Professional Documents
Culture Documents
Validation of The Fracture Mechanics Test METHOD EGF P1-87D (ESIS P1-901ESIS P1-92)
Validation of The Fracture Mechanics Test METHOD EGF P1-87D (ESIS P1-901ESIS P1-92)
00
Printed in Great Britain. All rights reserved Copyright 0 1993 Fatigue & Fracture of
Engineering Materials & Structures Ltd
Abstract-The elastic-plastic fracture mechanics test method EGF P1-87 I): EGF Recommendations for
Determining the Fracture Resistance of Ductile Materials has been validated by an extensive round robin,
covering fracture tests and their evaluation as well as scanning electron fractography. The results
confirmed the suitability of the procedure, but suggested some modificalions. The resulting procedure
ESIS PI-90 was then further modified and its actual designation is ESIS PI-92.
The present report provides comprehensive information on the determination of crack growth resistance
curves and initiation values. Particular emphasis is on single specimen techniques and on determining the
stretch zone width. Beyond the validation of the ESIS procedure the information given is pertinent to
elastic-plastic fracture testing in general.
CONTENTS
NOMENCLATURE
I INTRODUCTION
2 OUTLINE OF ROUND ROBIN
2.1 Draft procedure P1-87D
2.2 Round Robin
3 MULTIPLE SPECIMEN METHOD
3.1 First evaluation
3.2 Conclusions from the first evaluation
3.3 Second evaluation
3.4 Conclusions from the second evaluation
4 SINGLE SPECIMEN METHODS
4.1 Introduction
4.2 Evaluation procedure
4.3 Aa-prediction capability of the methods
4.4 Resulting shape of R-curves
4.5 Analysis of specific results
1231
1232 K.-H. SCHWALBE
et a1
Acknowledgements
7 REFERENCES
NOMENCLATURE
a = crack length
a, = initial crack length
a, = estimated initial crack length (when using single specimen method)
Aa = crack growth
AaB = crack growth due to blunting
Aa,,, = crack growth at termination of test
Aamax= valid crack growth range
AaSm = critical stretch zone width determined on one specimen, representing the average of 9 Aa,,,,, values
Aaszwi = refers to Aaszw for individual specimens as determined by the participants
&is,, = Aasm for a given material, averaged over at least three specimens
Zszwl = refers to &isZwfor a given material as determined by individual participants
A U , ~=, local
~ stretch zone measurement taken at 9 locations on the fracture surface
Aa$, = refers to the Aa,,, for a given material as determined in Action 4 in Section 5
B = specimen thickness
B, = net thickness of specimen
d, = factor relating Aa, to J as defined by Eq (5.1)
d,* = factor relating AaB to J as defined by Eq (5.2)
E = Young’s modulus
F = applied load
F, = maximum fatigue load applied during fatigue cracking
J = J-integral
J, = J uncorrected for crack growth
J,,,, = J corrected for crack growth
Jg = J at upper limit of J-controlled crack growth
J, = J at initiation of stable crack growth
JiBLiR = J, as determined by the intersection of the blunting line with the crack growth resistance curve
JisZwiBL = Ji as determined by the intersection of Au,,, with the blunting line
JisZwiR= Ji as determined by the intersection of Aa,, with the crack growth resistance curve
J,,% = validity limit for J
Validation of the fracture mechanics test method 1233
Abbreviations
ACPD = alternating current potential drop technique
BL, =blunting line based on formula given in PI-87D
EL, =blunting line based on diagram given in PI-87D
CTOD = crack tip opening displacement
DCPD = direct current potential drop
PUC = partial unloading compliance
S Z W = stretch zone width
1. INTRODUCTION
In 1984 the EGF (now ESIS) Task Group on Elastic-Plastic Fracture Mechanics set up the
Working Party on Fracture Mechanics Testing Standards for developing a fracture mechanics test
method based on European experience. A unified test method was the final goal, to cover as many
aspects as possible instead of having a number of independent standards. As a first step, the
Working Party developed a procedure covering resistance against initiation and crack growth, both
in terms of J integral and crack tip opening displacement (CTOD).
The discussion within the Working Party was first published at the 6th European Conference
on Fracture[l]. The first complete draft was ready in 1987[2] and was named
An extensive round robin was then defined aimed at verifying the Procedure and which had 21
laboratories from various European countries as participants. The management and evaluation of
the round robin was jointly carried out by The Welding Institute (TWI), Cambridge (management,
analysis of multiple specimen method), GKSS-Forschungszentrum Geesthacht, Geesthacht (analy-
sis of initiation related work), and FhG-Institut fur Werkstoffmechanik, Freiburg (analysis of single
specimen methods). It was financially supported by the Community Bureau of Reference (BCR)
of the European Commission.
The first evaluation of the round robin ended up with a number of unsolved problems which
where treated in a second evaluation.
This round robin analysis identified areas where the Procedure could be improved. Some of these
require further study, while others have been incorporated into the finalised version of the
Procedure, published under the designation ESIS P1-90, followed by the updated version ESIS
P1-92 [3].
This report, which is a condensed version of the project report presented to BCR[4], gives
the important details of the work undertaken to validate the Procedure and describe the
1234 K.-H. SCHWALBE
et al.
concIusions and recommendations drawn from this work. The report is structured in the following
sections:
Outline of round robin
Multiple specimen method
Single specimen methods
Blunting, Stretch Zone Width and Initiation
Conclusions and Recommendations for ESIS P1-92.
The authors of each section are identified separately, to aid in future communications, and all
references are given at the end of the report.
I I I I I I
1'
t .-
t load line of
CT specimen
fatigue precrack
stretch zone
~ ~~~
the multiple specimen method is considered the reference method. It requires the loading of several
nominally identical specimens to different displacements. The extent of ductile crack growth, Aa,
is measured on the fracture surface, as shown in Fig. 2.1. It should be mentioned here, however,
that the outline of the measurement positions gave rise to ambiguities in the round robin because
fractured specimens can exhibit a pronounced thickness reduction as a consequence of the
prefailure deformation.
Indirect techniques for measuring crack growth can be used if the difference between crack
extension measured on the fracture surface and the predicted value is no greater than 15% or
0.1 mm, whichever is the greater. In order to provide some guidance for this, an unloading
compliance and several electrical potential methods are described in an appendix to
P1-87D. However, the user is warned that those methods may not give accurate estimates of Ji
or di.
The evaluation of the J-integral or the CTOD, 6, is done using standard formulae, see Ref. [2],
and if the resulting J-Aa or 6-Aa data points are located within the validity box shown in Fig. 2.2,
they are regarded as characterising the crack growth behaviour of the material independent of
specimen size.
A curve of the type
or
6 = F(Aa + G ) H (2.2)
is then fitted to the valid data points. The resulting curve is valid, if the data spacing requirements
illustrated by Figs. 2.3 and 2.4 are met. The abscissa value of that point where the curve leaves
the validity box is used to define four equal growth regions, each of which shall contain at least
one data point.
A further condition is required for data points in the first crack growth interval, namely only
data exhibiting ductile crack growth shall be used. This can be achieved by excluding data
J.6
Aa
CrackGrowth M
Crack Growth - M
Fig. 2.2. Validity box forming the area for valid crack Fig. 2.3. Jvalidity limits a) controlled by Aumaxb) controlled
growth resistance data. by Jm,,.
1236 K.-H. SCHWALBEet nl.
J.6
Crack Growth Aa
* -
CrackGrowth Aa
Fig. 2.4. 6 validity limits a) controlled by Aumaxb) controlled Fig. 2.5. Schematic of differently defined crack initiation
by J,,,. parameters.
from specimens exhibiting less than 0.2 mm crack growth and/or obtaining a more accurate
estimate of the blunting behaviour of the material, where relevant. The weak formulations for
blunting line determination in P 1-87D were replaced by strengthened and more accurate provisions
in the final version of the procedure, [3].
The following three fracture parameters can be determined with EGF P1-87D:
Ji/Si corresponding to the fracture resistance at initiation of ductile crack growth.
Two alternatives are offered for determining 4 and Si, Fig. 2.5:
The critical stretch zone width, Aa,,,, is measured in a scanning electron microscope (SEM),
and averaged for at least three specimens. The point on the crack growth resistance curve
corresponding to Aaszw defines initiation.
If the critical stretch zone width cannot be determined by this method then the intersection of
the blunting line with the crack growth resistance curve defines initiation. (Throughout the report,
this parameter is referred to as J i B L to avoid confusion.)
