Documente Academic
Documente Profesional
Documente Cultură
Fatigue Behavior
Fatigue life was analyzed and a correlation was made between ultrasonic
inspection data and fatigue life distribution
BY H. D. SOLOMON
ABSTRACT. This paper describes the sta- fect. Since the actual product joints are [Refs. 3–9]) in fatigue data can be signif-
tistical distribution of brazed Cu joints fa- not machined after brazing, the tested icant even when considering carefully
tigue lives. It describes an approach for specimens were also not machined after machined laboratory fatigue specimens
setting requirements for braze coverage brazing. This left any small geometrical that do not contain joints. The situation is
and the determination of the resulting discontinuities (no large ones were even more complex when joints are pre-
statistical distribution of the fatigue lives. noted) as part of the specimen variation sent. The defects produced in joints and
Such an approach should greatly im- that led to the variation in the fatigue life. statistical distribution of these defects fur-
prove the reliability of brazed structures. Here, the focus was on the degree of cov- ther increase the scatter in the fatigue life
In this study, each joint was inspected erage and how this influenced the fatigue over that observed in specimens that do
ultrasonically and the results of these in- life. This data was then used to set a cov- not contain a joint. Thus, there generally
spections correlated with the fatigue be - erage specification. is a very large distribution in the expected
havior. Several transducers were used for All too often, a structure is designed fatigue lives of joints, which must be ac-
this ultrasonic measurement. Both the av- without properly accounting for the de- counted for when predicting the fatigue
erage of all transducer measurements grading influence of any brazed joints on behavior. This study describes one way to
and the single largest measurement were the design life of that structure. If account do this.
used to set joint quality standards. Limits is taken of the presence of the brazed Neither the braze process develop-
of 10% were set on the average reflection joint, it is often insufficient to guarantee ment nor the ultrasonic inspection sys-
of all transducers (corresponding to an a high degree of reliability. This paper tem was a specific subject of this study.
average incomplete brazing of the joint), shows a statistical approach for deter- Rather, the quality of the joints, as de-
and 13% for the maximum reflection of mining the fatigue life as a function of fined by an ultrasonic inspection, and
any transducer. Even with these stringent braze coverage. It describes the results of how this quality impacted the statistical
standards, the average projected fatigue a study performed to answer two simple, distribution of the fatigue lives were of
life had to be about 109 cycles to exceed but critical, questions: What is the statis- importance. Some aspects of the statisti -
a 104 cycle design life with a high degree tical variation of the fatigue life, and what cal evaluation of fatigue data are in-
of reliability. This observation points up is the acceptable ultrasonic reflection cluded as an appendix to this paper.
the high degree of scatter in the fatigue level required to statistically meet a re-
behavior of brazed joints and the need to quired design life? The answer to the first Experimental Procedures
overdesign them so that the fatigue life of question required a statistical analysis of
the poorest joint will meet the design life. fatigue behavior, which is described The joints tested in this study were
here. The answer to the second question made under the same conditions as ac-
Introduction required a correlation of ultrasonic in- tual product joints, by the same individ-
spection data with the distribution of fa- uals, using the same equipment. Propri-
It is well known that brazed joints tigue lives. This is also described. etary considerations prevent a more
generally have a lower fatigue life than The scatter (statistical distribution complete discussion of the joint process
the metals being joined (Refs. 1, 2). This development, but this is not an impedi-
degradation, relative to the properties of ment as we are concerned solely with the
the base and braze metals, stems from statistics of joint quality and the resulting
defects that develop in the braze. There fatigue life.
are three general types of defects: 1) a Two types of joints were made, tested
lack of coverage due to the largely un- KEY WORDS and analyzed. Some specimens, denoted
avoidable small voids that develop dur- as being large specimens, had a nominal
ing brazing, 2) more gross defects due to Brazing cross section of 1.8 in. wide (45.72 mm)
larger lack of coverage regions and 3) Brazed Joint by 0.375 in. thick (9.525 mm). A smaller,
geometrical discontinuities such as mis- Fatigue Life medium-sized specimen was also tested.
alignment, undercuts and crowns. Both Joint Clearance This specimen had a cross section 1.2 in.
