variation: An Analysis of
Stamping Process Capability and Implications for
Design, Die Tryout and Process Control," updates
ongoing research activities by the Body Systems
Analysis Team and the Manufacturing Systems
staff at The University of Michigan's Office for the
Study of Automotive Transportation.
An over-riding goal of this research is to develop
new paradigms that will drive automotive body-inwhite
development and production towards a total
optimized processing system. Previous reports
described fundamental research investigating
simultaneous development systems for designing,
tooling and assembling bodies, and flexible body
assembly. Since the inception of this research program,
considerable emphasis has been focused
on benchmarking key world class body development
and production processes. These benchmarks
created foundation elements upon which
further advances could be researched and developed.
This report summarizes recommendations for
moving toward a new "functional build" paradigm
by tightly integrating the many individual activities
ranging from body design and engineering
through process and tooling engineering. Revised
stamping die tryout and buyoff processes receive
special emphasis, as does the launch of stamping
and assembly tools.
The researchers are indebted to several global
automotive manufacturers for their on-going dedication
and participation in this research. They
include DaimlerChrysler Corporation, Ford Motor
Company, General Motors Corporation, Nissan,
NUMMI (Toyota), Opel and Renault. Each conducted
experiments under production conditions
involving hundreds of hours of effort and often
requiring the commitment of many production
workers and engineering personnel. Although it
may be impractical to mention each one of these
people individually, we do offer our sincere appreciation.
These reports represent a culmination of several
years of effort by the Body Systems Analysis
Project Team. Team membership, which has
evolved over the course of this project, includes:
J. Aube, General Motors Corporation
H. Bell, General Motors Corporation
C. Butche, General Motors Corporation
G. Crisp, DaimlerChrysler Corporation
T. Diewald, Auto/Steel Partnership
K. Goff, Jr., Ford Motor Company
T. Gonzales, National Steel Corporation
R. Haan, General Motors Corporation
S. Johnson, DaimlerChrysler Corporation
F. Keith, Ford Motor Company
T. Mancewicz, General Motors Corporation
J. Naysmith, Ronart Industries
J. Noel, Auto/Steel Partnership
P. Peterson, USX
R. Pierson, General Motors Corporation
R. Rekolt, DaimlerChrysler Corporation
M. Rumel, Auto/Steel Partnership
M. Schmidt, Atlas Tool and Die
The University of Michigan Transportation
Research Institute conducted much of the
research and wrote the final reports. The principal
research team from the Manufacturing Systems
Group was:
Patrick Hammett, Ph.D. (734-936-1121/phammett@
umich.edu)
Jay Baron, Ph.D. (734-764-
4704/jaybaron@umich.edu)
Donald Smith, Associate Director (734-764-5262)
vi
Executive Summary
The Auto/Steel Partnership (A/SP) is an innovative
international association that includes
DaimlerChrysler, Ford, General Motors and eleven
North American sheet steel producers. The
Partnership was formed in 1987 to leverage the
resources of the automotive and steel industries to
pursue research projects leading to excellence in
the application of sheet steels in the design and
manufacture of vehicles. The Partnership has
established project teams that examine issues
related to steel properties including strength, dent
resistance, surface texture and coating weights,
as well as manufacturing methods including
stamping, welding and design improvements.
Automotive manufacturers face the challenge of
identifying when a process is capable of producing
dimensionally acceptable stamped panels.
The non-rigid nature of many stamped parts has
always made them difficult to measure. Often
parts do not meet the dimensional quality objectives,
as measured by Cpk, seen in many other
vehicle components. In fact, no manufacturer has
successfully achieved a Cpk of 1.33 on all part
dimensions using the original specifications. This
is particularly true for the larger, lighter gauge
body panels. Furthermore, achieving a high Cpk
value alone is not necessarily a good predictor of
final dimensional quality. Factors, such as the
rigidity of the mating panels, the assembly locating
process and the clamp and welding effects,
influence how body panels build into an assembly.
Consequently, a number of automotive manufacturers
have opted not to use Cpk as the principal
measure of panel quality.
This stamping report analyzes dimensional data to
characterize stamping variation by short-term
(part-to-part), long-term (die set to die set), and
mean bias (long-term deviation from design nominal)
to better understand process capability.
Numerous factors affect the observed variation in
a stamping process, making stamping one of the
more difficult processes to control. The complexity
of stamping makes it extremely difficult to conduct
rigorous experimental studies that can be generalized
beyond a given part and process configuration.
Thus, the knowledge base of stamping variation
is very sparse, and a great opportunity exists
to learn and to apply this knowledge to automotive
body evaluation processes including die buy-off,
production validation and long-term process
capability analysis. Working within the constraints
of the production environment, this research evaluated
stamping variation for several processes
across the seven manufacturers. The research
found that stamping variation is related to:
• Check point location on a part: More rigid areas
tend to be closer to nominal and have less
variation.
• Measurement fixture design: Checking fixtures
with more clamps tend to reflect lower variation.
• Part size, complexity and thickness: Smaller, less
complex and thicker parts have less variation.
• Press process control: Different press lines
demonstrate higher die set to die set mean shift
control which often is reflected in the control of
process variables such as draw press tonnage.
• Shipping and handling: The shipping and handling
of parts tends to increase variation and
shift dimensions on the parts.
• Changes in stamping presses: Some dimensional
shifts occur as dies are moved from a tryout
press line to the home production press line.
Different automotive manufacturers manage variation,
in part, by how they manage these factors and
several examples are cited in this report. Although
the effects of steel material properties such as
gauge, yield strength, percent elongation and nvalue
were investigated, this factor is not included
in the list above because it had minimal influence
on variation. All of the manufacturers that supplied
steel coupons in this study had material properties
sufficiently controlled to virtually eliminate any influence
on stamping variation.
One of the objectives of this research is to understand
the amount of variation experienced by different
manufacturers and how they manage variation
issues. Together, this information may be used
to improve the overall validation process for
stamping and sheet metal assembly. The uncertainty
of sheet metal assembly clearly supports a
functional build approach where component quality
is determined by how it influences the assembly.
These methods are outlined in other reports by
the Body Systems Analysis Project Team.
1
1.0 Introduction
1.1 Motivation for Research
Leading global automotive manufacturers have
been challenged with applying traditional design
practices to sheet metal design and assembly.
The goal behind this effort is to help achieve high
quality car bodies with minimal lead-time and
development costs. These practices include geometric
dimensioning and tolerancing (GD&T), variation
simulation analysis, tolerance stack-up
analysis and setting quality standard targets for
process capability, such as Cp and Cpk. Most manufacturers
have expressed concerns over the limited
success these methods have had on sheet
metal processes including the assignment of
dimensional part tolerances, translating component
designs into tools that can make them and
predicting assembly conformance based on
stamping capability. Several important observations
account for the limited success of applying
traditional design principles to these processes:
• Manufacturers experience difficulties estimating
mean part dimensions, relative to nominal and
process variation because these attributes are
product and process co-dependent. Potential
attributes affecting variation include material
properties (steel variations in gauge, grade, and
coatings), part geometry (size and shape), die
engineering and construction, and stamping
press variables. The infinite number of design
and process possibilities make it nearly impossible
to accumulate sufficient historical knowledge
for a designer to accurately assign tolerances
that consistently meet future process capability.
• The lack of component rigidity allows less stable
panels to conform to more rigid ones, making it
difficult to predict final assembly dimensions
based on component quality.
• Component dimensions that deviate from their
design nominal cannot always be predictably
centered or shifted to the desired nominal without
excessive rework costs. Moreover, this
rework may correct one particular deviation but
adversely affect correlated points on the same
part.
• Part measurement systems often have limited
capability to measure non-rigid parts without
additional clamps beyond 3-2-1 requirements.
These additional clamps in the fixtures over-constrain
parts, thereby shifting mean dimensions.
• Stamping processes have so many input variables
affecting variation, with some estimates at
well over 100, that even world-class stamping
operations routinely operate outside of statistical
control, with non-stable process means between
die sets, especially on larger flimsy parts.
Consequently, measuring several parts from a
single die set or run does not provide sufficient
information about the expected long-term variation
of the process.
• Assembly processes often distort parts - sometimes
closer to and sometimes further away from
nominal - during assembly because of clamping,
spot welding, and inconsistencies of part locating
schemes. These distortions can shift panel
mean dimensions and affect process variation,
resulting in a low correlation between stamping
dimensions and assembly dimensions.
A purpose of this report is to provide a basic
understanding of stamping variation. The data in
this report are intended to illustrate general characteristics
of stamping variation, and are not
intended to be a comprehensive data base to support
design. World-class automotive manufacturers
that are most adept at designing, producing
and assembling sheet metal are those who have
effectively learned from past designs, while managing
the variation in new parts and processes as
they become known. By researching a number of
stamping and assembly processes across several
manufacturers, this report begins to establish
boundaries for the limits of variation that can be
expected under different situations. This report
examines the implications of this inherent stamping
variation on several design and validation
activities including:
• Tolerance assignment,
• Check point selection,
• Stamping process control limits,
• Process validation - die tryout,
• Production part approval process - stamping,
• Part measurement systems and measurement
strategies, and
• Assembly strategies with respect to part locating
and clamping.
2
The majority of the data in this report was collected
under production conditions, resulting in several
advantages and disadvantages over a more
controlled experiment approach. The advantages
are that the data actually reflect what can be
expected in production at normal line rates
and with typical levels of process control.
Generalizations about process variation are made
where similar observations are seen over several
case studies. The disadvantages are that the data
cannot be used, in most instances, to support
direct cause-and-effect conclusions, thus often
limiting observations to hypotheses. However,
given the infinite number of part and process
design options, controlled experiments in stamping
and assembly often have a limited value in
generalizing results.
1.2 Study Background
The seven automotive manufacturers noted participated
in this study by providing data about their
stamping and assembly processes. These manufacturers,
and the respective vehicles studied are
shown in Table 1 below. Note that companies
are referred to as A, B, C, D, E, F, and G in this
report and they do not correspond to the order
presented.
3
Table 1. Participating Automotive Manufacturers
Data Stamping Assembly
Company Model Collection Location Location
GM Grand Am 1996 Lansing, MI Lansing, MI
NUMMI (Toyota) Corolla 1996 Fremont, CA Fremont, CA
DaimlerChrysler Neon 1997 Twinsburg, OH; Belvidere, IL Belvidere, IL
Nissan Altima 1997 Smyrna, TN Smyrna, TN
Ford Taurus 1997 Chicago, IL Chicago, IL
Opel Vectra 1998 Ruesselsheim, Germany Ruesselsheim, Germany
Renault Clio II 1998 Flins, France Flins, France
Dimensional studies at each manufacturer are
based on the body side assembly and its major
stamped components. One participant provided
data for panels on both the right and left body
side, resulting in a total of eight body-side assembly
studies. The difference among manufacturers
was the type of body side outer. Three used body
side outers with an integrated quarter panel while
the remaining manufacturers used two-piece body
sides. The other panels chosen in each body side
assembly study, typically 4-5 mating parts,
depended on the design, but with the goal to
include critical rigid structural reinforcements
with thicker gauges greater than 1.25 mm.
The scope of body panels included from all the
automotive manufacturers are:
• One-piece body side outer,
• Two-piece body side outer,
• Center pillar reinforcement (B-pillar),
• Front pillar reinforcement (A-pillar),
• Quarter outer panel,
• Quarter inner panel,
• Roof rail outer,
• Wheelhouse outer, and
• Windshield frame reinforcement.
Figure 1 below illustrates a typical body side case
study for a two-piece body side.
4
Figure 1. Body Side Components Chosen for Company C
Windshield Frame
Reinforcement
Front Pillar
Body Side
Center Pillar
Roof Rail