In the draft document it was believed that these ways of determining initiation would give the
same values. But, as it was demonstrated in the round robin, this is not the case.
corresponding to the fracture resistance at 0.2 mm of ductile crack growth
Jo.l,eL/SO,Z/BL
These engineering approximations to the initiation values are determined by the intersection of
the R-curve with the parallel to the blunting line offset by 0.2mm, Fig. 2.5.
Two types of specimen were tested: compact (CT) and single edge notched bend (SENB)
specimens. The specimen thicknesses were 25 mm, with 2.5 mm deep sidegrooves on each face. The
initial crack length to specimen width ratio, a,/W, was specified as 0.6 and all testing was performed
at room temperature.
As the principal testing method in the Procedure is the multiple specimen technique, participants
were supplied with sets of eight or ten specimens. Laboratories with the necessary facilities to
perform single specimen testing using unloading compliance or potential drop methods were asked
to test individual specimens for the multiple specimen R-curve also as single specimens. In this way,
100 single specimen R-curves were produced in addition to the multiple specimen curves. The
multiple and single specimen results were analysed by TWI and Fraunhofer IWM, respectively.
Most participants tested two of the materials, depending on the availability of specimens. The
table below shows the number of sets of specimens for each material.
All the specimens in any one set were machined from one area of the original plate. The specimen
numbering system used was devised to relate the specimen to its location in the plate, thus
maintaining traceability.
In order to reduce possible sources of variability due to, for example, differences in initial
a,/ W ratios or sidegroove depths, the participants were asked to follow the guidelines, summarized
here:
1. Due to the different Aa spacing requirements for J-Aa curves and CTOD-Aa curves, it might
be difficult to produce valid R-curves in terms of both J and CTOD. Therefore, in the first
place produce a valid J R-curve and then, only if possible, derive a CTOD R-curve from the
data.
2. All specimens must be sidegrooved to a depth of 2.5mm on both sides.
3. After fatigue precracking, a,/ W should be 0.6.
4. The loading span for SENB specimens should be 200mm.
5 . For the high toughness steel (M3), test to the Aamueven though the J,,, limit would be exceeded.
6 . The tests should be carried out at room temperature which should be measured and reported.
7. The following values of material properties should be used:
8. For the determination of initiation parameters using the analytical blunting line approach, the
values of d, and ( T ~(eqn 5.1) to be used are as follows:
Finally, in order to evaluate the approach to determine initiation, eight laboratories carried out
stretch zone width (SZW) measurements on their tested specimens using a scanning electron
microscope (SEM). These results were supplemented by additional SZW measurements by GKSS,
who were responsible for the analysis of this part of the study.
R-curves which were based on data with a crack growth greater than the "stretch zone width"
determined either by measurements or by one of the analytical blunting lines based on d, or d,*
(equation 5.1, 5.2).
For each of the materials, the J and CTOD data were re-analysed as complete groups. Following
initial studies of the results, some errors were found in the data. These were corrected by
participants and the re-calculated R-curves based on data with a crack growth greater than 0.2 mm
have been plotted (solid line) along with the data points in Fig. 3.1 to 3.3. The validity limits for
each of these curves have been included in these figures. For the high toughness material, M 3 , the
J,,, limit was ignored and the curve fitted to all points below the Aamaxlimit.
The coefficients for the offset power law fit to each of the groups of data are included in the
captions. It may be noted that the offsets in the offset power law are either small or zero.
From these figures it can be seen that the scatter of the data is very low. Although the scatter
of the CTOD data for M2 appears at first sight to be large, it is in fact comparable with the scatter
in the CTOD data for high and medium toughness materials once the difference in scales has been
taken into account. The difficulties in measuring very small values of CTOD should, however, be
appreciated.
The data obtained from SENB specimens are indicated in the figures by solid symbols. There
appears to be no systematic difference between the data obtained from SENB specimens and CT
500 , , , 1 , , , 1 ~ 1 , 1 , 1 ~ 1 , ,
- c T SEN8
- OLlb 3 .Lab 1
Vbb4 *Lib5
Abb8 Ahb7
E "1 :::::: 0
, c <-ff
. '*-<-K-- 0-
- Obb16
xbb19
J max p :------
(a) -
1 ;
I , 1 , 1 , , , 1 , 1 , 1 , 1 , 1 ,
0 02 0.4 0.6 0.8 1.0 12 11 16 1.8 2.0
i
l
-
l
I c T
l ~ l
SENB
~ l ~ l l l l i ~ l l l / L l CT l SEN8
.
1.4 OLab3 OGbi
10 Lab2 0 Lab1 I vLib4 *Lab5
' 0 Lab4 Lab2
1.2 - I0 Lab8
A
Lab8
Lab7
I +Lab12 .Lib18
I A Lab9
0.6 -
g-
. o " -
0.4 + 0.2
7%
(bl - I
specimens. It is also pertinent to note that despite the various degrees of experience in this type
of testing on the part of the participants, low scatter was obtained with three materials of very
different toughness behaviour.
Although the data for each particular material lie within a fairly narrow band, the shapes of the
R-curves returned by the participants for their sets of data do vary from a straight line to a power
law with or without an offset. To give a feel for the variation in these curves they have been drawn
in Fig. 3.4 for J . These curves are in reasonably good agreement over crack growth ranges of about
tT SEN8
OLab3 mbb6
ObbS *Lab13
AbbB *Lab21
.Lab 18
I I I I I , I , I I I 1 1 I I I
04 0.6 08 1.0 1.2 14 1.6 1.8 2.0
CRACK GROWTH Aa mm
CRACK GROWTH rnrn
Fig. 3.4. Offsetpower law fits determined by participants,(a)
Fig. 3.3. Offset power law fit to data for the low toughness for high toughness, (b) for medium toughness, (c) for low
material, (a) J, = 45.8(Aa 0.009)0.503 +
(b) 6, = 0.136(Aa)0.457 toughness material.
1242 K.-H. SCHWALBE
et a1
0.2 mm to 1 mm. At crack growths of greater than 1 mm, the curves tend to diverge considerably
as they do at the lower end of the crack growth range.
In order to determine whether this variability was due to the particular fitting programmes used,
the data were re-analysed set by set at TWI. The curves calculated by TWI were found to be very
similar to the returned fits. The variability in the returned fits would therefore appear to be a
function of the approach used for curve fitting and not due to differences between the laboratories.
The TWI analysis program also checked the validity of the R-curves for each data set and in many
cases the curves were invalid due to incorrect data point spacing or too few points within the
validity box. This was a particular problem with the high toughness material for which the first
two quartiles were to the left of Au = 0.2 mm and hence excluded from the curve fitting range.
Furthermore, as anticipated, it was generally not possible to obtain valid curves for both J and
CTOD from the same set of specimens.
As there was a lack of consensus on the best type of fit to use for R-curve data, it was decided
also to consider a linear fit, using the ordinary least squares approach, and a fit of the type
J = C,/(Aa). The C,/(Aa) fit had been put forward in the final report on the Nordic Round Robin
on J-R-curve testing [5].
Linear, and C , / ( A a ) fits to the J data for all three materials have been plotted in Fig. 3.5. These
figures show both the fitted lines and the original data. For the low toughness material, the linear
and C , / ( A a ) fits correspond very closely over the fitting range Au = 0.2 to Aa = Aumax.The offset
power law calculated for the same range is virtually the same as the C,/(Aa) curve (cf. Fig. 3.3(a)).
The coincidence of the linear and C J ( A a ) fits for the medium toughness materials is less close
but still quite good, with the offset power law falling between the two. With the high toughness
material, the limitations of the linear and C J ( A a ) fits start to become evident. The fixed exponent
of the C,/(Aa) fit does not allow a good description of the curvature, tending to give higher
toughness values a t low crack growth and lower values closer to Aamax. The linear fit is better in
the region of Aa,,, where it agrees well with the offset power law, but at low crack growth it cannot
model the curvature shown by the data.
Although the C , / ( A a ) fit gave the best reproducibility in the initiation parameter (see report
to BCR [4]), this type of fit is restricted by its fixed exponent. For larger specimens and greater
crack growth, the limitations of the fit would probably become more apparent.
The linear fit has the advantage over the offset power law of being easy to apply and more
statistically correct when data sets are as small as four points, however, for small crack extensions,
the model may be poor.
The offset power law would appear to model the true R-curve behaviour better, allowing a range
of curves from a power law to a straight line. The disadvantages of this type of fit is that three
parameters are being determined from as few as four points. The offset power law is a nonlinear
model which, strictly speaking, requires sophisticated fitting routines. The fitting method proposed
in the ESIS procedure is approximate and hence may be open to criticism.
