Professional Documents
Culture Documents
Abstract
The experimental measurement of forming limit curves (FLCs) has become common practice in the process of evaluating the formability
of sheet steel and other sheet metals. In spite of the considerable amount of time researchers have spent on the subject of FLCs, the question
regarding the accuracy and precision with which a forming limit can be determined experimentally still has not been suf®ciently analysed.
The objective of the current work is to evaluate the intrinsic precision of the experimental procedure. The results of the present work give a
reliable estimate of the (precision wise) uncertainty on the locus of an FLC. In addition, an alternative approach for interpreting the results
of a forming limit characterisation experiment is evaluated. The new approach is a ®rst attempt to capture the expected tendency of a
material to deform along different but linear strain paths for the same geometrical boundary conditions for different sheet steel grades.
# 2001 Elsevier Science B.V. All rights reserved.
Keywords: Forming limit curve; Sheet metal forming; Steel; Deep drawing; Statistical analysis
1. Introduction geometry to another. The work reported here does not treat
the accuracy of the experimental determination of FLCs.
The experimental measurement of forming limit curves The objective of the current work is to evaluate the
(FLCs) has become common practice in the process of intrinsic precision of the experimental procedure as it has
evaluating the formability of sheet steel and other sheet been used by the authors for several years. The approach
metals. Methods like those proposed by Nakazima and used for the current investigation is to submit the same
Marciniak [1] are frequently used and, to some level, always material several times to a part of the total procedure which
seem to lead to an FLC, which can be used to analyse deep is normally used to determine the FLC, thereby obtaining
drawing operations from a material's point of view. some statistical data on the precision of the procedure, and at
However, in spite of the considerable amount of time the same time excluding the material variations themselves
researchers have spent on the subject of FLCs, e.g. [2±7] and as much as possible.
references therein, today the question regarding the accuracy The results of the present work give a reliable estimate of
and precision with which a forming limit can be determined the (precision wise) uncertainty on the locus of an FLC.
experimentally still has not been suf®ciently analysed. Given the result that the observed uncertainty is rather large,
The expressions ``accuracy'' and ``precision'' are used in instead of using FLCs, the concept of forming limit bands
this paper as de®ned in Fig. 1. The accuracy of the experi- (FLBs) is put forward, which is a more reliable tool in the
mental determination of an FLC is a complex matter to evaluation of the formability of sheet steel. This concept is
analyse as it depends largely on the experimental procedure not new in itself, as it is the approach introduced originally
used to obtain the curve. This is mostly due to the fact that by Keeler, Goodwin and others (see [8] and references
FLCs are determined in technological experimental set-ups, therein).
technological in the sense that the experiments involve not In addition, an alternative approach for interpreting the
only the material but also boundary conditions like friction results of a forming limit characterisation experiment is
and differences in strain path from one experimental set-up evaluated. The argument for introducing this new method is
that when using only the locus of the FLC, one discards a
*
Corresponding author. signi®cant amount of data available from the experimental
0924-0136/01/$ ± see front matter # 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 4 - 0 1 3 6 ( 0 0 ) 0 0 8 9 0 - 6
K. Janssens et al. / Journal of Materials Processing Technology 112 (2001) 174±184 175
Fig. 3. Example of Nakazima sample (left) and the Bragard method to determine necking limit of the material.
sample and the punch. In the case of a Marciniak test, a resulting in values for the major and minor strain compo-
spacer with a hole in the middle (diameter 35 or 50 mm) is nents.
put between sample and punch. In order to obtain the points of the FLC, the method of
Bragard [9] is used. For each sample, a parabolic curve is
3.2. Measurements and interpretation of results ®tted through the values obtained for the major strain, as
illustrated in Fig. 3.