fengwei-08

5
Table 2. Components Studied at Each Automotive Manufacturer
Company Part Number of Steel Steel
Identifier Description Dimensions Gauge Coupons
Body Side – One Piece 39 0.69
Quarter Inner 76 0.90
A Wheelhouse Outer 38 0.61 Yes
Front Pillar Reinf 69 1.70
Center Pillar Reinf 60 1.44
Body Side – One Piece 104 0.90
Body Side Inner 54 0.80
Wheelhouse Outer 42 0.75
B Center Pillar Reinf 17 1.00 Yes
Cowl Side 24 1.10
Roof Rail Inner 8 0.80
Roof Rail Outer 13 1.00
Body Side – Two Piece 60 1.10
Roof Rail Outer 22 0.90
C Front Pillar Upper 9 1.85 Yes
Front Pillar Lower 8 1.85
Center Pillar Reinf 14 1.87
Windshield Side Inner 30 2.70
Body Side – Two Piece 17 0.73
Quarter Otr 14 0.82
D Front Pillar Lower 6 Yes
Center Pillar Lower 6
Front Pillar Upper 15
Center Pillar Upper 4
Body Side – Two Piece 35
E Front Pillar Lower 2 No
Center Pillar Lower 2
Roof Rail 2
Body Side – One Piece 38 0.90
F Quarter Outer 11 No
Body Side Inner 6
Center Pillar 6
Body Side – Two Piece 54 Frt. 1.17;
(tailor welded blank) Rr: 0.77
Body Side Inner 13 0.67
G Windshield Side Inner 5 1.17 Yes
Front Pillar Reinf 6 0.97
Center Pillar Reinf 10 1.17
Table 2 below lists the body side components chosen
from each of the automotive manufacturers for
this study. Each of the seven manufacturers is
identified consistently by the same letter, A
through G, throughout the report. (See Appendix
for sketches of components in study.)
A consistent sampling plan was applied to each of
the stamped panels. This plan was designed to
ascertain both short-term and longer-term variation
under production conditions. Six panels were
taken during each die set or production run for a
given part: three consecutive panels near the
beginning of the run and three consecutive panels
near the end of the run. This six panel sampling
plan was then repeated over six separate die sets,
thus producing a total case study of 36 panels (6
per die set x 6 die sets = 36 panels). This sampling
plan was executed on all of the major panels
in each case study. A few smaller reinforcements
had less than 6 die sets. The length of each die set
varied by participant, but tended to be greater
than four hours. The time between each die setup
also varied and was typically between 2 and 7
days. This sampling plan allowed the calculations
of short-term variation, or variation across consecutive
panels, and long-term variation, or variation
both within and between die sets.
Several of the manufacturers also collected panel
coupons and process variable data to see if relationships
could be found between the material or
equipment setup and stamped panel variation.
They collected a steel blank at the destacking side
of the press while a contiguous sample of 3 panels
was collected. At the same time, several companies
collected data on the process, such as tonnage
and cushion pressure in the draw die. The
actual variables collected at each manufacturer
varied by part and die design. The steel coupons
were collected and tested for several properties,
including R-value, n-value and blank gauge variation.
The measurement of the parts was conducted in a
manner to reduce potential error. The 36 samples
were collected over a period of several weeks and
set aside for measurement at one time. This
approach was intended to reduce measurement
errors by using a single operator and a standard
measurement protocol of a loading, clamping and
measurement sequence. To measure the body
side outer and many of the other inner panels, five
manufacturers used CMMs, while E and F used
hard fixtures. A few of the smaller parts were
measured on hard checking fixtures using
datamyte collection devices and measurement
probes. In all cases, part locating was based on
the standard checking fixtures used by each manufacturer
for internal quality monitoring. A few
chose to modify their CMM measurement routines
to include additional dimensions to provide a more
comprehensive geometric database.
One challenge with comparing manufacturers in
this study was the significant differences in measurement
systems. These differences relate primarily
to the locating and clamping of parts in the fixtures.
Some manufacturers attempt to minimize
the influence of the fixture on the part by minimizing
the number of clamps and clamping pressure,
while others intentionally over-constrain their parts
for measurement. Manufacturers attempting to
reduce the influence of the measurement system
use a minimal number of clamps and locators to
obtain an adequate measurement system repeatability
and reproducibility, or gage R&R. Other manufacturers
more readily obtain high gage repeatability
and reproducibility by adding a larger number
of clamps. This approach, masks variation in
the panel, making the measurement system less
able to detect variation. The difference in measurement
systems requires caution when generalizing
variation across manufacturers.
6
7
2.0 Stamping Variation
2.1 Components of Variation Explained
Dimensional variation from the stamping process
may be categorized into a number of components.
Generally, different variation components are
attributable to different sources and often have a
different impact on downstream operations. The
following are the general components of variation
that will be used throughout this report. Figure 2
below illustrates each variation component.
• Mean bias deviation is the process bias relative
to the design nominal. Mean bias is the absolute
value of the average deviation from nominal.
When a process is centered exactly at its nominal
dimension, its mean bias is zero. If, after a
single die set, for example at the die source tryout,
a mean dimension is -0.65 mm or 0.65 mm,
then its mean bias is 0.65 mm. If the mean from
two die sets are -0.40 mm and 0.10 mm, assuming
equal sample size from each die set, then the
grand mean is -0.15 and the mean bias of those
two die sets is 0.15 mm ([-0.40 + 0.10] ÷ 2 =
0.15).
• Part-to-part variation is also referred to as the
short-term or inherent variation. It is the amount
of variation that can be expected across consecutive
parts produced by the process during a
given run. The assumption is that the variation is
a reflection of numerous incidental random variables
over a short-term and is not affected by
any special causes of variation such as a
change in the steel coil or process settings. This
part-to-part variation is denoted as part-part.
Figure 2. Components of Variation
Tryout Regular Production
Die Source to Home Line
Mean Shift
Mean Shift
Die Source Tryout
Mean Bias
Upper
Specification
Lower
Specification
Nominal
Legend:
Individual Measurements
Mean of the Stamping Run
Part-to-Part Total Variation
Mean Bias
1.5
1
0.5
0
-0.5
-1
8
Estimates for part-to-part variation for the
36-panel study are based upon the 12 subgroups
of 3 consecutive panels. Again, the
assumption is that the process is stable during
three consecutive parts.
• Run-to-run variation is commonly referred to as
mean-shift variation. It is the measure of the
repeatability of the die setting process, and its
derivation is based on the variation of the mean
dimension across two or more die sets. Run-torun
variation is denoted as run-to-run. Estimates
for run-to-run variation are based on the variations
in mean dimensions between die sets.
• Begin-end of run variation is another type of
mean-shift variation in that it is a measure of the
stability of the process mean within a run. Since
stamping production runs can be long, the mean
of the run can change from the beginning to the
end, which may be several hours later. This
change in a mean dimension may occur due to
process changes during a run such as a steel coil
change, changes in operating speeds or tonnage,
or adjustments to draw lubrication. If a mean
dimension significantly shifts during a run due to
some special cause, the stable mean assumption
is violated and begin-end variation is greater than
part-to-part variation. We denote this variation as
begin-end. Estimates for begin-end of run variation
are based on the variation of the mean dimension
from the beginning to the end of each die set.
Figure 3 below illustrates a run chart for a single
stamping dimension with unusually large variation.
Each of the three variation components, part-topart,
run-to-run, and begin-end variation, is illustrated
in the plot.
Figure 3. Potential Sources of Stamping Variation
Run Chart of a Stamping Check Point
Part -
to- part
Measurement Values Mean of Group
4
3
2
1
0
-1
-2
within run
mean shift
run-to-run
mean shift
run 1 run 2 run 3 run 4 run 5 run 6
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
Measurement Value (mm)
9
• Mean shift variation is the sum of the run-to-run
variation and begin-end of run variation. Since
run-to-run variation and begin-end variation are
both forms of mean instability, they can be combined
into one variation number that is called the
mean shift variation. The mean shift variation is
denoted by mean shift, where 2 mean shift = 2 runto-
run + 2 begin-end. In most cases with stamping,
the run-to-run variation dominates the within run
variation ( 2 run-to-run >> 2 begin-end), so rather
than separate the two, the total mean shift variation
( 2 mean shift) is used.
• Total variation is the sum of part-to-part variation
and mean shift variation. This represents the
total variation that the downstream assembly
process is subject to over the long-term. The
total variation is denoted total. Equation 1 and
Figure 4 below summarize variation partitioned
into components.
Total variation is equal to the sum of the components
of variation:
2 total = 2 part-to-part + 2 mean shift, or
Equation 1
2 total = 2 part-to-part + 2 run-to-run + 2 begin-end
2.2 Calculating Components of Variation
Using ANOVA
An efficient method for estimating the components
of variation is through Analysis Of Variance, or
ANOVA. Briefly, the parameters for an ANOVA
model for this sampling plan were defined in the
following manner:
d = number of die sets = 6
s = number of groups = 2 samples of 3
per die set per batch
n = sample size per group = 3 consecutive
panels
The total number of panels sampled is equal to
dsn, or 6x2x3 = 36. This ANOVA model estimates
part-to-part, run-to-run and begin-to-end of run
variation using the expected Mean Squares (MS).
The equations shown in Table 3 on page 10 are
used to estimate the sources of variation. If
both factors of run-run and begin-end of run are
statistically significant, then a begin-end of run
and a run-run variance may be calculated. If only
one of the two factors is significant, then only that
variable will have a variance estimate. Finally, if
neither of the two factors is significant, then all of
the total variance may be attributed to part-part
variation.
2
(inherent process
variation)
total
2
part - to - part
2
run - to - run
2
begin-end
2
mean shift
Figure 4. Total Variation Partitioned into Components
10
The following table illustrates an application of an
ANOVA analysis for a stamping dimension.
Equation 2 2 = MSE mean squared error
(mean squared – mean
squared error) – sample size
= (MSBE – MSE)
n
Variation Formula Description
=
=
Equation 3
Equation 4a*
Equation 4b*
part-to-part
2
begin-end run begin-end
2
run-to-run
2
run-to-run
(MSRR – MSBE)
sn
(MSRR – MSE)
sn
begin-end
run-to-run
run-to-run
(mean squared – mean
squared error ) – (number
of samples x sample size)
(mean squared – mean
squared error) – (number of
samples x sample size)
Table 3. Formulas for Calculating Components of Variation
*Note: If all variation sources are significant, use Equation 4a. If begin-end factor is not significant, use Equation 4b.
Table 4. 36-Data Samples for a Stamping Dimension
Die Set Group Panel 1 Panel 2 Panel 3 Sample Average
1 begin run 0.13 0.16 0.17 0.15
1 end run 0.10 0.03 0.03 0.05
2 begin run 0.06 0.16 0.08 0.10
2 end run 0.13 0.08 0.23 0.15
3 begin run 0.18 0.72 0.14 0.34
3 end run 0.19 0.49 0.16 0.28
4 begin run -0.35 -0.43 -0.47 -0.42
4 end run -0.41 -0.39 -0.38 -0.40
5 begin run -0.32 -0.31 -0.35 -0.33
5 end run -0.32 -0.33 -0.29 -0.31
6 begin run -0.16 -0.10 -0.21 -0.16
6 end run -0.20 -0.17 -0.20 -0.19
Grand Mean (Mean Bias) -0.06 (.06)
11
Table 5. SPSS Output Calculations for Mean Squared Errors (all factors)
a. Mean square run-to-run
b. Mean square begin-run of run
c. Mean squared error
Source df Mean Square F Significance
Run-Run Hypothesis 5 .479a 103.81 .000
Error 6 4.6E-03b
Run-Run x Hypothesis 6 4.6E-03b .352 .901
Begin-End Run Error 24 0.013c
Table 6. SPSS Output Calculations for Mean Squared Errors without Begin-End Factor
Source df Mean Square F Significance
Run-Run Hypothesis 5 .479 25.836 .000
Error 30 0.011
The ANOVA output for this data is summarized in
Table 5 below, or using the Statistical Software
Package SPSS based on Type I error, = 0.05.
Note that a significant variable has a value less
than . For this data set, the Mean Squared Error
for the begin-end factor is not significant. That is,
the mean does not significantly change from the
beginning to the end of the stamping run.
Since the begin-end factor is not significant, the
ANOVA model must be revised and the Mean
Square Errors recalculated. The revised SPSS output