V h b 4 *Lab2
OhbB a h b e
AhbS Abb7
OLablO * L i b 1 7
Fig. 3.5. Linear and C& fits to J, data for (a) high toughness, (b) medium toughness, (c) low toughness
material.
The results of the comparison showed that it was very difficult to achieve a strictly valid R-curve.
Firstly, the validity requirements for crack growth measurements were excessively rigorous for
the small crack extensions studied here. Secondly, the data point spacing requirements were
impossible to meet for the high toughness material. This led to the data spacing quartiles being
modified and in ESIS P1-92 [3] they are now based on the material’s analytical blunting line, see
Section 6 . In addition, the exclusion of points with toughness values greater than J,,, or 6,, from
the curve fitting within the valid Au range caused unnecessary difficulties in obtaining sufficient
points. This was also changed in P1-92 where the J,,, or d,,, limit is applied after the curve has
been fitted.
1244 K.-H. SCHWALBE
et al.
J= A + CAa (3.11
J = CJAa (3.2)
J = C(Aa)” (3.3)
J=A(Aa+C)D (3.4)
J =A + C ,Aa + C,(Aa)*+ C,(Aa), (3.5)
The linear fit was chosen for its simplicity as was the CJAa fit which had originally been put
forward in the Nordic J-R-curve round robin test programme [5]. The ordinary power law was
felt to be a strong contender to replace the offset power law in the ESIS Procedure as the first
evaluation of the round robin had shown the value of the offset to be zero or very small. The
third order polynomial was included as polynomial fits are allowed in the DVM procedure
002, [6].
Validation of the fracture mechanics test method 1245
A suite of five computer programs was written to produce and analyse the simulated data sets
using the different proposed regression equations. The creation of the simulated data sets was done
by pooling all the data for the material of interest and sorting them into the four spacing quartiles
required in ESIS P1-92. One point would then be selected at random from each quartile to form
a simulated data set. For data sets with more than four points, additional points would be randomly
chosen from anywhere. In accordance with P1-92, the J,,, limit was ignored at this stage. Data
sets of 4, 5, 6, 7, 10 and 20 points were studied.
For each data set size 500 simulated data sets were created. The fracture parameters, Jo,2,Jo.z,eL
and Jg were calculated from each simulated set. The J0.2,BL values were determined using the
analytical blunting line. In accordance with P1-92, the slope of the blunting line was determined
using d,*.
J, was evaluated as this was a parameter of interest to some of the participants. Although strictly
J,,, varied from data set to data set, the calculation of J,,, for each set would have been excessively
complicated. Therefore a fixed value of J,,, was used based on the nominal dimensions of the
specimens (B = 25, a / W = 0.6). These values were:
All of the simulations were done using J, data as it was assumed that the findings would be equally
valid for J,,,,, 6, and 6,,,,.
For each material, the means and standard deviations of the fracture parameter, JO,Z,BLhave been
plotted for the different data set sizes for Aumax= 1.2 mm, Fig. 3.6. For the high toughness material
the mean values are fairly insensitive to the number of points per data set, however, they do depend
on the type of fit. As severe problems were encountered with the polynomial fit for this material
1 B H"%,
9
9
a
8 a h &&n
P
A 0 0
m
rMMpar!w.**
v)
0 4 6 6 7 10 ?o
and the results from the other two lower toughness materials showed this type of fit to have
serious drawbacks, it was decided to only consider the linear, C,/Aa, power, and offset power law
fits.
Figure 3.6 also shows the standard deviations of the simulated fracture parameter. There is both
an effect of type of fit and of data set size. The linear and offset power law fits produce the greatest
scatter in the fracture parameter and the CJAa fit the lowest. The standard deviations for the
C J A a are fairly constant for data sets up to 10 points. The other fits show a reduction in scatter
from 4 to 5 to 6 points. Increasing the number of points per set to 20 greatly reduces the scatter
for all fits.
With the medium toughness material, the mean fracture parameter values are fairly constant with
data set size. All five fits give very similar mean values. The ordinary power law shows an
unexpected result in that the standard deviations rise as the data sets are increased to 10 points.
However, as the standard deviations for 7 and 10 points are similar to those from the offset power
law, perhaps the unusual feature is the low scatter for sets of 4 and 5 points. Looking at the
coefficients of the power law fits, these are fairly constant with size of data set but the exponents
do change. The spread of exponents increases with the number of points and this will give an
increasing spread in fracture parameters.
The mean fracture parameter values for the low toughness materials also shows insensitivity to
the number of points per data set. Except for the polynomial fit, the standard deviations are
reasonably constant irrespective of data set size and insensitive to type of fit.
4. I Introduction
The first goal for all participants was to produce a series of tests following the multiple speci-
men method which would result in a valid J-Aa curve and, if possible, a valid 6-Aa
curve also. Participants were allowed to test individual specimens within the multiple speci-
men series using single specimen methods. For this it was proposed to test two specimens up
to Aamax( M 1 mm, depending on specimen size) and to stop one additional test in the crack
growth range 0 < Aa < 0.3mm in order to validate the R-curves for evaluation of initiation
parameters.
Validation of the fracture mechanics test method 1247
The three different single specimen methods, used by the participants, are abbreviated in the
following by
PUC Partial Unloading Compliance
DCPD Direct Current Potential Drop
ACPD Alternating Current Potential Drop
MSU Multiple Specimen Unloading
A total of 100 single specimen R-curves for the three different materials were delivered for this
evaluation:
8 1 CT-specimen tests
19 SENB-specimen tests
56 PUC-evaluations
44 PD-evaluations (ACPD or DCPD)
Some of these tests were performed using partial unloading compliance and a potential drop
method in parallel, thus resulting in two R-curves. The main effort was done on the reference
material M3 using CT-specimens.
The R-curves were defined by sets of J,-Aa points. The number of measured points varied from
about 10 to about 60, depending on the amount of total crack extension, the material, the method
and the laboratory. This resulted in a significantly greater number of valid data points for each
R-curve than the minimum number of tests specified for multiple specimen evaluations.
4.2 Evaluation procedure
The first goal of the evaluation was to unify the data handling after the J-Aa points had been
defined by the participants. Then the data were re-evaluated as defined in the draft procedure [2]
and in addition with a number of modifications including the final specifications of P1-92[3].
Part of the evaluations done for the multiple specimen series was repeated for those single
specimen tests loaded up to a sufficient amount of crack extension. Though this discussion is limited
to J-Aa curves using Jo without crack growth correction, the results will be valid for the
qualitatively equivalent R-curves in terms of other parameters as well.
Some of the participants established their individual blunting line from SZW measurements,
others calculated the slope of the blunting line following APPENDIX 7 of the procedure.
The influence of the blunting line definition on critical J-values is beyond the scope of this
chapter. Therefore the calculated blunting line slopes following APPENDIX 7 of the procedure [3]
were used for these evaluations.
Selected “extreme” R-curves were specially investigated and partly re-evaluated based on more
complete data sets than originally returned in order to identify problems that might be specific for
the type of material
the toughness level
the experimental method for monitoring crack growth
the laboratory.
4.3 Aa-prediction capability of the methods
The most important criterion for the applicability of any single specimen method is the capability
to predict the crack extension during the test. The only direct check is the comparison of the last
estimate of the indirect method (Aa (predicted)) with the crack extension measured on the fracture
surface after the test (Aa(measured)).
1248 K.-H. et al.
SCHWALBE
The correlation of measured and predicted Aa-values is shown in Figs 4.l(a) to (c) for the three
materials. The lines mark the ideal correlation and the upper and lower limits of errors accepted
by the procedure PI-92, i.e. the greater of 0.15 mm or 15% of the measured Aa-value. Different
methods and specimen types are plotted by different symbols.
For the medium toughness material (MI, BS 4360) positive and negative errors are nearly evenly
distributed. Positive errors dominate slightly, mainly due to the DCPD-results. With only one
exception all errors are within the tolerable error band.
‘1 M 1 / BS 4360 Gr.5OE //
- (a) s
2-
CT DCPO
1- + CT ACPO
7 M 2 / A15083 -0
E
1 -
3- Ibl
4
8 -
2 2-
3
3 - CT DCPD
P CT ACPD
u+n 1-
W
a -
n.
0
0 1 2 3
M 3 / BS1501-224 Gr.4
CT DCPD
MEASURED Aa (mm)
Fig. 4.1. Correlation of measured and predicted crack growth; (a) M1 steel (b) M2 Aluminium alloy (c)
M3 steel
Validation of the fracture mechanics test method I249
For the low toughness material (M2, A1 5083) all errors are negative with only one exception.