After deformation, strains are measured along a line The maximum value of the parabola returns the limiting
perpendicular to the crack or local neck. The strain level value of the major strain for local necking. The correspond-
is evaluated in ®ve squares at both sides of the crack, ing value for the minor strain component is taken from the
4. Material properties
Table 1
Steel grades used in the experiments and their mechanical properties
to the instability of the process of local necking of a where m equals 1, 2 or 3 for a confidence interval of 68,
material, i.e. the same sample width may lead to 95 and 99% on the exact location of the FLC locus.
different values for major and minor strains at necking,
but the results tend to spread more along the curve than
perpendicular to the curve. Using the described pro- 6. Results and discussion
cedure allows a better estimate of the locus of the FLC.
3. An estimate of the precision factor of the uncertainty of The raw data can be found in Table 2, and is depicted on
the locus of the curve at each data set is then constructed an FLD in Fig. 7. All data except one point for the DC04
as follows: grade has been accepted without further investigation. The
one point which is observed in the biaxial strain mode at
s of the experimental measurements is approximated by
e2 0:41 and e1 0:44 seemed abnormally high at ®rst
the standard deviation s of the data set, calculated accord-
sight. Further investigation however showed that the
ing to the method described above.
increased deviation is entirely inherent to the method of
The locus is approximated by the average of a data set.
strain measurement implemented and/or the irregularity of
This data set is limited to 10 points, thus a confidence
the rupture pattern, which is frequently observed for this
interval on this approximation should be calculated
sample width and thus cannot be considered anomalous.
using a correction factor ta,n according to the Student
Using this data and the approach described in Section 5,
t-distribution, resulting in the following formula:
the results given in Table 3 and depicted in Fig. 8 are
ta;n 1 s obtained when analysing the uncertainty on a per data set
p (1)
n 1 (10 samples) basis, i.e. the width of the FLB is varied locally
according to the z-values observed locally for each data set.
with a the confidence level, n the number of samples in a
As can be seen in Fig. 8, this leads to an increase of the
data set, and s estimated by the standard deviation s of the
uncertainty on the FLC locus from the deep drawing to the
data set.
stretching quadrant of the FLD. This is no surprise since
In addition, an uncertainty exists on the estimate of s by
samples rupture in a more irregular fashion in the stretching
the standard deviation s of the data set. For a confidence
mode than they do in the deep drawing mode. Some
level of a this is given by the following formula using the
examples of samples are shown in Figs. 9 and 10.
w-squared distribution:
v However, as only 10 samples per deformation path have
u ! been used, and the advised minimum for a statistically sound
u 1 1
t
n 1s2 (2) analysis is about 30 samples, the signi®cance of the trend
w2a=2;n 1 w21 a=2;n 1 shown is very small. In addition, when looking at other data
sets from earlier experimental work, this trend is not con-
An estimate of the global precision z on the position of the ®rmed.
FLC at the data set is then given by An alternative approach is to disregard any dependency of
v
! the uncertainty to the minor strain, and to evaluate the global
u
u 2
ta;n 1s
2
1 1 uncertainty of the FLC locus by grouping all the data sets
t 2
zm s
n 1s 2
n 1 w2a=2;n 1 w21 a=2;n 1 into a single one. This global data set, containing 60
samples, leads to a statistically much more reliable analysis.
(3) Using the procedure as described in Section 6, one obtains
Table 2
The true strain at local necking for the different sample widths for both steel grades, as determined using the Bragard method [9]
e2 e1 e2 e1 e2 e1 e2 e1 e2 e1 e2 e1
0.275 0.5 0.04 0.3 0.3 0.33 0.41 0.675 0.04 0.385 0.315 0.38
0.31 0.51 0.03 0.31 0.31 0.34 0.43 0.66 0.01 0.36 0.34 0.36
0.29 0.5 0.04 0.305 0.32 0.365 0.41 0.675 0.03 0.38 0.33 0.36
0.25 0.455 0.05 0.295 0.3 0.325 0.4 0.67 0.045 0.385 0.375 0.4
0.25 0.465 0.04 0.31 0.32 0.35 0.41 0.66 0.055 0.375 0.31 0.375
0.31 0.505 0.045 0.305 0.32 0.36 0.42 0.68 0.015 0.38 0.31 0.39
0.295 0.505 0.03 0.27 0.32 0.365 0.38 0.625 0.04 0.37 0.335 0.37
0.31 0.495 0.01 0.29 0.335 0.365 0.425 0.665 0.04 0.41 0.345 0.37
0.26 0.465 0.04 0.295 0.32 0.365 0.42 0.675 0.03 0.385 0.365 0.395
0.3 0.5 0.04 0.31 0.41 0.44 0.44 0.675 0.025 0.385 0.365 0.39
Fig. 7. Experimental results for the DC04 and DC06 grade in an FLD.