is shown in Table 6, where again significance
is based on a Type I error of 0.05.
The mean squared errors in Table 6 above may be
used to estimate the variation for each of the components
of variation present using Equations 2
through 4 (note: begin-end = 0 because this factor
is not significant for this dimension). The variation
estimates are shown in Table 7 below.
Table 7. Summary of Components of Variation Calculations
Variation Source Mean Squared Error Calculation Variance (mm2) Standard Deviation (mm)
Part-to part MSE 0.011 0.11
Begin-end run Not significant 0.00 0.00
Run-to-run (0.479 – 0.011) ÷ (2)(3) 0.078 0.28
Total Process (.078+.011) 0.089 0.30
Table 8 below summarizes the components of variation
and the mean bias at 12 measurement locations
for the Company A Body Side Outer as,
shown in Figure 5. One noteworthy finding is the
large range in variation for part-to-part, 0.01 to
0.48, and run-to-run, 0.00 to 0.18, across different
measurement locations. This contrast is attributable
to the differences in location/axis on the part
and also to the proximity of measurement system
clamps. It will be shown in the following section
that as the number of clamps increases on a
checking fixture, the amount of observed variation
decreases due to the masking of variation by the
clamps. Table 8 also indicates that part-to-part
variation, 65.4%, and run-to-run variation, 30.3%,
are much greater than begin-end run variation of
4.3% for these dimensions.
12
Figure 5. Body Side Outer for Company A: 12 Measurement Locations
#10 & #11
#12
#8
#9
#7
#5
#4
#1
#3 #2
#6
Table 8. Variance Summary for 12 Body Side Dimensions
Measurement Component of Variation (mm2)
Part-to-Part Begin-end Run-to-run Total Process Mean Bias (mm)
Location Direction ( 2 part-to-part) ( 2 begin-end) ( 2 run-to-run) ( 2 total)
1 Y 0.09 0.00 0.06 0.16 0.313
2 Y 0.04 0.00 0.00 0.04 0.342
3 Z 0.01 0.01 0.00 0.02 0.786
4 Y 0.05 0.03 0.00 0.08 1.187
5 X 0.48 0.00 0.00 0.48 0.851
6 Z 0.05 0.05 0.00 0.10 3.673
7 Z 0.06 0.00 0.03 0.09 1.609
8 Y 0.14 0.00 0.18 0.32 2.530
9 X 0.34 0.00 0.17 0.51 1.139
10 Y 0.01 0.00 0.00 0.01 0.618
11 Z 0.02 0.00 0.03 0.05 0.837
12 Y 0.14 0.00 0.18 0.32 0.675
Average 0.12 0.10 0.05 0.18 1.130
Percent of Total 65.4% 4.3% 30.3% 100.0 ––
2.3 Description of the Sources of Stamping
Variation
Extensive research has been conducted regarding
dentification and elimination of the sources of
variation associated with stamping sheet metal.
The stamping process is complex, with many
variables that can influence variation. One related
research effort, by John Siekirk,(1) identified 30
major factors, and then classified them into the
following seven categories:
• Blank condition,
• Blank lubrication,
• Stamping press variables,
• Metal properties,
• Die condition,
• Miscellaneous and
• Interactive variables.
Because this body side research project investigated
process variation under production conditions,
only a limited number of process and material
variables could be collected. More importantly,
process variables were not purposely altered.
Therefore, only inferences can be made between
stamping variation and the observed variability in
process variables. In many of these case studies,
the process and material variables were under
control. As a result, our findings do not necessarily
identify those variables that could affect part
variation, but rather which variables explain the
dimensional mean shifts in these case studies.
One area of this research that is often not examined
is the effect of process variables on mean
conformance. Most of the research on reducing
stamping mean biases has been directed toward
metal forming and die design. Little research
exists on eliminating mean biases once a die has
been made and the actual mean biases become
known. Even less attention has been given to nondie
related influences on mean bias such as the
measurement system effects. Among the factors
that influence mean bias include:
• Measurement System:
• Clamping sequence
• Clamping forces
• Part locating (datum)
• Product Design:
• Part geometry (size and complexity)
• Part rigidity (shape and gage)
• Check point location
• Process:
• Press setup and control of process
variables (see above)
• Changes in stamping presses
(e.g., tryout to production presses)
• Material handling and storage
13
1 Process Variable Effects on Sheet Metal Quality, Journal of Applied Metalworking, American Society for Metals,, July 1986.
14
3.0 Analysis of Stamping Variation
3.1 Mean Conformance
One of the greatest challenges in die making and
stamping is minimizing mean biases for dimensions
on stamped parts. As defined earlier, the
mean bias is the absolute value of the average
deviation from nominal. Ideally, manufacturers
would produce every stamped component such
that each dimension is, on the average, at the
specification nominal. By doing so, design capability
(Cpk) would be maximized for a given level of
process variation. Achieving minimal mean biases
in stamping also facilitates the “tune-in” of assembly
tooling, which is initially designed for parts at
nominal, and increases the likelihood of producing
dimensionally acceptable assemblies within the
shortest possible lead-time. The problem is that no
manufacturer in the world has demonstrated the
capability to produce stamped body parts without
mean biases.
Manufacturers who have minimized their mean
biases relative to their competition appear to
maintain a competitive advantage in terms of cost,
quality and lead-time. To achieve lower mean
biases, manufacturers employ a combination of
technology and applied learning and limit the evolution
of product design to reduce uncertainty.
Future product designs with uncertain forming
challenges might be subject to soft tool evaluation
in order to evaluate metal forming and die design
before production tools are machined.
Modifying hard dies, or die rework, after they have
been machined to reduce mean biases represents
one of the most difficult tasks in getting dies
approved for production. Manufacturers attempting
to rework dimensions to reduce mean biases
face several challenges. First, since a stamped
part has a continuous surface, reworking a die to
shift one dimension may affect other areas of the
part. Many areas of a part are interdependent, so
that when one dimension changes another area
does as well, sometimes in an unpredictable way.
Another difficulty is trying to rework dimensions
exactly to their design nominal. Basically, there is
a limited ability to hit the nominal dimension even
after rework. A final difficulty concerns the ability to
measure a part and to know precisely what the
mean bias really is. In addition to die processing,
mean dimensions also are affected by the number
and positioning of clamps in measurement fixtures.
Because of the many variables in forming a
part, such as changes in stamping press variables
and steel properties, and the limited ability to
measure sheet metal, ascertaining the precise
mean bias can be very difficult - both before and
after a die change. Manufacturers often face a
complicated decision in determining when to
rework a die or when to allow a mean bias to
remain (see Body System Analysis Project Team
report “Event-Based Functional Build: An
Integrated Approach to Automotive Body
Development”).
3.1.1 Benchmark Comparison - Body Side
Outer and Inner Panels
Figure 6 on page 15 shows a histogram for 143
mean dimensions across 5 parts at Company C.
Several observations may be made from these
data:
• The distribution of mean dimensions is approximately
normal. Assuming the measurement system
does not unfairly influence mean deviations,
this finding suggests an inherent variation in the
ability to design and construct dies to produce
part dimensions at nominal.
• The distribution of mean dimensions is centered
approximately at zero (i.e., average mean bias
is near zero). This is as expected since the
distribution is normal and the die maker's target
is to have zero bias.
• Approximately 10% of mean values have a bias
greater than 1.0 mm, and about 35% have a
bias greater than 0.5 mm.
15
In general, the amount of spread in the mean distribution
will vary significantly by type of part.
Larger, less rigid panels like the body side outer
often have significantly more dimensions with
large mean biases than rigid panels, or panels
with blank thickness greater than 1.5 mm.
Figure 7 below compares the mean deviations for
smaller rigid reinforcement panels at Company A,
such as A and B pillar reinforcements, to larger,
less rigid panels, such as one-piece body side
outer and quarter inner. The less rigid panels have
larger mean biases and also a greater dispersion
in mean deviations than the rigid panels. This is
evident in comparing reinforcements to a onepiece
body side, and it also occurs in comparing
a one-piece body side of 0.69 mm gauge to a twopiece
body side of 1.10 mm gauge as shown in
Figure 8 on page 16.
45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
Range of Mean Deviations
% of Dimensions (143 total)
<-1.25 -1.25~-.75 -0.75~-.25 -0.25~+.25 0.25~.75 0.75~1.25 >1.25
65%: [Mean] <0.5 mm
Figure 6. Histogram of Mean Values across 5 Parts for Company C
Figure 7. Mean Conformance: Rigid vs. Non-Rigid Panels
100%
80%
60%
40%
Body Side Otr/Qtr Inr (Non-rigid) Front/Center Pillar Reinforcements (Rigid)
Range of Mean Deviations
Non-rigid: 44% [Mean] <0.5
Rigid: 83% [Mean] <0.5
20%
0%
<-1.5 -1.5 ~ -0.5 -0.5 ~ 0.5 0.5 ~ 1.5 > 1.5
% of Dimensions
Table 9 below summarizes the mean bias for the
body side outer panels at each of the automotive
manufacturers. These data suggest several generalizations.
First, larger one-piece body side outers
with integrated quarters tend to exhibit greater
biases than two-piece body sides. Second, manufacturers
using constrained measurement systems
such as excess part locating clamps have significantly
less mean deviations. Companies E, F, and
G use constrained measurement systems, and all
have lower biases for the body side outer panel. In
addition, the same body side panels at Company
B exhibited less mean bias when measuring the
parts in a more constrained fixture. A major influence
of the constrained checking system is that
extreme mean biases, those greater than 1.0 mm,
are greatly reduced.
16
Figure 8. Mean Conformance: Two-Piece Body Side Panel vs. One-Piece
80%
60%
40%
One-Piece Body Side Two-Piece Body Side
Range of Mean Deviations
One-Piece: 31% [Mean] <0.5
Two-Piece: 65% [Mean] <0.5
20%
0%
<-1.5 -1.5 ~ -0.5 -0.5 ~ 0.5 0.5 ~ 1.5 > 1.5
% of Dimensions
Table 9. Mean Conformance by Company
Body Side # cross car Average % Dimensions % Dimensions % Dimensions
Company Type clamps in fixture [Mean] [Mean] <.5 [Mean] >1 [Mean] > tol (t)
A Integrated Quarter 11 1.10 34% 56% 66%
B (remeasured) Integrated Quarter 14 0.73 49% 29% 39%
C Two-piece 7 0.51 65% 15% 5%
D Two-piece 8 0.88 42% 39% 39%
E Two-piece 22 0.36 74% 3% 14%
F Two-piece 16 0.31 84% 3% 39%
G* Integrated Quarter 17 0.37 69% 2% 28%
* Over-constrained (excess clamps) during measuring
The effect of a constrained measurement system
is limited to larger, less rigid panels since additional
clamps beyond 3-2-1 on rigid parts with
gauge greater than 1.5 mm have little or no effect.
Table 10 below compares mean conformance
across several part types. Although the clamping
strategy may be correlated with mean bias in the
body outer panels, the same cannot be done for
rigid panels.
17
Table 10. Mean Bias by Type of Part
Body Side Non-Rigid Rigid Inner Body Side Non-Rigid Rigid Inner
Outer Inner Panels Panels Outer Inner Panels Panels
Average Average Average % Dimensions % Dimensions % Dimensions
Company [Mean] [Mean] [Mean] [Mean]>1 [Mean]>1 [Mean]>1
A 1.10 0.79 0.22 56% 30% 1%
B 0.90 0.56 no data 33% 16% no data
C 0.51 0.31 0.34 15% 0% 3%
D 0.88 0.32 0.27 39% 0% 0%
E* 0.36 0.35 no data 3% 0% no data
F* 0.31 0.38 no data 3% 17% no data
G* 0.37 0.39 no data 2% 6% no data
* Over-constrained (excess clamps) during measuring
3.1.2 Mean Bias and Part Tolerances
Another contrast across manufacturers is the
assignment of part tolerances. When comparing
manufacturers, the same physical dimension on a
body side may have a tolerance of +/- 0.3 mm at
one manufacturer and +/- 1.25 mm at another.
Table 11 on page 18 shows the typical tolerance
for the body side panel and the percentage of
dimensions whose mean biases exceeds the tolerance
limit. On average, more than 30% of the
dimensions in companies A through G have their
mean bias outside the tolerance. It is important to
note that whenever the mean bias exceeds the tolerance
limit, at least 50% of the panels have that
dimension outside of tolerance. It is clear that a
significant number of vehicles are being produced
with acceptable final body quality, but with a significant
number of body panel dimensions out of
tolerance.
Another observation across study participants is
that although Companies A through C use Cpk as
their principal buyoff criteria, they do not achieve
greater mean conformance. In fact, it might be
argued that the use of Cpk at Company C has led
primarily to wider tolerances to achieve greater
Cpk conformance, not greater mean conformance.