Especially for small amounts of crack extension up to about 1 mm nearly 50% of the predicted
values are significantly smaller than the fracture surface measurements done by optical microscopy.
About 30% of the measurements are outside the tolerable error band.
In contrast to this, for the high toughness material (M3, BS 1501) the majority of the errors are
positive; negative errors appear only for crack extensions greater than 1 mm. Nearly 30% of the
measurements are outside the tolerable error band.
Some participants delivered a series of 5 to 10 single specimen tests, covering the crack growth
range from less than 0.5 mm to more than Aumax.Laboratory specific results of unloading
compliance tests for the materials M2 and M3 are plotted in Figs 4.2(a) and 4.2(b), respectively.
For the material M2 both series are underestimating the crack extension by about 0.15 mm for
the CT-specimens of laboratory 9 and by about 0.25 mm for the SENB-specimens of laboratory
6. Post examination of some fracture surfaces of M2 specimens by scanning electron microscopy
(SEM) for SZW measurements (see Section 5 ) indicated that the standard light microscope
measurements as done by the participants tend to overestimate at least small amounts of crack
extension by about 0.15 mm. Though this difference was not checked for larger crack extensions
it is likely that a correction of this order has to be applied to all the tests compared in Fig. 4.2(a).
Then the agreement of predicted and measured crack extensions would be nearly perfect for the
CT-specimens and acceptable for the SENB-specimens.
Fig. 4.2. PUC results, (a) from 2 laboratories from Fig. 4.l(b), and (b) from 3 laboratories from Fig. 4.l(c)
1250 K.-H. SCHWALBE
et al.
For the material M3 the three series in Fig. 4.2(b) differ by less than 0.15 mm. But below 1 mm
of crack extension all predictions are overestimating. Only the CT-specimen tests of laboratory 9
are underestimating beyond 1 mm. At least the first finding seems to be systematic and independent
of the specimen type. It is not known whether this is due to compliance evaluation problems for
large deformations and small crack extensions or to the optical measurement of crack extensions
as found for M2.
These results show that the prediction capabilities of the single specimen methods are sufficient
to meet the requirements of the procedure in the majority of tests.
The quality of PUC-evaluations may depend on the material toughness, but apparent inaccura-
cies may be due to the light optical measurement as applied in the multiple specimen evaluation
as well. The agreement is best for the medium toughness material. There is a tendency to
underestimate crack extension for lower toughness and to overestimate for higher toughness,
0 1.0 20
1W
E "
E
t w
40
20
0 10 20
x
x
200
0 0.5 1.0
/ I I .
1.5
, , , , ,
2.0
CRACK GROWTH Aa mm
Fig. 4.3. Final J-An points of all single specimen tests, power law regressions for these points (solid lines)
and MSU regressions (dashed lines): (a) MI, (b) M2, (c) M3.
Validation of the fracture mechanics test method 1251
respectively, as compared to the light optical measurements. If this trend is true critical J-values
derived from these R -curves may be overestimated for low toughness and underestimated for high
toughness materials as compared to multiple specimen results.
Similar checks for the potential drop evaluations are not possible, because a large number of
tests from some of the laboratories were calibrated using the measured final crack extension of each
individual test.
..
. - .. .
0 05 10 1.5 2.0
Fig. 4.4. All data points of all single specimen tests with valid ha prediction error: (a) M1, (b) M2,
(c) M3.
1252 K.-H. SCHWALBE
et al.
through the valid data points (full line) and the respective regression curves calculated for all
multiple specimen tests discussed in Section 3 (dashed line). As a basis for comparison of individual
R-curves or groups of R-curves these two multiple specimen regression curves for the respective
material are plotted together.
All R-curves were replotted and evaluated using the J-Aa data points returned by the
participants. In the first step R-curves not meeting the Aa-prediction accuracy, i.e. lying outside
the new tolerance bands in Figs 4.l(a) to (c), were excluded. All data points of all tests with a valid
Aa prediction are plotted in Figs 4.4(a) to (c). Only part of these tests having at least one data point
beyond the 3rd validity box and a valid data point distribution resulted in valid power law
regression curves shown in Figs 4.5(a) to (c), respectively.
When comparing single specimen results with the multiple specimen regression curves it should
be considered that
(i) the individual data points of the multiple specimen tests are distributed in scatter bands on
both sides of the regression curves with a width of about:
MI k 25 N/mm
M2 k 5 N/mm
M3 k 60 N/mm
(ii) points outside the regression range should not be compared with the regression curves fitted
to data within the range only
(iii) deviations of single specimen data points from regression curves do not only reflect errors
of the single specimen test method, but in addition contributions from e.g. material scatter
and errors in optical crack length measurement.
Scatter of data points
Especially in the initial region the scatter of single specimen data points is significantly larger
than the widths of scatter bands mentioned above. This is not due to tests with small amounts of
crack extension with the relatively mild validity condition for the Aa prediction. For the three
materials the highest and lowest data points of Figures 4.4(a) to (c) are close to or even identical
with those of fully valid R-curves.
Considering the scatter along the Au (instead of the J ) axis, i.e. approx. k0.1 mm, this is within
the expected experimental error of individual data points of single specimen evaluations; and of
the optical crack length measurements for multiple specimen series. Neglecting the occurrence of
negative Aa -points the resulting R -curves do not contradict the multiple specimen regressions.
Scatter of R-curves
Some of the R-curves showed deviations from the behaviour normally expected. Problems were
caused by, for example:
(i) scatter, crack growth steps forward or backward in the initial part of PUC-tests causing
problems of how to define ai (point of Aa = 0)
Validation of the fracture mechanics test method 1253
0 0.5 10 15 20
Fig. 4.5. Valid power law regressions for single specimen tests: (a) M1, (b) M2, (c) M3.
(ii) “negative crack growth” in the initial part of PUC tests and DCPD tests causing problems
of how to define Aa = O .
(iii) S-shape of the initial part of R-curves from DCPD-tests
Only part of these curves were invalidated because of unacceptable Aa prediction errors, but some
others remained valid, e.g. PD-curves calibrated with the individual final crack extension, thus not
allowing any accuracy check for an individual specimen.
An additional criterion to assess tests encountering these problems, added in the final procedure
P1-92, requires two specimens to be unloaded in the initial and intermediate range of ductile crack
extension of each series of single specimen tests. Not all of the participants met this condition, as
it was not required by the original specification of the round robin.
1254 K.-H. SCHWALBE
el al.
0
05 1.0 1.5
most conservative but normally unrealistic way. For cases of pronounced “negative crack growth”
the improvement of this definition compared to (i) is not quite clear.
A test showing negative crack growth beyond scatter and an invalid under-estimation of the final
crack growth was re-evaluated based on a mechanical model for a possible experimental effect [7].
Figure 4.7 shows the original R-curve (0-symbols) and the final correction (.-symbols) on the
reference curves. The correction of the measured compliance is based on the geometry effects
obtained from the rolling of the loading pins (8= 12.3 mm) in round holes of the specimen
(8= 12.5 mm) and of the clevis. Instead of a flat clevis surface an effective clevis diameter of 13 mm
caused by plastic indentation of the pin was assumed. This result does not prove that the model
and the assumed geometry conditions are correct, but it indicates that the effect of negative crack
growth may cause an underestimation of the final crack growth and consequently an overestimation
of derived fracture toughness parameters. As any correction needs knowledge of the actual effect
causing the negative crack growth problem, the schematical application of this option cannot be
supported.
(iii) From the apparent blunting line.
Defining a, in such a way that the early stages of the crack growth behaviour follow the apparent
blunting line seems to be the best if option (i) does not work.
In fact this option is normally used to “check by eye” whether a, is properly defined. In many
tests the exactly defined option (i) did work when the initial data points scattered around the
calculated blunting line. In other cases the positions of the first data points indicated that a, was
not properly defined by this method and the R-curve might be improved by shifting all data points
along the Aa axis.
Figure 4.8 shows one of the R-curves of Fig. 4.6. The 0-symbols are J-Aa points as
returned by the participant adjusted following option (i). The curve meets the condition of
the procedure requiring “at least five data points. . . within 0.2 mm of u,”.In fact there are ten
points in this region, and they show that a shift by about -0.06 mm might improve the definition
of a,. The valid regression through these points is nearly straight (dashed line with one point),
resulting in low fracture toughness parameters, while the shifted data (.-symbols) and the
respective regression curve (dashed line with two points) agree much better with the reference
curves.
0 0.5 10 15
CRACK GROWTH Aa mm
Fig. 4.7. Effect of negative crack growth (open symbols) and pin rolling correction (closed symbols).
FEMS 16111-E
1256 K.-H. SCHWALBE
et ai.