Table 3
Estimated precision as derived for each of the data sets
DC04 DC06
Fig. 8. FLBs for indicated con®dence levels, based on local data set uncertainty estimates.
180 K. Janssens et al. / Journal of Materials Processing Technology 112 (2001) 174±184
an FLB width of about 7%, true strain at a con®dence level that both grades are proven nearly equally formable in the
of 99%. It is exactly this band width which has been used in pure stretching mode.
drawing the FLBs as shown in Fig. 8. The results shown in The evolution of the width of the FLB as a function of the
Fig. 8 clearly indicate that for the given number of samples total number of samples used in the experimental procedure
per sample width, it is possible to discriminate DC04 from can be estimated from the current data using Eq. (3). The
DC06 in the deep drawing and the plane strain modes, and resulting curve is depicted in Fig. 11. A plausible conclusion
Fig. 11. The evolution of the width of the FLB as a function of the total number of samples used in the experimental procedure.
K. Janssens et al. / Journal of Materials Processing Technology 112 (2001) 174±184 181
from this ®gure is that, in order to limit the uncertainty to an higher r values promote deep drawing as a deformation
acceptable level, at least 20 samples should be included in mode.
the experimental procedure. Furthermore, one can also see The question is whether such information can be extracted
that the uncertainty would not decrease substantially when from a standard Nakazima or Marciniak experiment. To ®nd
one would increase the number of samples to even higher out whether this is the case, the data collected from a forming
volumes. limit experiment has to be analysed taking into account the
sample width. All the strain points measured on all of the 60
samples used in the analysis reported as in the former sec-
7. New interpretation approach of FLC experiments tions are depicted in the diagram shown in Fig. 12. Assum-
ing that the strain path is linear, every point reveals the strain
The results presented in Section 6 prove that one has to path for a speci®c location on the sample: it is the vector
perform a substantial amount of experiments in order to starting at the origin of the FLD pointing at the data point.
discriminate the FLBs of improved deep drawing steel To validate the assumption of linearity, the experiment has
grades like DC06 and standard deep drawing grades like been evaluated at different punch depths, the result of which
DC04, even when the differences in mechanical parameters can be found in Fig. 13. This ®gure shows all the data points
are signi®cant as for the materials used in this work. Con- for a 40 mm wide sample, measured at 20, 40, 60, 80 and
sidering the effort that is put in FLB characterisation, the 100% punch depth (100% is the punch position at sample
question arises whether one can get more out of the raw data failure). The chart clearly illustrates that no deviation from
than with the current procedure. linearity can be observed. Similar experiments have been
An alternative way of looking at the raw data, is to performed for other sample widths, and none of them
evaluate the direction of the strain path as a function of revealed any deviation from the linearity assumption.
the sample width. In this context it is frequently suggested One way of looking at the data in Fig. 12 is to determine
that for different materials, not only the locus of the FLC the strain path direction as a function of the steel quality. It is
moves, but also the strain distribution in a formed part quite evident from looking at the strain clouds that no
is modi®ed. This is correct, as for example an increase in difference in strain path direction can be observed. For all
work hardening rate n of the material leads to a more three sample widths, the direction is the same within a few
homogeneous strain distribution and thus a lowering of degrees. This is not as one may expect, as higher r and thus
the strain peaks implying a decrease of potential problems higher q
ewidth =elength e2 =e1 implies that under
in a part. Also the Lankford parameter r is expected to have tensile test conditions, a more negative minor strain is
an in¯uence on the strain distribution, i.e. higher r values preferred to an increase of the major strain, thereby promot-
allow deeper drawing, from which one could derive that ing the deep drawing mode in the material.