Another finding is that only those manufacturers
using constrained measurement systems
assigned tolerances less than ±0.70 mm.
Company F assigns the tightest tolerance at ±0.3,
but uses a constrained measurement system and
also has a two-piece body side which tends to
have lower mean bias than the larger one-piece
design.
18
3.1.3 Benchmark Comparison - Tryout versus
Production
Many manufacturers apply common dimensional
validation procedures and criteria to all body panels
even though the expected mean bias differs by
type of panel, whether rigid versus non-rigid or
small/simple form versus large/complex form.
Table 12 below depicts mean conformance across
multiple parts during regular production at
Companies A, B and C. These data suggest that
manufacturers produce stamped parts with 50-
70% of dimensions within 0.5 mm. Comparing
these findings with mean biases experienced at
production buyoff, the data are consistent.
Although the other four manufacturers did not provide
tryout data, discussions with their personnel
suggest that their mean conformance distributions
in production also corresponded to die tryout. The
main point is that even though manufacturers may
adjust some mean biases to correct build concerns,
the overall ability to produce mean dimensions
at nominal does not significantly change
from die tryout.
Table 11. Mean Conformance and Tolerances
Body Side Typical # cross car Average % Dimensions % Dimensions
Company Type tolerance clamps in fixture [Mean] [Mean] >tol (t) Cpk > 1.33
A Integrated Quarter +/- 0.7 11 1.10 66% 15%
B Integrated Quarter +/- 0.7 14 0.73 39% 80%
C Two-piece +/- 1.25 7 0.51 5% 75%
D Two-piece +/- 1.0 8 0.88 39% 23%
E Two-piece +/- 0.5 22 0.36 14% 43%
F Two-piece +/- 0.3 16 0.31 39% 29%
G Integrated Quarter +/- 0.5 17 0.37 28% 37%
Table 12. Summary of Mean Bias: Tryout vs. Production (Case Study Parts)
Company % Dimensions % Dimensions % Dimensions % Dimensions
[Mean]<0.5 [Mean]<0.5 [Mean]>1 [Mean]>1
A 59% 63% 26% 22%
B 51% 53% 15% 23%
C 64% 66% 13% 10%
Tryout Production Tryout Production
3.1.4 Mean Bias Stability over Time
Another important consideration regarding mean
bias concerns its stability over time. Most automotive
manufacturers first evaluate mean bias during
die source tryout. A decision is eventually made to
move the die to the production press, often
referred to as the “home line”, where another estimate
of the mean bias is made. Finally, as the dies
are repeatedly run on the home line for production,
each die setup provides another opportunity to
estimate the mean bias. Most of the data in this
study was collected during production, a year or
more after the dies were initially brought to the
home line. An important question affecting dimensional
validation is how does the estimate of mean
bias change from tryout to the home line and then
to future production.
19
Table 13 below examines changes in part dimensional
means between die source tryout and home
line tryout. The first two data sets are based on the
case study parts at two of the manufacturers. Two
more extensive studies of die source to home line
mean shifts are also included. These data suggest
that approximately 30% of dimensions shift at least
0.5 mm when the dies are moved from the tryout
presses to the home line. The amount and uncertainty
of change is one reason manufacturers recognize
that it is necessary to re-evaluate the
dimensions on a part when the dies are transferred
to the home line. Interestingly, a similar
number of dimensions shift toward nominal as
opposed to away from nominal, or the shift in
mean bias from tryout to the home line appears
random.
Table 13. Comparisons of the Change in Mean Bias from Tryout to Home Line
Company # of Parts/ Median Shift [Die source % Dimensions of the Dimensions
# Dimensions mean. Home line Mean] [Mean Shift] >0.5 with shift >0.5
% closer % away
B-1 4/104 0.30 30% 40% 60%
C-1 5/86 0.20 25% 68% 32%
C-2 47/652 0.23 30% 63% 37%
C-3 26/182 0.27 28% 50% 50%
Overall 0.25
Die Source to Home Line
Figure 9 below compares the mean dimension at
time of part approval versus the production mean
approximately one year later at Company C. For
the parts in this study, nearly 50% of the dimensions
shifted more than 0.5 mm over the life of the
program. Table 14 on page 20 shows further that
of the dimensions with significant mean differences,
a similar number shifted closer to nominal
than away from nominal.
Figure 9. Correlation of Mean at Part Approval vs. Production Mean
2.00
1.50
1.00
0.50
0.00
-0.50
-1.00
-1.50
-2.00
-2.00 -1.00 0.00 1.00 2.00
Correlation, R = .12
Home Line Mean at Part Approval
Production Mean
These findings suggest that automotive body
parts continue to evolve from die source tryout,
through home line tryout, and even through regular
production. Although some of these dimensional
changes are intentional based on die rework to
correct a problem, the majority are not. They shift
because of lack of process control or die rework or
maintenance in another related dimensional area.
Interestingly, some dimensions shift away from
nominal with no apparent impact on the assembly
process. In addition, some dimensions may shift
significantly closer to nominal, greater than 1 mm,
but with an adverse affect on the final assembly.
Figure 10 below depicts a stamping dimension
that shifts 1.3 mm between stamping runs four and
five. Even though this shift is toward nominal, the
variation observed in assembly actually becomes
higher.
In this example, maintaining a stable mean over
time appears more important than the magnitude
of the mean deviation. Similar to stamping,
assembly processes evolve over time to match
stamping mean deviations. If these mean deviations
change significantly, assembly processes
will likely experience problems. Thus, manufacturers
must develop a better understanding of how to
minimize mean instability. Fortunately, mean instability
is not inherent to a process like part-to-part
variation, rather it is caused by some special influence
such as a process variable change or die
rework. Thus, a potential exists to control these
special causes.
20
Table 14. Change in Mean from Home Line to Long-term Production
Company Median Shift % Dimensions % of the Dimensions
[Home line mean- [Mean Shift] >0.5 with shift >0.5
Production] % closer % away
A 0.48 48% 41% 59%
B 0.57 53% 45% 55%
C 0.35 40% 44% 56%
Home Line Approval Mean to Production Mean
Figure 10. Effect of Stamping Mean Shift on Body Side Assembly
(Note: above dimension is coordinated between stamping and assembly)
4.5
1 2 3 4
Stamping
Stamping .30
.24
.19
.80
run 1-4 run5-6
1.3 mm mean shift
Assembly
Assembly
5 6
Measurement, mm
Stamping Run #
4
3.5
3
2.5
2
1.5
1
0
0.5
21
3.1.5 Impact of Shipping on Mean Bias
One final investigation into the factors influencing
mean bias looked at the impact of material handling,
including:
&#8226; Racking of parts and container design for consistency
and impact resistance,
&#8226; Time lag as stressed parts become stress
relieved and
&#8226; Movement of parts - including manual, forklifts,
and truck and rail mass transit, all of which
impact the distortion of parts through vibration.
An experiment was performed at Company A
where four inner panels were measured both
before and after shipment. The measurement system
used the same locating fixtures and CMM programs
at both the production and assembly
plants; however, different operators performed the
actual measurements. The panels were shipped in
their specified containers, via truck over several
hundred miles. Two small parts were dropped into
bins, one larger wheelhouse outer was stacked
and the fourth part, the quarter inner, was shipped
in a special rack. The results are shown in Table 15
below.
Table 15. Summary of Panels Measured Before and After Shipping
Number of Average Mean % of Dimensions With
Part Check Points Bias Shift (mm) Mean Bias Shift>0.2mm
Wheelhouse Outer 69 0.89 76%
Quarter Inner 91 0.10 15%
B-pillar Reinforcement 59 0.16 19%
A-pillar Reinforcement 70 0.08 6%
This experiment indicates a potentially significant
impact of shipping on mean bias. It should be
noted, however, that the impact of shipping is confounded
by different measurement operators.
Reproducibility is a potential source of gage error
in this study because the same operator did not
measure the panels before and after shipping.
However, reproducibility of a CMM based on an
automatic program is generally insignificant.
The wheelhouse outer suffered the greatest mean
shift with an average change of 0.89 mm. The
wheelhouse panels at the bottom of the stack had
the largest dimensional differences, reflecting the
accumulated weight effect. For the quarter inner,
special racks are used which clearly help reduce
the shipping effect. The more rigid panels also
experienced less shipping impact, with most
mean dimensions shifting less than 0.2 mm.
Similar to the die source tryout to home line analysis,
the direction of the mean shift appears random,
or equally likely to get closer to or further
away from nominal.
3.2 Stamping Process Variation
3.2.1 Benchmark Comparison - Part-to-Part
Variation
Part-to-part or short-term variation is a measure of
the inherent variation for a particular product, or
set of dies, and process, or stamping press line.
Key variable set-up parameters, such as shut
height, lubrication, cushion pressure, etc, and
incoming steel coils or blanks are presumed to be
constant or consistent. Several variables may
explain differences in part-to-part variation across
companies. Some of these differences were investigated,
including:
&#8226; measurement and clamping system,
&#8226; check point location/axis on the part, and
&#8226; part rigidity, size and material thickness.
Table 16 on page 22 summarizes part-to-part variation
for the body side outer panels for each of the
manufacturers studied. A comparsion of variation
across manufacturers is again difficult because of
the different measurement strategies as demonstrated
by the number of clamps. The three manufacturers
using the most clamps (E, F, and G) have
the lowest part to part variation. In addition, part to
part variation at Company B is significantly lower
when using a more constrained measurement system.
These data suggest that adding measurement
clamps will likely reduce the observed part
to part variation for large, non-rigid parts. Both
the average and the extreme variation points, or 6
part-to-part greater than 1.5 mm, appear to be significantly
reduced by the additional secondary
locating clamps.
The type of body side, one-piece versus twopiece,
also appears to affect variation. Companies
A through D use roughly the same number of
clamps, but have two different body side styles,
integrated quarter panel and two-piece. It appears
that the two-piece body side results in lower average
part-to-part variation than the larger and more
complex integrated quarter body side by about
10%. The same relationship is seen among companies
E, F, and G using the more constrained
measurement approach. Of these, company G
with the larger body side has the highest 95th percentile
part to part variation. Follow-up analysis at
company G indicates that most of their high variation
dimensions are in non-stable measurement
areas in the tail area of the body side outer panel.
Since part-to-part variation differs according to
body side style, it would be expected to vary
according to part size and rigidity for non-body
side outer panels. The panels in these case studies
may be grouped into three categories: body
side outer panel, non-rigid body side inner panels
and rigid body side inner panels. The body side
outer panel is the largest and one of the lightest
gauge panels, varying from 0.69 mm to 0.90 mm
thickness. The non-rigid body side inner panels
are arbitrarily limited to 1.5 mm thickness and are
smaller than the outer panel. These panels include
the quarter inner, wheelhouse outer and roof rails.
The third category consists of small, heavy-gauge
parts, including the A- and B-pillar reinforcements.
Figure 11 on page 23 plots the average standard
deviation, or sigma, for all parts studied at the
seven manufacturers. The changes in these
groupings from large and flimsy to smaller and/or
more rigid can be seen to correlate with the average
amount of part-to-part variation. As panels
become smaller and more rigid, their part to part
variation decreases. In addition, Figure 11 suggests
that the body side panels with the lowest
variation are from manufacturers using more
measurement clamps, thus masking some of the
actual process variation.
22
Table 16. Part-to-Part Variation for Body Side Outer Panels
Note: 95th percentile is the level of variation where 95% of the dimensions on the part are less than this amount.
Body Side # cross car Average 95th Percentile % Dimensions
Company Type clamps in fixture 6 part-part 6 part-part 6 part-part >1.5
A Integrated Quarter 11 1.14 2.90 20%
B (remeasured) Integrated Quarter 14 1.09 2.35 23%
C Two-piece 7 0.99 1.89 18%
D Two-piece 8 0.99 1.57 10%
E* Two-piece 22 0.48 0.81 0%
F* Two-piece 16 0.32 0.50 0%
G* Integrated Quarter 17 0.40 1.08 0%
* Over-constrained (excess clamps) during measuring