Fig. 4.8. Effect of shift in Au of -0.06 mm on regression result, (open symbols and curve A-original data,
closed symbols and curve B-shifted data).
It seems possible to demonstrate the sufficiency of quality by five points within 0.2 mm of ai,
but it is obvious that more points will be needed to measure the improvement of quality obtained
by option (iii) or any other more precisely defined correction procedure that may be recommended
in the future for cases with a badly defined ai. Ten points in the blunting region seem to be more
appropriate and should be recommended in order to facilitate subsequent corrections with only
little additional test effort. After more experience with the procedure a proper algorithm to check
and improve the ai definition should be appended.
Selection of regression points
Scatter in Aa in the low Aa-region may cause problems in the selection of data points for the
regression. The example R-curve in Fig. 4.9 is not valid because the final crack extension is not
CRACK GROWTH Aa mm
Fig. 4.9. Effect of point selection in first quartile of validity box on resulting regression.
Validation of the fracture mechanics test method 1257
sufficient. Nevertheless it is selected because the initial part allows several different selections of
regression points all of which could be valid according to the procedure for determination of Jo.z,
JO.ZIBL and even Ji. Here a, was defined by option (i), and this definition is supported by the following
four data points with low scatter.
With increasing J , the positions of the following points seem to be influenced by scatter in Aa.
The program originally used for producing the regression curve selected points No. 11 to 19.
The rule applied was that only points which fell within the validity boxes in the order in which
they were obtained could be included in the analysis. Thus although points 7 , 8 and 9 are themselves
valid, as point 10 is not, they were excluded. The effect of this was tested by calculating a number
of regression curves, each excluding the invalid point 10 but starting from point 7, 8, etc. The
respective regression curves are plotted in Fig. 4.9, identified by the respective starting point. It
is apparent that using starting points which sequentially follow the highest invalid point does not
influence the regression result significantly, while including the other nominally valid points
irrespective of order increases the slope.
The regressions starting from points 9 or 11 are closest to the reference curves. This alone
however, cannot validate the rule on excluding out-of-order points. Nevertheless, a procedure such
as this should be defined in order to better standardise the selection of points and to help distinguish
between scatter and real crack extension.
Material specific interpretation
M1
The R -curves are generally in good agreement with the multiple specimen regressions. Possible
corrections would even diminish the resulting scatter band of regression curves.
M2
The R-curves and the resulting regression curves (Fig. 4.5(b)) are very close to the reference curves,
but there is a trend to overestimate J or, more appropriate, to underestimate Aa in the low Aa range.
The deviations are in the range of the overestimation of crack extension by the optical
measurement, mentioned above (see also Section 5). Possibly the multiple specimen reference curves
are not better at low Aa than the single specimen results.
M3
Many of the R-curves reflected the trend shown in Figs. 4.l(c) and 4.2(b), namely underestimation
of crack extension in the blunting region (including “negative crack growth”), overestimation of crack
extension up to about 1 mm and a change back to underestimation for larger amounts of crack
extension. As a consequence the regression curves tend to be more straight than the reference curves
and deliver lower initiation toughness parameters than the multiple specimen series. The lowest J-Aa
points in Fig. 4.4(c) and the lowest regression curve in Figure 4.5(c) are due to a test not rejected
by the present criteria but showing obvious errors in the definition of ai by option (i) because of
significant scatter. Rejection of this test would diminish the scatter of the regression curves
significantly. Further investigations are necessary to clarify whether this behaviour is typical for
other high toughness materials. Improvement of the compliance evaluation, e.g. by improving the
rotation correction for the large geometry change by plastic crack opening may be needed.
Suggestions for improving the reliability of the partial unloading method
( i ) Measure more J-Aa-points in the blunting region than specified at present (e.g. 10 instead
of 5) in order to estimate the quality and improve the definition of a,.
K.-H. SCHWALBE
et al.
Exclude low load effects like crack closure from the compliance evaluation.
A better measure of the quality of the a,-definition and a procedure for improvement or
rejection is needed.
The option (ii) of the procedure seems not to be adequate to define a, if “negative crack
growth” is found in a test.
The influence of the significant geometry change of specimens on the measured compliance
when testing tough materials needs special consideration, e.g. by an improved rotation
correction.
Do not select all potentially valid J-Aa points in the first validity box. Where there are a
large quantity of data, consider including only one or two points which are in the upper
limit of the data range within the box, and which are below a Au = 0.2 mm offset. Note
that such a selection can only be made if there are a sufficient number of data points
obtained at low crack extensions.
4.5.2 Potential drop methods.
Most of the results were consistent within the respective set of curves of each participant.
Therefore differences should be mainIy caused by differences of the experimental techniques and
of the evaluation procedures.
As for the P.U.C. method, three methods for the definition of ai and the evaluation of Aa from
potential drop measurements are described in the procedure [3]:
(i) at the minimum of the potential; linear calibration between Aa = SZW/Acp = 0 and
Aaend /AVend.
(ii) at an abrupt change in slope of the potential displacement record; linear calibration as in (i)
(iii) from the definition of a reference line by a linear part of the potential vs. load curve in the
range of elastic load; calculation of Au from Acp by a given formula.
Each of the three different methods was used by some of the participants. Some of them used
two different methods for the two different materials tested. The results showed that the selection
of the evaluation method is not a free choice and does not depend only on the method (AC or
DC) or on the position of contacts. Groups of specimens tested in the same laboratory by the same
method with identical contact positions showed different behaviour; the steel specimens needing
0 70 20 30 60
APPLIED FORCE F kN
Fig. 4.10. Example of V , # vs. F curves for M3 ACPD: reference lines for minimum method (O),real
elastic slope dq5/dF ( + ) and apparent elastic slope ( x ) .
Validation of the fracture mechanics test method 1259
0 5 10 15
APPLIED FORCE F kN
Fig. 4.1 1. Example of V , 4 vs. Fcurves for M2 ACPD: real elastic slope dd/dF(dashed line) and apparent
elastic slope (full line).
YI-
5 40-
u -
Lu-
0
u 30-
B
a -
-
W
-1
P 20-
0.
a
/
/
10 - /
/
/
J ' ' I I I I
08
I I I I
12
I I I I
1.6
I I I ~
2.0
I
1260 K.-H. SCHWALBE
el al.
,
I
/ /
CRACK GROWTH Aa mm
Fig. 4.13. J-Aa curve evaluated using reference line of Fig. 4.12 ( x ) , and the curve returned by the
participant (open symbols) for comparison.
scatter of all multiple specimen data. Figures 4.5(a) to (c) show the valid non-offset power law
regressions.
The scatter band covered by the regression curves is diminished by the new regressions especially
in the low Aa range. Considering that the extremes of the regressions are due to tests that could
be improved by re-evaluations proposed above or tests that should be rejected by improved validity
checks to be specified, the single specimen results are comparable to the multiple specimen
evaluations.
4.7. Fracture toughness parameters
For all valid R-curves critical fracture toughness parameters Jo,2,Jo.z,BL and Ji were calculated.
Despite the great number of specimens included in the evaluation, the number of valid fracture
toughness parameters is too small for a statistically significant assessment of their quality related
to specimen type, experimental method or laboratory. Some general conclusions may be derived.
The broadening of the regression curve range below 0.2mm of crack growth causes a great
scatter of derived fracture toughness parameters, especially for J , , the value of the curve at a small
Aa value in the range of 0.1 mm. This effect was even more pronounced for the offset power law
regressions proposed in the draft procedure, and was one reason for the change to the non-offset
power law in the final procedure. Additional critical checks, not specified in the present procedure,
possibly causing improvements or rejection of the candidate tests, can reduce the scatter band
significantly.
As a consequence of the greater scatter of the single specimen regression curves when compared
to reference curves the scatter ranges of derived fracture toughness parameters are greater as well.
Nevertheless the mean values of the single specimen tests are comparable with the reference results
within one standard deviation.
The deviations between results of different methods are smallest for M1 (medium toughness)
while they tend to be greater for the other materials (with changing signs).
Each fracture toughness parameter derived for a reference multi-specimen series reflects one
average number for the properties of a set of specimens while the scatter band of single specimen
results shows the scatter within the set of specimens. This band reflects scatter (and possibly sys-
tematic errors) caused by the methods but in addition it contains information on the material scatter.
4.8 Conclusions
The preliminary conclusions from the first step of this evaluation gave rise to some proposed
improvements. Some of them are already included in the final procedure after additional more
specific analyses in the second step.