of the first two types of defects were pre- Statistical Distribution wide (30.48 mm) by 0.275 in. (6.985
sent in this study. This study did not Ultrasonic Reflection mm) thick. In both cases, the joints were
specifically deal with the third type of de- Lap Joint made by induction butt joint brazing two
Cu bars together. The joint area was cal-
H. D. SOLOMON is with General Electric culated from the nominal area of the
R&D Center, Schenectady, N.Y. joint. All of the joints were ultrasonic
A C
Fig. 2 — Fracture surface for a typical joint. A — Low magnification of the fracture surface; B — higher magnification showing the voids in the braze
that were produced during brazing; C — higher magnification showing the voids produced during brazing and those smaller ones (on the ridges
between the large voids) produced by the ductile failure of the braze; D — ductile failure in the Cu.
Fig. 6 — Fatigue life vs. the average reflection (LOB) for specimens Fig. 7 — Fatigue life vs. the average reflection (LOB) for specimens cy-
cycled ±8500 lb/in.2 (±59 MPa). cled ±9750 lb/in.2 (±67 MPa).
phased nature of the braze, and at the fine an acceptable braze. This assess- braze. The service stress was assumed to
braze/base metal interface (due to elastic ment was made based on the correlation be ±5500 lb/in. 2 (±38 MPa). (The actual
property differences). It was beyond the of fatigue life distribution with this re- stresses being applied were more com-
scope of this study to define the source of flection data. plicated, involving a range of stresses and
the reflections in each specimen. Rather, One of this study’s goals was to pre- mean stresses, but this value is appropri-
reflection information was used as a met- dict the component life as a function of ate for the analysis being illustrated.) Un-
ric that a fabricator could employ to de- the ultrasonically measured lack of fortunately, this small a stress level re-
sulted in a fatigue life that was too long where Nf is the average fatigue life for the late this many cycles. Extrapolation was
to allow testing in a reasonable amount applied stress amplitude, σ. This equa- thus the only way to estimate the fatigue
of time. It was, therefore, necessary to test tion was used to extrapolate the data life at this stress amplitude. It should be
at higher stress levels and then to extrap- down to ±5500 lb/in.2 (±36 MPa). The kept in mind, however, the value of 3 x
olate this data to the lower stresses of in- analysis was also performed without the 109 cycles is the average fatigue life, but
terest. (This extrapolation involves cer- two highest stress points, where only sin- from a reliability standpoint, what is crit-
tain inherent assumptions that will be gle tests were run (i.e., the results at ical is the time for the first failures. Thus,
discussed shortly.) Figure 5 shows fatigue ±14,000 lb/in.2 and ±17,000 lb/in.2) with what is critical is the statistics of the fail-
life as a function of stress amplitude for no significant change in the life at ±5500 ures. This statistical correlation will be
fully reversed R = –1 cycling. A series of lb/in.2. considered after the correlation of the re-
tests was run at different stress ampli- There is, however, generally some flection data and the fatigue life is dis-
tudes, with numerous specimens being curvature in an S-N curve (Ref. 10), but cussed.
run at three of these amplitudes. The data none was observed here. Furthermore, it Figures 6-8 show the average trans-
of this figure was obtained with speci- was assumed there was no curvature be- ducer reflection for the joints as a func-
mens that were deemed to be of highest tween the lowest stress employed in test- tion of the fatigue life. As can be seen, the
quality, i.e., the average reflection was ing and ±5500 lb/in. 2. The extrapolation greater the average reflection (denoting
≤10% and no single transducer channel down to ±5500 lb/in.2, therefore, pro- more defects in the joint), the lower the
had a reflection (LOB) that was >15%. duces a conservative approximation, as fatigue life. The net section stress model
This data was used to develop a stress- any curvature in the S-N curve would re- that correlates this data utilizes Equation
fatigue life plot that was used to extrapo- sult in a longer fatigue life than what is 1 and the assumption that the average re-
late to the lower stress levels. predicted when neglecting this effect. flection correlates with the load bearing
The curve fit was developed from the The 99% confidence band for the corre- cross section, which equals the nominal
individual data points obtained at lation with Equation 1 is shown on the cross-sectional area times (1 - average re-
±14000 lb/in.2 (±97 MPa) and ±17000 figure. The figure also lists the value of the flection). For instance, in a sample with a
lb/in.2 (±112 MPa), and the averages of fatigue lives obtained at different stress 0.25 average reflection, it is assumed the
numerous tests run at lower stress ampli- amplitudes. As can be seen, cycling at cross-sectional area that can support a
tudes. The least squares fit gave the fol- ±5500 lb/in. 2 (±36 MPa) would yield an load is only (1–0.25) = 0.75 of the nom-
lowing: average fatigue life of 3 x 109 cycles. At inal cross-sectional area, so the stress is
20 Hz it would take more than 41,000 (1/0.75) times the nominal stress. This
Nf = 10^(61.88-14.01Logσ) (1) hours (more than 4.5 years) to accumu- very simple model appears to do a rea-
Fig. 10 — Cumulative probability for specimens tested at ±9750 lb/in.2 Fig. 11 — Cumulative probability for specimens tested at ±9750 lb/in.2
(±67 MPa). (±67MPa).