fengwei-08

23
Another difference among manufacturers is the
number and location of dimensions measured.
Two manufacturers, companies D and E, collect
less data on their stamped panels than the other
manufacturers, and primarily collect data from
points located in more rigid localized part areas.
Although a body side outer panel tends to be flimsy,
certain areas in highly formed sections of the
part, such as the door openings, are typically
more rigid than the tail or wheelhouse areas.
Control of these more rigid areas often is more
important than other areas because they are less
likely to conform to reinforcements during assembly.
As has been shown, dimensions on less rigid
parts tend to have greater variation. In order to
illustrate the impact of dimension location, Table
17 below shows the body side variation for companies
D and C. At company D, 24% of their body
side dimensions have an average standard deviation
greater than 0.2 mm. Company D measures
near the A- and B-pillars and on the flanges in the
door openings. In contrast, company C measures
dimension throughout the body side and has 73%
of their dimensions exceeding 0.2 mm. However,
when comparing dimensions in similar locations,
the variability at company C more closely resembles
company D. Thus, the expected variation on
a stamped panel appears dependent upon where
the dimension is located and how rigid the part is
at that location.
Figure 11. Average Variation (Standard Deviation) by Type of Part
0.30
0.25
0.20
0.15
0.10
0.05
0.00
integrated
quarters
Body Side Non-Rigid Rigid (guage>1.5)
Average part-part
small,
E, F, G simple
6 =1
Table 17. Effect of Dimension Location on Variation
Selected
Company Dimension <0.2 >0.2
D 14 76% 24%
C 40 27% 73%
C 14 (common with D) 60% 40%
3.2.2 Variation Over Time
In theory, part to part variation produced from a set
of dies on the same press line should remain constant
over time. In practice, part-to-part variation
does vary for some dimensions. Variables that
may affect part-to-part variation over time include:
&#8226; The condition of the press line, a function of the
level of maintenance of the presses,
&#8226; The condition of the dies, a function of die maintenance
and engineering change rework, and
&#8226; Processing variables, such as the control of
cushion pressure, material handling, automation
between presses, etc.
Although many of these changes often are associated
with mean shifts, part-to-part variation can be
affected as well. Table 18 below shows that partto-
part variation typically increases from part
approval runs to regular production. These data
suggest that the average six sigma increases from
0.8 mm to 1.2 mm after more than a year in production.
The most likely explanation for this difference
is that operating conditions at buyoff are
substantially more controlled than in regular production.
Although the overall variation increases,
not every dimension exhibits an increase. Figure
12 below compares the observed part-to-part
standard deviation at buyoff versus regular production.
This illustration indicates a general lack of
correlation between part approval variation and
regular production. For some dimensions, the variation
increases and for others it decreases,
although more dimensions have higher part to part
variation in production.
24
Table 18. Part-to-Part Variation: Home Line Approval vs. Production by Company
Note: production data 1 year + after home line buyoff
Company # Parts Average Average % Dimensions % Dimensions
(# Dimensions) 6 part-part 6 part-part 6 part-part>1 6 part-part>1
A 1 (37) 0.79 1.16 14% 48%
B 5 (132) 0.96 1.32 26% 48%
C 39 (327) 0.84 1.14 23% 38%
Home Line Production Home Line Production
Figure 12. Part-to-Part Variation: Home Line Tryout Approval vs. Production by Dimension
0.80
0.60
0.40
0.20
0.00
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80
Correlation, R = .21
part-part – Home Line Tryout Approval
part-part – Production
25
3.2.3 Impact of Shipping on Variation
As previously mentioned, part shipping from the
stamping plant to the assembly plant caused several
mean dimensions to shift, particularly for the
non-rigid wheelhouse outer panels. This section
examines the effects of shipping on variation. As
noted earlier, some potential operator noise exists
because different operators measured the panels
before and after shipping. However, this operator
effect is unlikely to be significant, as the panels
were measured using the same fixtures and automated
CMM programs.
Table 19 below indicates that part-to-part variation
increased on 87% of the dimensions for the four
parts: wheelhouse outer, quarter inner, A-pillar
reinforcement, and B-pillar reinforcement, with a
lesser increase on the more rigid components, the
A- and B-pillar reinforcements. Clearly, part shipment
increases part-to-part variation.
Table 19. Summary of Remeasured Data Before and After Shipping (via truck)
Panel Measurement Points Variation Increased
Wheelhouse Outer 69 91%
Quarter Inner 91 92%
B-Pillar Reinforcement 59 86%
A-Pillar Reinforcement 70 76%
3.2.4 Components of Variation: Part-to-Part,
Run-to-Run, and Begin-End of Run
Stamping variation may be broken down into three
components of variation: part-to-part, run-to-run,
and begin-end of run (see Section 2.0). Total variation,
total, is a statistical summation of these three
variation components. One reason for looking at
the components of variation individually is that the
each is a reflection of different root causes. Table
20 below shows the part-to-part and total variation
for each auto company's body side outer panel.
Table 20. Summary of Part-to-Part and Total Variation for the Body Side Outers
Body Side # cross car Average Average % Dimensions
Company Type clamps in fixture 6 part-part 6 total 6 total >1.5
A Integrated Quarter 11 1.14 1.41 29%
B Integrated Quarter 14 1.09 1.93 57%
C Two-piece 7 0.99 1.88 42%
D Two-piece 8 0.99 1.21 23%
E Two-piece 22 0.48 0.52 0%
F Two-piece 16 0.32 0.49 0%
G Integrated Quarter 17 0.40 0.77 3%
Companies E, F, and G, which used the most constrained
measurement systems at 16, 17, and 22
clamps respectively on the body side outer panel,
have the lowest part-to-part and total variation.
Comparing companies C and D, excluding the
clamping effect, showed that although the two
manufacturers exhibit similar part-to-part variation,
company C has much higher total variation. Figure
13 below shows that company C has significantly
more run-run and begin-end of run mean shifts.
Thus, company C does not appear to control their
process as well as company D. A similar finding is
observed in comparing companies A and B with
company D. Among companies E, F, and G, the
constrained measurement companies, company
G appears to have less control over their mean
shift variation. In general, manufacturers with similar
panels and similar checking systems should
have similar levels of dimensional variation. When
they do not have similar levels of variation, the difference
typically is not related to inherent part-topart
variation, but rather to how well one manufacturer
controls its process over time.
Table 21 on page 27 shows the amount of variation
for each of the parts studied at company A, by
source of variation. The sample size for each type
of panel is 36 right and 36 left, or 72 total for each
type. The numbers expressed in Table 21 are
averages across all the dimensions on a part and
therefore are non-additive. These data indicate
that less rigid panels exhibited the largest part-topart
and mean shift variation. Interestingly, the
variation for a particular component is not always
the same for right and left mirror image parts. At
company A, the right hand body side outer
exhibits significantly less variation than the left
side. Overall, variation at company A is relatively
low with the exception of the left body side.
Although part-to-part variation is typically larger
for a one-piece body side, the principal reason
that the left side has significantly higher variation
than the right side is due to mean shifts between
stamping runs.
26
Figure 13. Components of Variation for Body Side Panel at Company C and D
(Note: total is greater at Company C due to mean shifts not part-part variation)
% of Total Observed Variation
Part-part Run-run Begin-end of run
Company C Company D
80%
100%
60%
40%
20%
0%
90th Percentile part-part C: 0.27 D: 0.24
90th Percentile total C: 0.47 D: 0.29
46%
35%
20%
71%
27
Table 22 below shows the percentage of total variation
at company C according to variation source:
part-to-part and mean-shift (run-run and/or beginend).
The effects of mean shifts at company C are
more significant than company A. The variation of
the body side, front pillar and center pillar reinforcements
are approximately doubled due to
mean shifts. An analysis of the roof rail and windshield
frame suggests one potential challenge in
assessing mean shifts. Because analysis of variance
methods are used to estimate mean shift
variation, higher part-to-part variation will mask
mean shift variation. In other words, the true mean
shift variation cannot be effectively evaluated if the
inherent variation is unstable, a violation of the
homogeneity of variance assumption used in
ANOVA models.
Table 21. Sources of Variation by Part for Company A
Average Average Average Average % of Variation
Part run-run begin-end part-part total Explained by
Mean Shifts
Body Side - RH – 0.15 0.19 0.24 31%
Body Side - LH 0.26 0.15 0.26 0.34 43%
Quarter Inner 0.07 0.07 0.08 0.10 43%
Wheelhouse Outer 0.11 0.09 0.09 0.14 50%
B-Pillar Reinforcement 0.04 0.07 0.05 0.08 59%
A-Pillar Reinforcement 0.06 0.05 0.07 0.08 28%
Table 22. Sources of Variation by Part for Company C
Average Average Average % of Variation
Part mean-shift part-part total Explained by
Mean Shifts
Body Side - RH 0.26 0.17 0.31 79%
Roof Rail 0.23 0.28 0.34 32%
Front Pillar Upper 0.16 0.09 0.18 76%
Front Pillar Lower 0.15 0.09 0.18 76%
Center Pillar 0.21 0.08 0.23 92%
Windshield Frame 0.15 0.20 0.22 22%
3.2.5 Steel Properties and Press Setup
Control and Stamping Variation
These case studies under production conditions
provide an opportunity to investigate possible root
causes of mean shift variation. Short-term or partto-
part variation is assumed to result from several
factors related to product design, part size and
rigidity, die design, stamping press condition or
the measurement system. Mean shift variation runto-
run and within run, however, is generally related
to changes in the process over time, such as the
repeatability of press setup or changes to material
properties. Although this study did not provide an
opportunity to rigorously control variables to
ascertain direct cause and effect relationships
between process input variables and variation, it
does allow for some general conclusions regarding
the causes of mean shifts.
Five manufacturers collected input data for both
process and material variables across thirty parts.
(companies E and F did not participate). They collected
this data for each sampling of three panels,
or, in some cases, once per run. The material
coupons were analyzed later, either at an independent
test laboratory, three participants, or inhouse,
two participants. The following variables
were collected when possible:
&#8226; Process data (at each setup)
- Draw press shut height
- Draw Tonnage
- Die cushion pressure (if applicable)
- Outer ram tonnage (if double-action
press used)
&#8226; Material data (a steel coupon was sampled
when a sample of parts was taken from the
production run)
- Gauge
- Yield strength
- Ultimate strength
- n-value
- Percent elongation
Due to data collection limitations, it was not possible
to match process and material variable data
directly to a particular panel. For example, the
material properties of the steel for each individual
panel are unknown. Thus, the analysis is limited to
trying to explain mean shift variation and not partto-
part variation. For instance, if mean shifts
account for only 20% of the total observed variation,
then the most variation that can be explained
with the input variables collected is 20%. This
analysis only identifies relationships between
control of input variables and mean shifts.
Of the thirty parts with process input data, approximately
33% of the dimensions, 330 out of 1135,
had at least one large mean shift greater than 0.5
mm over the data collection period. Thus, prior to
any mean shift analysis, over two-thirds of the
dimensions studied were found robust to the variability
of their respective process and material