The results presented show that the capabilities of the single specimen methods applied to predict
Aa and to derive R-curves is sufficient to meet the requirements of the procedure in the majority
of tests. To ensure this, specific experimental problems of the single specimen methods (e.g.
“negative crack growth”, calibration for potential drop) contributing to crack growth prediction
errors and badly defined initial parts of R-curves have to be handled carefully. When comparing
single specimen results to multiple specimen series it should be considered that the Aa-prediction
error allowed by the procedure is comparable to the accuracy that can be expected for the optical
fracture surface measurement itself. Thus differences slightly greater than the allowed error do not
necessarily indicate serious deficiencies of the applied indirect methods. The errors seem to depend
on toughness systematically, but possible errors in the light optical measurement as indicated,
especially for the aluminium alloy M2, should be considered as possible sources of error for
multiple specimen series as well.
1262 K.-H. SCHWALBE
et al.
Because of the impact of such problems in the initial part of R-curves on derived fracture
parameters the procedure requires one test unloaded at a small amount of crack extension. This
ensures that the R -curves resulting from single specimen evaluations are comparable with multiple
specimen results and the scatter bands are of similar width, if R-curves with obvious problems in
the initial part are consequently excluded from further evaluation. Improved criteria for acceptance
or rejection should be developed for appending to the procedure after more experience.
The J fracture toughness parameters derived from the single specimen R-curves show more
scatter than the reference multi-specimen results, but the mean values are comparable. To some
extent this is due to the fact that each multi-specimen result reflects an average property of a set
of specimens while the single specimen results show the scatter within this set of specimens.
Some minor changes may be recommended on the basis of the present evaluation:
(i) Measure more J-Aa points in the blunting region than specified at present (e.g. 10 instead
of 5 ) in order to measure the quality and improve the definition of a, (mainly for PUC).
(ii) Exclude low load effects like crack closure from the evaluation (PUC and PD).
(iii) A better measure of quality of the ai definition and a procedure for improvement is needed
(PUC).
(iv) The option (ii) of the procedure for PUC does not seem to be adequate to define ai if
“negative crack growth” occurs.
(v) The influence of the significant geometry change of specimens on the measured compliance
when testing tough materials needs special consideration, e.g. by an improved rotation
correction (PUC).
(vi) Do not select all possibly valid J-Aa points in the first quartile of the validity box: exclude
those which are out of sequence with the first one, in order to reject points that might reflect
only scatter instead of real crack extension (PUC and PD).
(vii) The quality of the derived R-curves for the PD method close to initiation is highly
dependent on the proper definition of the reference potential or reference line. This is very
important for derived fracture toughness parameters as well. The results presented indicate
that a linearly load dependent reference line similar to method (iii) in the procedure, may
be a basis for a unified evaluation procedure, independent of the material, the method (AC
or DC) and of the contact positions (PD).
(viii) Based on the differences seen in the slopes of the reference line according to (iii) in the
PD method and of the unloading cycles, an improved reference line may be found by
recording cp vs. F in some unloading cycles in the initial elastic load range (above crack
closure) even in tests that are not intended for unloading compliance evaluation.
Stretch zone width method: the crack growth resistance curve is intersected with the critical stretch
zone width, GSzw.
Blunting line method: the R-curve is intersected with the analytical blunting line.
In the draft document it was believed that the two ways of determining initiation would give
the same values. However, it was demonstrated in the round robin that this is not the case.
Three elements are needed for determining the fracture resistance at initiation:
Determination of the analytical blunting line.
Measuring GsZw in a scanning electron microscope.
Intersecting the R-curve with either the blunting line or with Gszw.
These were checked with regard to their physical background and to their reproducible determi-
nation.
5.2 Dejinition of task
According to the above mentioned three elements the task consisted of the following items:
1. Check of the blunting line predictions.
2. Analysis of the stretch zone width data returned by the participants.
3. Additional SZW measurements on the fracture surfaces using a scanning electron microscope
(SEM).
4. Determination of the initiation parameters Ji/Si according to Fig. 2.5, using the R-curve fits
provided by the Welding Institute, and comparison of the two definitions of initiation.
5. Re-formulation of the blunting line procedure, if necessary.
6 . Re-formulation of the SZW determination, if necessary.
7. Re-formulation of the J, /hi determination, if necessary.
A particular point of interest was the expected scatter in stretch zone width data.
Items 1 and 4 required tests in the blunting and near initiation phases which were not included
in the round robin. Therefore, additional experiments were carried out, which will be referred to
in the following as “subsidiary tests”.
5.3 Blunting
5.3.1 Determination according to EGF P1-87D [2].
The slope of the blunting line in J terms is determined from either
J
Au=
~ O.4dn-
60
or
J
AaB = d,* -
E
In CTOD terms the blunting line slope is given by either
(5.3)
or
(5.4)
1264 K.-H. SCHWALBE
et al.
where d, and d,* are functions of the strain hardening exponent, n, and oolE (ao is the reference
stress); ay is the yield strength.
The factor d , is evaluated from
d,, =
rca,(l
1.185E
+ n ) ( l - v’)
2a0(1+v)(1+n)
1 Itn
(5.5)
Material M 1
Ductile tearing did not initiate in a uniform manner along the crack front. Initiation rather
occurred in the form of localised peaks as shown in Fig. 5.1. For crack growth Aa > 0.1 mm a light
microscope (LM) was used. Heat tinting helped to improve distinction of the boundary between
stable crack growth and re-fatigue fracture surfaces. From all fracture surfaces photographs were
taken along the complete crack front in order to determine crack growth by the area method: the
area covered by stable crack growth divided by the specimen’s net thickness was taken as the
average amount of crack growth. Crack growth GO.1 mm was determined in an SEM.
Material M 2
Oxidation by heat tinting is not a suitable method to determine crack growth in aluminium
specimens. Therefore, high R ratios and low loads during re-fatigueing were used to generate a
fatigue fracture surface with a very fine structure in order to obtain a well marked boundary
between stable crack growth and re-fatigue crack growth. However, a safe distinction in the light
microscope turned out to be very difficult so that all measurements were done in the SEM. At each
Fig. 5. I . Fracture surface of material MI. Dark area = fatigue pre-crack and localised stable crack growth.
Bright area = re-fatigue crack surface.
of the nine measurement locations the local amount of crack growth was determined by the area
method, in order to obtain very accurate results.
The material data needed for determining the blunting line are compiled in Table 5.3. The data
for material M3 are also included because they will be used for the determination of initiation in
Section 5.5.
5.3.3Results
Material M 1
Table 5.4 contains the measured Aa values which refer to either pure blunting or blunting plus
some ductile tearing, and the related J and 6 values. These data represent the very beginning
of the R-curve, and they are plotted in Figs. 5.2 and 5.3, together with the data points contributed
by GKSS to the round robin, and the R-curve fits through the data of all participants of the round
robin. These fits were forced through the origin, i.e. a non-offset formulation was applied.
It can be seen that the R-curve fits which are based on data Aa > 0.2 mm match perfectly the
data points below 0.2 mm.
The area method, which uses the whole crack front contour for averaging crack growth, seems
to provide a smoother slope of the R-curve than the 9 point method. The blunting line was
evaluated using d,* from Fig. A7.2 of EGF P1-87D [2] (see equations (5.2,5.4)), resulting in the
slope BL, in Figs 5.2 and 5.3, and using equation (5.5), resulting in the slightly higher slope BL,.
Looking at the diagrams, a slight preference may be given to BL,, so that it is recommended to
drop equation (5.5). It is seen:-
( i ) that the blunting line is a very good representation of the SEM data points from the blunting
phase, and
(ii) that the blunting line in its CTOD formulation (Fig. 5.3) has the same degree of accuracy
as on the J level (Fig. 5.2).
It should be noted that this is the first validation of the ESIS CTOD blunting line, whereas the
formulation on the J level has already been verified previously [18-201. It can thus be concluded
that the ESIS blunting line provides a realistic description of the blunting phase.
200
of this specimen
175
150
125
5E 100
7-
75
50
25 f S E M i
I
0 9 point average
0 areamethod
0 GKSS data points from round robin
Aa, mm
Fig. 5.2. J-R-curve of material MI close to initiation.
Validation of the fracture mechanics test method I267
0.275
0.250
0.225
+v
0.2w
0.175
Y I
0.150 , .....
E Fitted by TWI LhrOWh ! hdata of
E participants aceording m W o n 6
L 6 of the EGF-pmcadurn.