Fig. 12. Strain clouds of DC04 and DC06 for sample widths 60, 180 and 300 mm.
182 K. Janssens et al. / Journal of Materials Processing Technology 112 (2001) 174±184
As this observation is not as expected, an extended Although the data presented in Fig. 14 show a faint
number of mild steels have been analysed to see whether difference in the strain path direction of DC04 and DC06
the strain path direction is varying at all. In order to be able grade steels, the difference is too small in comparison to the
to manage the large amount of data involved, the strain path scatter on the data to draw any decisive conclusions, i.e. one
direction is evaluated based on the necking limit points as is not able to discern a rotation of the strain path direction.
determined with the Bragard method (described in detail in This observation is con®rmed in all other data sets we have
Section 4). This approach has been thoroughly evaluated and available, concerning a variety of steel grades (including
there is no indication of systematic differences, i.e. when bake hardening, structural, high strength and stainless steel
evaluating the strain path direction either with the strain grades).
cloud regression approach or with the Bragard points, the A separate experiment was performed, comparing two
same strain path direction is found. steel grades, which have exactly the same sheet thickness but
Using the Bragard points, a chart is constructed as shown have very different mechanical properties (mild steel DC06
in Fig. 14 and micro-alloyed HSS, ZStE 420), as listed in Table 4. The
results, depicted in Fig. 15, show that a difference in strain
The chart shows the angle between the strain path direc-
path is observed in the deep drawing mode, where the strain
tion and the vertical axis in the FLD as a function of
path turns about 108 more into deep drawing mode. In plane
sample width. Positive angles are defined as clockwise
strain and stretching mode, there is no difference in strain
and drawn against the Y-axis on the right of the chart in
path direction.
Fig. 14.
Based on the above ®ndings, one can conclude:
The S-shaped line in the chart is a reference representing
the mean strain path direction for a DC06 grade of Mild steels with approximately the same thickness (ran-
thickness 0.8 mm. This ``strain path direction reference'' ging from DC04 to DC06, with 1:5 < r90 < 2:3) seem to
has been constructed based on more complete forming show the same strain path direction, independent of their
limit experiments, totalling about 40 samples with dif- r-value for all deformation modes.
ferent sample widths ( extended test). Plotted along the A micro-alloyed high strength steel with r90 0:9 does
Y-axis on the left of the chart is the deviation of the current have a strain path direction different from mild steels, but
experiment to this reference line. only in deep drawing mode.
K. Janssens et al. / Journal of Materials Processing Technology 112 (2001) 174±184 183
Fig. 14. Strain path direction chart for DC04 and DC06 grades.
Table 4
Steel grades used in the additional experiment and their mechanical properties
These observations suggest that there might be a transi- seems to ``saturate'' to a constant level. Further research
tion in r-value, below which the deep drawing strain on this hypothesis is necessary. It also should take into
path direction is varying with r, and in deep drawing account the in¯uence of sheet thickness. Experimental
mode only. Above this transition value, strain path direction results not presented in detail in this paper suggest that
Fig. 15. Strain path directions observed for steel grades DC06 and ZStE 420, both having a sheet thickness of 0.91 mm. A small deviation is observed in the
deep drawing mode strain path direction.
184 K. Janssens et al. / Journal of Materials Processing Technology 112 (2001) 174±184
the dependence of the strain path on the sheet thickness is stamped in a single step, points of strain below the FLB are
non-linear. 99% safe, points in the FLB itself are potentially proble-
matic. Evidently, one can still judge the robustness of the
process based on the distance between the strain points
8. Conclusions observed and the lower limit of the FLB.