input variables.
The next step was to examine the relationship
between process variable control and mean-shift
variation. Table 23 below compares mean shift
variation with process input variation using
allowed ranges. Allowed ranges are essentially the
tolerances of the process and material input variables.
Thus, if manufacturers control their
process-input variables within these ranges, they
should not observe significant mean shifts related
to these variables. Generic allowed ranges are
used instead of tolerances to permit comparison
among manufacturers with different process and
material variable specifications. Furthermore,
since this analysis only looks for relative variation
differences, the nominal or average value of each
variable is not important.
Table 23. Summary of Product and Process Variation Compliance
% Parts within Correlation, R,
Variable Robust Range Robust Range to mean-shift
Material Gauge 0.06 mm 96% 0.23
Yield Strength 6 ksi 95% 0.22
Ultimate Strength 6 ksi 92% 0.24
% elongation 9% 100% 0.19
n-value 0.04 100% 0.09
Inner Tonnage 60 tons 45% 0.69
Outer Tonnage/Cushion 50 tons/+/- 10% 48% 0.47
28
Table 23 shows that most steel variables, 92% to
100%, fall well within their expected ranges of
variation. Consequently, it is not surprising to see
that their correlation with mean shifts is relatively
low with values ranging from 0.09 to 0.24, where 0
has no correlation, 1.0 is a perfect correlation and
a value greater than 0.6 is considered correlated.
In general, the steel manufacturers studied had
control of their variation, and even when they did
not, material property variability could not be
correlated with dimensional mean shifts.
The process variables of inner tonnage, outer tonnage,
and cushion pressure, had considerably
more variation and operated within the allowed
range only 45% to 48% of the time. The result was
a much higher correlation to mean shifts.
Presumably, the opportunity for variation reduction
for these part dimensions is significant if the press
setup variable of tonnage and cushion pressure
can be controlled more tightly. Figure 14 below
suggests an observed threshold of around 75 tons
as the limit to the allowable range of variation for
controlling dimensional mean shifts. Note that
these observed ranges relate only to dimensional
mean shifts and do not consider potential impacts
on formability issues such as splits or wrinkles.
A few additional comments with respect to this
analysis are appropriate. First, tonnage readings
may be affected by several setup variables such
as lubrication, die placement in press, shut height,
etc., and thus the correlation to mean shifts should
be viewed principally as an indicator of lack of
process control. Second, the relationship between
tonnage and mean shifts over a continuous range
of tonnage settings was not analyzed scientifically
for every part. Therefore, these data should not be
used to identify tonnage specifications for a particular
part. Rather, simply recognize that those
parts in this study exhibiting large mean shifts
tended to have relatively poor control of the
process variables but good control of the material
variables.
Figure 14. Relationship between Press Tonnage and Mean Shift Variation ( mean shift)
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00
0 100 200 300 400
Range of Draw Die Tonnage
mean-shift
(<75)
Range 90-300
3.2.6 Effect of Mean Shifts on Statistical
Process Control Techniques
All manufacturers in the benchmark study exhibited
some level of mean shift variation for the majority
of their part dimensions as shown in Table 24 on
page 30. Of the 1287 dimensions examined,
approximately 80% would have at least one subgroup
plot out-of-control on an X-bar chart.
However, only 20% of the dimensions had a mean
shift greater than 0.5 mm. Note that the majority of
these mean shifts occurred on the parts at
companies B and C. Again, these mean shifts
largely explain why certain manufacturers have
more variation in their process than others.
29
The fact that such a large percentage of dimensions
would plot out-of-control on an X-bar chart
has serious implications for process control. One
interpretation is that stamping and die processes
by nature are not stable enough to produce parts
with stable mean dimensions, even at world-class
facilities. Another interpretation is that the part-topart
variation of a stamping process often is so low
that even well maintained processes will exhibit
some process drifts over time. Assuming, for
example, that the inherent standard deviation of a
stamping process is 0.10 mm, a process will be
deemed statistically out-of-control if a mean shifts
by more than 0.15 mm(2). Most manufacturers
would not want to adjust a process for a 0.15 mm
mean shift.
Assuming that small mean shifts are inevitable
with the die changeover process, the traditional
use of X-bar charts to assess mean stability may
be unnecessarily stringent. The small part-to-part
variation results in tight control limits, and this in
turn results in many out-of-control dimensions.
Since small mean shifts rarely affect assembly
builds, manufacturers using control charts often
ignore the results. This is true even if larger shifts
are observed. The main concern with X-bar/
Range charts for stamping is that they do not
effectively separate problems from insignificant
process changes. One approach to desensitize
charts is to replace X-bar/ Range charts with
Individual and Moving Range charts.
Individual and Moving Range charts are based on
subgroup sizes of one. Control limits to assess
mean stability are then based on moving ranges.
Because moving range values are based on consecutive
subgroups, variation estimates reflect the
part-to-part variation and some mean shift variation.
Table 25 on page 31 presents process control
data for a stamping dimension. Using traditional Xbar
charts, this process would be considered
unstable or out-of-control as shown in Figure 15,
on page 32. Interestingly, if only the first observation
in each subgroup is measured and Individual
and Moving Range charts are used, this same
process would be deemed in control. The reason
is that Individual charts based on moving ranges
are less sensitive than X-bar charts if small mean
shifts are inherent to the process. Of course, with
individual and moving range charts, large significant
mean shifts may still be identified.
Table 24. Summary of Mean Shift Variation across Companies
# of Average % Dimensions w/ % Dimensions
Company Dimensions total Significant Mean Shift [mean shift> .5]
A-RH 329 0.12 80% 3%
A-LH 282 0.15 88% 12%
B 262 0.36 80% 51%
C 143 0.28 84% 31%
D 62 0.19 85% 3%
E 41 0.09 34% 0%
F 61 0.10 82% 3%
G 107 0.15 82% 14%
Total 1287 0.18 81% 19%
2 The control limit for an X-bar chart is equal to A2(n) x d2(n) x part-part, where A2 and d2 are functions of subgroup size. If the subgroup
size, n, is equal to 4, then the control limits are +/-0.729 x 2.059 x 0.1 or +/- 0.15mm.
30
Table 25. Process Control Data
Subgroup Sample Sample Sample Sample X-bar Range X Rm
(i) 1 2 3 4 (i) (I) (I=2) (I)
1 0.40 0.30 0.20 0.50 0.35 0.30 0.30 0.00
2 0.25 0.50 0.40 0.30 0.36 0.25 0.50 0.20
3 0.25 0.25 0.05 0.15 0.18 0.20 0.25 0.25
4 0.50 0.20 0.10 0.20 0.25 0.40 0.20 0.05
5 0.90 0.75 0.85 0.70 0.80 0.20 0.75 0.55
6 0.65 0.40 0.50 0.90 0.61 0.50 0.40 0.35
7 0.20 0.40 0.25 0.25 0.28 0.20 0.40 0.00
8 -0.10 0.10 0.25 0.20 0.11 0.35 0.10 0.30
9 0.25 0.30 0.30 0.25 0.28 0.05 0.30 0.20
10 0.40 0.25 0.10 0.20 0.24 0.30 0.25 0.05
11 0.40 0.65 0.50 0.30 0.46 0.35 0.65 0.40
12 0.30 0.25 0.20 0.25 0.25 0.10 0.25 0.40
13 0.10 0.10 0.00 0.10 0.08 0.10 0.10 0.15
14 0.40 0.30 0.70 0.50 0.48 0.40 0.30 0.20
15 0.30 0.25 0.30 0.30 0.29 0.05 0.25 0.05
16 0.35 0.60 0.50 0.40 0.46 0.25 0.60 0.35
17 0.15 0.15 -0.05 0.05 0.08 0.20 0.15 0.45
18 0.60 0.30 0.20 0.30 0.35 0.40 0.30 0.15
19 0.70 0.55 0.65 0.50 0.60 0.20 0.55 0.25
20 0.75 0.60 0.90 1.00 0.81 0.40 0.60 0.05
21 0.15 0.20 0.35 0.40 0.28 0.25 0.20 0.40
22 0.30 0.50 0.25 0.60 0.41 0.35 0.50 0.30
23 0.15 0.20 0.20 0.15 0.18 0.05 0.20 0.30
24 0.30 0.55 0.40 0.50 0.44 0.25 0.55 0.35
25 0.75 1.00 0.85 0.65 0.81 0.35 1.00 0.45
Average 0.38 0.26 0.39 0.25
31
The use of Individual and Moving Range charts for
stamping processes solves the problem of oversensitive
control charts; however, it does not necessarily
result in better process control. The fundamental
problem with statistical process control
charts for stamping is that they merely expose
mean shifts. Effective process control requires an
understanding of the robustness of dimensional
measurements to input variables and then the discipline
to control the variation within these robust
levels. For example, manufacturers should identify
safe operating windows for draw tonnage, cushion
pressure, shut height, counterbalance pressure,
air pressure, n-value, material thickness etc. They
then need to operate their processes within these
windows. If they can meet this objective, there is
little need to measure stamped parts during regular
production. However, many manufacturers
either have insufficient knowledge of the robustness
of their processes to input variables or are
not consistent in monitoring them.
Ultimately, whether a non-stable mean is acceptable
depends on the influence that the variation
will have on the assembly. In these case studies,
most assembly dimensions were robust to the variability
of their coordinated stamping dimensions.
Figure 15. X-Bar/Range Chart vs. Individuals/ Moving Range Charts
(Note: charts based on the same process data)
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
subgroup #
x bar Chart
UCL
CL
LCL
0.10
0.00
1
xbar (i)
3
5
7
9
11
13
15
17
19
21
23
25
1.2
1
0.8
0.6
0.4
0.2
0
subgroup #
Individuals Chart
UCLx
CL
LCLx
-0.2
-0.4
1
Individuals, xi
3
5
7
9
11
13
15
17
19
21
23
25
0.70
0.60
0.50
0.40
0.30
0.20
subgroup #
Range Chart
UCL
CL
LCLR
Rbar
R
0.10
0.00
1
xbar (i)
3
5
7
9
11
13
15
17
19
21
23
25
0.9
0.8
0.7
0.6
0.5
0.4
0.3
subgroup #
Moving Range Chart
UCL
CL
LCL
Rm
Rm
0.1
0.2
0
1
Moving Range
3
5
7
9
11
13
15
17
19
21
23
25
32
Table 26 below indicates that relatively few dimensions,
less than 5%, exhibited strong correlation.
Although stamping-to-assembly correlation is low,
some stamping dimensions with mean shifts
greater than 0.5 mm corresponded with assembly
dimensions demonstrating higher variation. Thus,
the elimination of large stamping mean shifts
would likely lead to a reduction in some assembly
variation.
Table 26. Effect of Stamping Mean Shifts on Assembly Variation
# Dimensions with # Dimensions with
Company # Coordinated Significant Stamping Mean Median assembly Median assembly
Dimensions Correlation Shift > 0.5 if shift < .5 if shift < .5
A 33 1 1 0.18 0.30
B 104 8 62 0.16 0.23
C 32 2 14 0.19 0.38
D 31 0 1 0.22 0.21
E 32 0 0 0.16 none
F 8 2 0 0.20 none
G 77 1 9 0.13 0.13
Totals 317 14 (4%) 87 (27%) Average=0.16 Average=0.25
Effect of Stamping Mean Shifts
33
4.0 Tolerance Considerations
4.1 Tolerances
Two objectives for assigning sheet metal tolerances
are to help insure that final assembly quality
will be met and to minimize productivity losses
during assembly because of large stamping variation.
Assigning tight tolerances help achieve this
goal. The tradeoff to assigning overly tight tolerances,
however, is that die and stamping costs
may become excessive trying to meet them. In
some respects, the tolerance has the effect of
shifting costs from stamping to assembly or from
assembly to stamping, depending on the tolerance
assigned. A reasonable and meaningful
sheet metal tolerance needs to consider the following
three factors:
&#8226; Stamping process capability:
The tolerance must reflect what a stamping
process is capable of achieving, otherwise
unnecessarily high stamping costs will accrue.
There are many current instances where
stamped parts are out of tolerance, but are being
assembled successfully. All of the benchmark
automotive manufacturers had body side outer
panels with a significant number of points out of
tolerance. This is evidence that manufacturers
tend to assign unnecessarily tight tolerances on
stamped parts, particularly for less-rigid outer
panels. When stamping plants have difficulty
meeting assigned tolerances, there is a tendency
to overlook the tolerance and wait to hear if
assembly generates build problems. This waitand-
see approach would be improved upon if
the tolerances were known to be meaningful.
&#8226; Impact on assembly:
Unlike many other rigid assembly processes, the
assembly of sheet metal affects final part geometry.