0.125
0.100
'!ME; I I
0.075
0.050
0 areamethod
0 GKSS data points from round robin
V Bluntingwntribution to average Aa
0.025
,
I -
t,aw -0.055
I
/
I
Aa. m m
Fig. 5.3. 6-R-curve of material MI close to initiation
Material M 2
Tables 5.5 and 5.6, and Figs. 5.4 and 5.5 contain the relevant data for material M2. When
comparing the data of the round robin with those from the subsidiary tests, it is immediately
obvious that there is a problem: the round robin curve does not match the data points of the
subsidiary tests. Since the generation of these data points in the SEM did not give rise to any
45
6=0.210 Aa
- BLD Curvefit of GKSS in me
40
So.ml=
0.120
35 ._
I
=32.8 /
./ I
I n
30
25 curvefit 01 GKSS in me
E range0 < M S O . 2 6 mm
E
t ................67%...... . . . 6796.
7- E
20 Lc-
I x?
I/ I
I 1
I
I I
0 0.1 0.2 0.3 0 0.1 0.2 0.3
&, rnm &a, rnm
Fig. 5.4. J-R-curve of material M2 close to initiation. Fig. 5.5. 6-R-curve of material M2 close to initiation.
ambiguities, it was suspected that the light microscope measurements carried out in the round robin
were erroneous.
In order to verify this statement, the specimen M2123D, which was investigated by GKSS in
the round robin and which then gave a light microscope value of Aa = 0.19 mm, was again looked
at in the SEM: the resulting value was 67% lower and fell now right into the scatter band of the
specimens from the second evaluation (which were only investigated in the SEM), see Figs. 5.4 and
5.5.
These other specimens were then investigated in a light microscope to reproduce the conditions
of the round robin. The light microscope revealed a dark seam at the beginning of the re-fatigue
fracture surface which had the same colour as the ductile tearing area and which could not be seen
in the SEM. This seam was interpreted as an oxidation product on the initial re-fatigue area. It
is thus concluded, that the optical investigations overestimated the amount of ductile tearing by
the width of the seam described. The SEM showed without any doubt, that the seam was not on
the ductile tearing surface. The crack orientation has no effect on the blunting phase, but ductile
tearing is significantly affected.
The findings and conclusions relevant to the blunting line are the same as those for material M 1.
A further conclusion-which is not directly related to blunting, but to the early stages of ductile
tearing-is that there is the danger of overestimating ductile tearing in aluminium alloys. It is
therefore recommended to measure the early stages of crack growth (Aa < 1 mm) in an SEM.
(5.6)
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
r
I l l l l l l l l
0.01 B
(No
machined notch
Fig. 5.6. Measurement of crack lengths on single edge notch bend specimens: a) plain sided specimens
b) sidegrooved specimens.
1270 K.-H. SCHWALBE
et al.
r----l
fatigue preuac&
-stretch zone
--stable crack growth
i=l i=k
to fatigue surface
Furthermore, the stretch zone formation along the crack front exhibits substantial scatter;
therefore, the following averaging procedure is required:
A local Aaszw,Lis determined at the nine positions used for measuring the crack length, Fig. 5.6:
1 9
=-
A ~ z w C Aaszw,L,i
9,,,
(5.7)
Each of these nine local values has to be determined by averaging at least five individual
measurements, Fig. 5.7.
5.4.2 Work performed.
The SZW data returned by the participants were analysed with respect to their scatter. As a first
step, a compilation of the data and the illustration of their variability were performed, Section 5.4.4.
Detailed examination of the marked photos returned by the participants was then carried out to
reveal the sources of scatter. In order to obtain information on a possible non-symmetry of SZW
development, the two halves of one specimen of each material were examined. Using the experience
Table 5.7. Names of participants involved in the stretch zone width investigation
~ ~~~~~
Prof. Dr. Ing. A. Bakker, Technische Hogeschool Delft; Dr. E. Roos Staatl. Material-
priifungsanstalt Stuttgart; Dr. B. K. Neale CEGB, Berkeley Nuclear Laboratories,
Berkeley; Dr. A. Cornec, GKSS-Forschungszentrum Geesthacht GmbH Geesthacht;
Dr. R. L. Jones, A R E , Poole; Dr. S . C. Druce, Harwell Laboratory, UKAEA, Harwell,
Dr. K. Wobst, BAM, Berlin; Dr. Ing. J. M. Amo, CENIM, Madrid, E. Morland, Risky
Nuclear Laboratories, United Kingdom Atomic Energy Authority, Risky
Validation of the fracture mechanics test method 1271
gained during the analysis of SZW data, a “final” bas,, for each material was derived. This work
will be described in Section 5.4.5.
The experience gained in the foregoing item was supposed to give some hints on generating more
consistent SZW data, see Section 5.6.
5.4.3 Round robin results.
The laboratories participating in the round robin did the testing using single or multiple specimen
methods according to their facilities. Nine of these laboratories (see Table 5.7) took SEM pictures
of the broken specimens and provided the respective Aaszw and baszwi values. All three materials
were evaluated this way. Table 5.8 shows the Gszwi values of the individual laboratories. From
these aszwi values, i.e. those values which were returned by the participants as their ~,,,,
the
average ZsZw was calculated which is supposed to be a characteristic value of the material under
consideration, see Table 5.8. Fig. 5.8 shows representative pictures of the stretch zone of the three
materials investigated.
with N being the number of Gszwi values used for the analysis. In addition the standard deviations
of the AasZwivalues of each participant from his own average, daszwi were determined the same
way.
Fig. 5.8. Typical stretch zone formation at crack tips for steels ( M l , M3) and aluminium alloy (M2). The
micron markers shown on each photograph are 100 microns in length.
1372
Validation of the fracture mechanics test method 1213
0.12-
R
0.10-
0.08 --
0.06.
0.04 -
0.02 -
I I
1 0.01 J
3 5
1 Participant
-251
-50 J
3 5
m
Laboratory numbel
5
I2
1 3 4 5 7 8
Laboratory number
Figs. 5.9(a), 5.10(a) and 5.1 l(a) show the individual Aaszwimeasurements (each value represent-
ing one specimen) in relation to the total average, zsZw,
for the three materials, as extracted in
Action 1. The diagrams also show the standard deviations obtained by each participant as well
as the standard deviations of the baszwivalues from the total average, by the error bars. zsZw,
The relative deviation -
- -- Aaszw * 100 (in percent)
Aas~wi
Aaszw
1274 K.-H. SCHWALBE
et al.
0.35
0.30
2 0.25
$. 0.20
2 0 I5
0.10
0.05
Participant
$
-2
-501
-100
,
CT
;
,[y(, SEN8
@I
1 2 2 4 6 7 8
Laboratory number
Fig. 5.11. Round robin results of M3 data in comparison with the f i g 1 value, together with the SZW
scatter for each participant in relation to Aa,,, .
0.15
0.20
Laboratory 1 Laboratory6 *)
0
I V
V
8
.0--
V
B
2
100
0.05
O l
b Gszw
=0.120
,100 ’ 1 2
Action
+
3
Aam
1
centre line
-
One specimen of Action 1 with +230% is out of the range.
(material MI) and the scatterof each specimen in relation (material M3; laboratories I, 4 and 6) a@ the scatter of each
to Aaszw. specimen in relation to Aa,, .
1276 K.-H. SCHWALBE
et al.
Table 5.10. Bas, measurements in mm on the same micrographs by 7 GKSS staff members.
Specimen MI44 MI54 M2113B M2113D M3-4 M3-3
Evaluator
A 0.0538 0.0599 0.0286 0.0286 0.0675 0.0815
B - 0.0527 - 0.0206 - 0.0693
C - 0.0747 - 0.0250 - -
D - 0.0642 - 0.0229 - 0.0607
E 0.0690 - 0.0332 - 0.0871 -
F 0.0923 0.1044 0.0293 0.0392 0.0856 0.1190
G 0.0768 - 0.0352 - 0.0858 -
-
All k z w 0.0730 0.0712 0.0315 0.0273 0.0815 0.0257
Evaluators
U 0.0160 0.021 2 0.0032 0.0073 0.0094 0.0257
Fig. 5.11 shows that the standard deviation of Au,,,~, i.e. the specimen to specimen variation,
of a participant may be larger than the standard deviation of dasZwi. The relative deviation
of da,,,, provided by the individual participants from the total average, Zszw, is less than
f25%.
The total averages, z,,,,
for materials MI and M2, as determined in Actions 2 and 3, are listed
in Table 5.9 together with the values from Action 1. The table shows that in five out of six cases
the standard deviations of zsZwi
from Actions 2 and 3 are smaller than those from Action 1. This
may be the consequence of the learning process resulting from analysing many more cases than
each individual participant had the opportunity to look at. Consequently, this kind of training led
V
_-
0 $ 0
0
3 V
B
0
0.05 0
OJ
1w
l v o I
IN131
a)
100
8
0-
0
50 -
B
V .