The assembly process has the ability to add
or reduce variation depending upon the components
and the assembly process. Many assembly
processes are robust to a wide range of
stamping variation showing virtually no impact
on assembly quality due to stamping variation. In
these instances, it would benefit manufacturers
to widen stamping tolerances, at least to the
point where they begin to impact assembly.
&#8226; Measurement system limitations:
Because of the impact the measurement system
has on the ability to measure stamped panels,
part tolerances need to reflect the measurement
system design. It was shown earlier that measurement
fixtures with more clamps tended to
have parts with tighter tolerances than those
measured with fixtures using fewer clamps. The
amount of observed variation with constrained
checking fixtures is less than that of less-constrained
fixtures and therefore, tighter tolerances
can be achieved.
4.2 Cp and Cpk (Pp and Ppk)
The predominant tolerance strategy used by automotive
manufacturers is to assign tolerances
which may be difficult to achieve but are believed
to help final assembly quality while reducing
assembly problems. In some cases, overly tight
tolerances are assigned; if not readily achieved,
they can be re-evaluated during development. An
advantage of this strategy is that certain parts
where all the tolerances are met are approved
without special intervention. One concern with this
strategy, however, is that many dies are unnecessarily
reworked to meet the original tolerances
even though they may not impact assembly build.
This unnecessary rework leads to delays. Since
these manufacturers often use process capability
indices to measure conformance to tolerance,
they will be discussed next.
Two process capability indices often used to compare
how well a process is achieving the design
tolerances are Cp, process capability, and Cpk,
design capability. These indices are a function of
the tolerances, part-to-part variation and mean
bias, and were developed to measure the capability
of a process relative to design intent. The formula
for Cp is:
Equation 5 Cp = USL - Nominal
3 part-part
The Cp index is determined by dividing one half of
the tolerance, where one half the tolerance equals
the upper specification limit (USL) minus the nominal,
by three standard deviations of part-to-part
variation. The formula for Cpk is:
Equation 6 Cpk = USL - Mean Bias
3 part-part
(Note: mean bias = process mean - nominal)
34
The Cpk index is determined similarly to Cp, except
that any mean bias is first subtracted from the
numerator. If there is no mean bias and the
process is operating exactly at the design nominal,
then Cp = Cpk. For the purpose of these calculations,
part-to-part is estimated using statistical
tables and the formula:
Equation 7 part-part = R
d2
If the sample standard deviation is used to estimate
part-to-part rather than the above formula, then
the Cp and Cpk indices are referred to as Pp and
Ppk. Their interpretation, however, is the same
regardless of the method used to estimate part-topart.
Figure 16 below illustrates differences in Cp
and Cpk for three different scenarios.
Figure 16. Illustration of Cp and Cpk calculations for three scenarios
Mean
Mean
Mean
nominal
tolerance
nominal
tolerance
nominal
tolerance
0.5
Tol ±1.0 ±1.0 ±1.0
0.25 0.33 0.25
Cp 1.33 1.0 1.33
Cpk 1.33 1.0 0.67
4.3 Recommended Tolerances for Sheet Metal
The tolerance guidelines shown in Table 27 on
page 36 are based on these empirical benchmark
studies. These guidelines allow consideration for
process capability, or achieving a Cp = 1.33, influence
on assembly dimensions and measurement
system limitations. They also assume that the data
is obtained without over-constrained measurement
systems. Furthermore, these tolerances only
reflect manufacturing variation about the long-term
process mean, and do not consider the ability to
hit the design nominal. Since dimensions routinely
deviate from design nominal, initial specifications
may account for both mean bias and process
variation, resulting in wider tolerances than those
shown in Table 27.
These case studies also suggest that rigid components,
typically with material gauges greater
than 1.5 mm, have greater process capability, or
smaller variation, exhibit more influence on the
assembly, and therefore warrant smaller tolerances.
Rigid components also tend to exhibit
greater repeatability from die set to die set, so both
short-term and long-term tolerances are smaller
than other components. Dimensions for non-rigid
panels are divided into two groups; mating surfaces
and non-mating surfaces. Mating surfaces
often are more critical for assembly, and thus may
have tighter tolerances than non-mating surfaces.
In all cases represented in Table 27, a tolerance
range is shown because the ability to control variation
may differ around the part. These general tolerances
are a function of the inherent sigma and
assume that a Cp of 1.67 is desired. For all three
categories, manufacturers should be able to at
least meet the high end of the tolerance guideline
based on the capability of stamping processes.
35
4.4 Part Tolerances and Functional Build
The assignment of part tolerances often hinges on
whether to allow for mean bias, or deviation from
nominal. The previous section recommended part
tolerances based on manufacturing variation without
consideration of mean bias. Since mean bias
is not considered, the Cp index may be used to
measure conformance to design, but Cpk is not
used. This development strategy relies on two
steps: minimize variation to an acceptable level
and evaluate the impact of mean bias on the
assembly to determine which points, if any, require
rework. Here, the assembly build is used to identify
dimensional shifts and not product specifications.
Several manufacturers use this functional
build strategy and the advantages include:
&#8226; Less die rework is needed because only dimensions
that adversely affect the final assembly are
identified for rework.
&#8226; Development lead-time is saved because less
die rework is required.
&#8226; Lower overall process variation is achieved both
in stamping and in assembly. Many engineers
believe that as the amount of die rework increases
from shifting many dimensions toward nominal,
the less robust the die becomes.
This functional build strategy may also help
improve process control because the final specifications
for mean bias and process variation are
determined during tryout and thus better reflect
process capability and the influence on assembly.
The consequence of not meeting the final tolerances
is better understood without waiting to hear
from assembly.
Table 27. General Recommended Tolerances for Stamped Parts Based on Process Capability
(Note: data based on measurements systems without over-constrained clamping)
Part Location of Inherent Tolerance to Achieve Cp > 1.67
Rigidity Dimension Sigma (tol > 3Cp or +/- 5sigma)
rigid dimensions
(~gauge > 1.5 mm) all .06 ~ .15 0.3 to 0.75
Mating Surface .10 ~ .20 0.5 to 1.0
non-rigid dimensions
(~gauge > 1.5 mm) Non-Mating .10 ~ .25 0.5 to 1.25
Surface
36
5.0 Conclusions and Summary
The following conclusions are based on the analysis
contains in this report and from observations
made throughout the study. Since much of the
data collection was obtained under production
conditions in a non-statistically structured manner,
the analysis is not sufficiently rigorous to establish
conclusive results in many areas. Due to the number
of product and process variables seen at a
single manufacturer, rigorous experimentation
would have severely limited the breadth of analysis.
The following general conclusions provide
insight from several manufacturers, and reflect differing
design and manufacturing strategies structured
around common operating principles of
sheet metal design, die construction and metal
forming. These conclusions also provide guidelines
to developing more rigorous research
deemed necessary at particular manufacturers
choosing to develop a more scientific approach to
stamping variation and measurement.
1. An important distinction across companies was
the type of panel measuring system used on
large, non-rigid parts like the body side and
wheelhouse outer panels. The greater the number
of clamps, the less observed variation and
mean biases were seen in the measurement
data. Constrained measurement systems had
between 16 and 22 in/out clamps, whereas the
lesser-constrained systems used from 5 to 11
clamps. The use of clamps and their location is
indicative of different dimensional validation
and process control strategies not discussed in
this report. It is important to note the difference
because of the impact on the measurement
data for large panels. Manufacturers using constrained
measurement systems also assigned
tighter tolerances to the body side. The constrained
tolerances varied from 0.3 mm to
0.50 mm, where the unconstrained tolerances
varied from 0.70 mm to 1.25 mm.
2. There are significant differences in the amount
of variation seen in larger, less-rigid parts such
as a body side outer panel versus smaller reinforcements
such as A and B pillar reinforcements.
Larger parts experience from 20% to
500% more mean bias on the average, from
0.5 mm to more than 1.0 mm for unconstrained
measured parts. The amount of mean bias
varies considerably across manufacturers
depending on many factors, including measurement
strategy, panel size and die buyoff
strategy. Large parts also demonstrate up to
twice as much variation as small parts, and the
variation is distributed across part-to-part and
mean shift variation.
3. Short-term variation is relatively small with the
95% 6-sigma less than 1.0 mm for rigid parts
and less than 2.0 mm for the body side outer,
using unconstrained measuring. If the mean
bias could be eliminated, many parts would
readily achieve a Cpk = 1.33. A significant
challenge during dimensional validation is
eliminating mean bias, particularly for large or
small complex panels.
4. Large, less-rigid panels also are more susceptible
to changes in variation due to transferring
the dies from the tryout source to the home line
and from home line tryout to future production.
In both cases, both the mean bias and the
amount of variation are likely to increase. Small,
rigid panels have smaller changes in variation
when transferring from tryout presses to the
home line. In some cases during production,
they show a decrease in mean bias from the
home line tryout. It is likely that die rework has
taken place during the production life to reduce
mean bias, and attention may have been
focused more on the rigid panels than on the
larger ones. Small, rigid panels are also less
susceptible to increased variation and mean
bias due to shipping influences than are larger
panels. Several small panels experienced from
6% to 19% of the dimensions shifting at least
0.2 mm due to shipping, whereas the wheelhouse
outer had 76% of the dimensions shift at
least 0.2 mm. The difference in the variation
increase was not as significant, where the small
panels averaged 85% of their dimensions
increasing in variation and the wheelhouse
increasing 91%.
37
5. Manufacturers with similar part design and
measurement systems, but with different levels
of total variation, usually experience varying
degrees of run-to-run mean shifts. As expected,
part-to-part variation is similar for manufacturers
with similar measuring strategies and product
designs. The two setup-related variables investigated
in this study, tonnage and cushion pressure,
showed a correlation with dimensional
mean shifts. No significant relationship could be
found between material property variation and
mean shift variation.
6. All stamping processes in this study operated
out of statistical control. Stamping processes
have inherent complexity making it difficult or
impossible in production to set up repeatedly
with a constant mean value on all panel dimensions.
For this reason, conventional X-bar and R
charts are inappropriate for process control
because they would routinely indicate that the
processes are out of control, despite the capability
to be assembled into acceptable bodies.
Some manufacturers are better than others at
minimizing mean shift variation, but all manufacturers
in this study are producing a significant
percentage of parts with dimensions outside
of their assigned tolerances. The meaningfulness
of currently assigned tolerances to
sheet metal part dimensions is suspect, particularly
for less rigid panels.
38
39
Appendix

Appendix A - Part Sketches by Company
Locating Pin
U/D & F/A
Front
Pillar
Center
Pillar Quarter
Inner
Clamps
Detail Fixture
Pin U/D
Clamps
Detail Fixture
Bodyside Outer
Wheelhouse
Outer
Center
Pillar
Quarter
Inner
Locating Pin
U/D & F/A
Front
Pillar
Figure 17. Part Sketches at Company A
Figure 18. Part Sketches at Company B
41
Bodyside Outer
Center
Pillar
Not included:
A-Pillar Upper and Lower Reinforcement
B-Pillar Upper and Lower Reinforcement
Quarter Outer
Clamps
Detail Fixture
Bodyside
Center
Pillar
Windshield Frame
Reinforcement
Locating Pin
U/D & F/A
Front Pillar
Figure 19. Part Sketches at Company C
Figure 20. Part Sketches at Company D
Roof Rail
42
Center Pillar Bodyside Inner
Reinforcement
Quarter Outer
Clamps
Detail Fixture
Bodyside Panel
Center Pillar
Assembly
Locating Pin
U/D & F/A
Pin U/D
Front Pillar
Assembly
Figure 21. Part Sketches at Company
Figure 22. Part Sketches at Company F
Roof Rail
Quarter Outer
Clamps
Detail Fixture
Bodyside Outer
43
Clamps
Detail Fixture
Bodyside
Outer
Center Pillar
Locating Pin
U/D & F/A
Front Pillar
Figure 23. Part Sketches at Company G
Quarter Inner
44
AK Steel Corporation
Bethlehem Steel Corporation
DaimlerChrysler Corporation
Dofasco Inc.
Ford Motor Company
General Motors Corporation
Ispat/Inland Inc.
LTV Steel Company
National Steel Corporation
Rouge Steel Company
Stelco Inc.
U.S. Steel Group, a Unit of USX Corporation
WCI Steel, Inc.
Weirton Steel Corporation
Auto/Steel
Partnership
This publication was prepared by:
Body Systems Analysis Project Team
The Auto/Steel Partnership Program
2000 Town Center, Suite 320
Southfield, Michigan 48075-1123
248.356.8511 fax
http://www.a-sp.org
A/SP-9030-3 0100 2M PROG
Printed in U.S.A.

Triangel

英语看起来还真是费劲儿啊。

happyromeo

好牛的帖子啊！真是考验英语水平！

huarengan

gaoxialai慢慢看

南宫霹雳

太难了，太长了啊，有附件可以下载吗

wf5577677

太难咯嘛！～～太考英语水平咯嘛！～～有没有高手能翻译一下呢！～～～

jiatang0839

LZ：翻译一下嘛，的确看不懂耶！