I$
-
B O -
I$
?
0 2 -50-
V 0
.looJ I )
1 2 3 1 2 3 -100
A B C D E F G
Action Action
differentevaluators
Fig. 5.14. Aaszw measurements in Actions 1, 2 and 3
(material M3; laboratories 2 and 7) andthe scatter of each Fig. 5.15. Scatter of the GKSS staff member measurements
specimen in relation to Aa,,, . on all three materials in relation to &is,, .
Validation of the fracture mechanics test method 1277
the evaluators from GKSS to more consistent definitions of beginning and end of stretch zone.
Furthermore, some of the participants stated little experience in SZW determination prior to the
round robin.
The scatter of the Aaszwivalues supports the observation from Table 5.9, Figs 5.12 and 5.13:
the specimen-to-specimen variation in Action 1 is in all cases greater than that of Actions 2 and 3.
Action 4 was then undertaken in order to obtain the variability within a control group from one
laboratory. Seven GKSS staff members measured Aaszw on identical photographs. Each of the
three materials was represented by 18 photographs, hence 54 photographs were evaluated.
Table 5.10 lists the results obtained from the individual specimens (two specimens per material).
Again, as in Actions 2 and 3, the average stretch zone widths are smaller than those from Action 1,
see Table 5.8. A graphical representation is given by Fig. 5.15, which shows the relative deviation
of each GKSS staff member from da,,, as of Action 1 for all three materials.
Table 5.1 1 compares the average stretch zone widths of all four actions and presents also values
averaged over all four actions.
If, however, a comparison is made only among the evaluators of Action 4, i.e. if each evaluator’s
values are compared with the average, zszw*,
obtained exclusively from Action 4, the result shown
in Fig. 5.16 emerges. It can be clearly seen that six out of seven evaluators agreed very closely in
their interpretation of the stretch zone. But due to evaluator F, the scatter amounts up to about
50%, which is not much different from the result of Action 1.
5.4.5 Origin of scatter.
Analysis of photos
Considerable effort was spent on examining the marked photos returned by the participants in
order to reveal the sources of scatter. Two kinds of problems could thus be identified: technical
faults and individual interpretations. Due to the limited space, they can only be briefly summarised
here. For details the reader is referred to the final BCR report [4].
Technical faults
In some cases dirt on the fracture surface obscured the details of interest.
Some photos were improperly exposed or
loo I
50 I
1
a
a
- 100
1, A
~
B C
.b,A
D
~ camre
I line
E
~
F G
I 0.0754 0.0310 0.1103
2 0.0626 - 0.0808 different evaluators
3 0.0770 - 0.0879 Fig. 5.16. Analysis of scatter of the GKSS staff member
4 0.0720 0.0294 o.0820 measurements (Action 4) in g a t i o n to newly calculated
All actions 0.0718 0.0302 0.0903 average Aa&, .
1278 K.-H. SCHWALBE
et al.
0.1 mm
Fig. 5.17. Fracture surface photo with SZW-marking of specimen M337 made by participant 6.
Interpretation
In many case details such as dimples and broken particles which belong clearly to ductile tearing,
were included in SZW measurements. Fig. 5.17 shows an example. But also the interpretation
already shown in Fig. 5.8 yields SZW values too high. These items were carefully avoided in the
determination of the final SZW values, which thus led to substantially lower SZW values, see below.
individual final values are well within *25% which is supposed to describe an acceptable scatter
band.
The term “final” SZW means that these values are the result of just the laboratory analysing
the SZW part of the round robin. There is still enough room for individual interpretation, so that
other laboratories may come to different conclusions.
5.5 Initiation
The Aa measurements generated for checking the theoretical blunting line served also for
validating the initiation procedure.
Figures 5.2. to 5.5 contain all the information required for determining initiation: in addition
to the crack growth data and the blunting line the critical SZW is shown. These are the final values
as shown in Tables 5.12 and 5.13. The diagrams-in particular those for material M1-show clearly
the gradual transition from blunting to ductile tearing. It is also seen, that the blunting process
continues to follow the theoretical blunting line after localised initiation, up to the point where the
critical SZW, Zszw, has been reached.
In addition to the two definitions of initiation outlined in Fig. 2.5, a third definition may be
formulated: it is the definition of initiation used by JSME SOO1-1981 [21], see also Ref. [I], which
intersects the blunting line with daszw.For better distinction of the two SZW methods, two
different subscripts were used in Figs. 5.2 to 5.5: JiSZWIBLand JlsZwiR.Fig. 5.18 gives a schematical
representation of these three definitions.
The constructions show clearly that the intersection of zszw with the R-curve provides the best
description of the initiation process, and it is therefore recommended to use only this method.
5.6 Conclusions
Careful re-analysis of round robin data and additional experiments in the initial phase of crack
growth including comprehensive SEM examinations have confirmed that the ESIS Procedure is
based on very reasonable assumptions concerning the physical events and their modelling. As a
result of the work described in this report, the following recommendations are derived in order to
improve the accuracy and reproducibility of the determination of initiation parameters.
Blunting line
A theoretical blunting line is needed for defining the region of valid crack growth data and for
determining J0.21BL/80.2,BL.
The blunting process is very well described by the theoretical blunting
line. It is recommended to drop the formula in A7.4 of Pl-87D [2] (equation (5.5) of this report)
to avoid ambiguities.
The 9 point averaging method with five individual measurements at each of the nine locations
is adequate.
It is sufficient to examine one specimen half for determining the SZW.
Technical faults may unnecessarily lead to erroneous results. Thus
clean the fracture surface before examining the specimen in the SEM,
0 adjust the SEM magnification such that the SZW covers most of the photo and that its
beginning and end are visible on one photo,
0 use vertical view on the fracture surface,
focus SEM properly,
expose photos properly.
It is recommended to gain experience in interpreting SEM photos and fracture features before
stating specific SZW values.
The main problem seems to be the definition of the beginning and in particular the end of the
stretch zone. When fixing the transition line between the stretch zone and the tearing surface,
dimples, broken second phase particles, and similar features should be carefully identified and
left aside. This serves to make sure that only stretching or blunting contributes to what is
measured as the stretch zone width. Examples were shown in Fig. 5.8.
Whenever possible, more than one person should be asked to carry out the SZW determination.
Examining more specimens may help to balance lack of experience.
1 2 3 4 5 6 7 8 9 hontfaCeOf
i I
-
fatigue crack
front lnnial
uadr front
Anal crack
hOnt
I
I
-
Aa sm 0.2 rnm
Aa
Fig. 6.1. Measurement of crack growth on a deformed
Fig. 5.18. Schematic of three definitions of initiation. specimen.
1282 K.-H. SCHWALBE
et al.
Determination of initiation
The R-curve fit performed for data with Aa 2 0.2 mm is adequate for describing near-initiation
data.
A light microscope for measuring low Aa values may not be adequate, as it may give rise to
mis-interpretation of the features on the fracture surface which in turn results in erroneous Au
measurements. This was found for the aluminium alloy, so that this kind of material requires the
use of an SEM measuring small amounts of crack growth (Aa < 1 mm).
Of two alternatives for determining Ji/Si, the stretch zone width method-which intersects Zszw
with the R-curve-was found to correspond well with the SEM observations. It is therefore
recommended to use this method as the only method in the ESIS procedure.
&nu
Aa
OlmlLJ
Fig. 6.2. Data spacing and curve fit. Fig. 6.3. Single specimen crack growth termination require-
ment.
Validation of the fracture mechanics test method 1283
beyond blunting are used. At least one point is required in each of these regions. The crack growth
limits, Aamax,shown in Fig. 2.2 have not been altered.
7. REFERENCES
K.-H. Schwalbe, B. K. Neale, and T. Ingham (1986) The development of a European Group on Fracture
procedure to measure elastic-plastic fracture parameters, in Fracture Control of Engineering Structures,
ECF6, ed. H. C . van Elst and A. Bakker; vol. 1, pp. 277-299.
K.-H. Schwalbe, B. K. Neale, and T. Ingham (1988) Draft EGF recommendations for determining the
fracture resistance of ductile materials: EGF Procedure EGF P1-87D, in Fatigue Fract. Eng. Mater. Strucr.
11, 409-420.
ESIS P1-92 (1992) ESIS Recommendations for determining the fracture resistance of ductile materials,
January 1992, available from ESIS office, University of Technology, Delft, PO Box 5025, 2600 GA Delft,
The Netherlands.
1284 K.-H. SCHWALBE
et al.