sonably good job of explaining most, but ations of 1–2 orders of magnitude in fa- reflection of 18% and maximum reflec-
not all, of the variation in the fatigue life. tigue life with respect to the model’s tion of 26% is utilized, the fatigue life is
All but three of the data points in Figs. trend line. Thus, even with knowledge of predicted to be about 2000 cycles. This
6–8 were within about two orders of the LOB for each joint, there can still be means that at 2000 cycles, it would be
magnitude of their expected value. There a considerable uncertainty as to the fa- expected that 0.001% of the specimens
was only one data point in all those tigue life of an individual joint. It is thus or parts experiencing ±9750 lb/in.2
shown in Figs. 6–8 that was off the model necessary to consider the statistical vari- (±67MPa) would have failed. If a criteria
line by as much as about three orders of ation of the fatigue lives in order to de- of an average reflection of 10% and max-
magnitude. This was the point in Fig. 8 termine the predicted fatigue behavior. imum reflection of 12% are utilized, then
with an average reflection of 8%, but a This is done in the cumulative probabil- it would be expected it would require
life of only 200 cycles. The fracture sur- ity plots of Figs. 9–12. more than 300,000 cycles for 0.001% of
face of this specimen is shown in Fig. 3. Figures 9–12 show the cumulative the samples or parts to fail. A similar
While the average of all transducers was probability curves (discussed in the ap- analysis can be made for each of the
only 8%, there was a large local un- pendix) are altered by considering data stress ranges utilized, for the different cri-
brazed region (with an individual chan- sets that use only the results obtained teria utilized and for the different cumu -
nel reflection of 19%) that caused the fa- from specimens exhibiting a restricted lative probabilities for failure.
tigue life to be much less than what was range of reflections (limited LOB al- The data of Figs. 9–12 are not all log
expected. It is thus necessary to consider lowed). The effect of restricting the al- normal, i.e., there are considerable devi-
the reflections of each transducer (which lowable reflections is to decrease the ations from linear behavior in many of
defines the LOB at a specific area) as well standard deviation of the data set. This the figures. This difficulty can be over-
as the average of all of the transducers. causes the curves to be steeper. The av- come, in part at least, by considering a
Hence, quality limits were placed on the erage fatigue life is not necessarily in- Weibull distribution function. This will
maximum allowable reflection of an in- creased by an appreciable amount, so be done shortly, but first we shall con-
dividual transducer, as well as on the av- the probability plots can cross. The be- sider the behavior at the design stress of
erage of all of the transducers. This re- havior when an extrapolation is made to ±5500 lb/in.2 (±38 MPa). To do this, it is
flects the fact that lack of brazing a low probability of failure is critical, and necessary to translate all the data to
distribution is important, as well as the here, the more restricted the data set, the ±5500 lb/in. 2, which is done in Fig. 13.
average LOB. longer the fatigue life. For instance, let us This translation was accomplished by
Figures 6–8 show that while the consider the data of Fig. 10. If a cumula- using the data of Fig. 5, which was used
model explains much of the systematic tive probability of failure of only 0.001% to develop Equation 1. The fatigue life of
variation of fatigue life, there can be vari- is desired, then, if a criteria of an average each data point was corrected using
Fig. 12 — Cumulative probability for specimens tested at ±11,000 Fig. 13 — Cumulative probability for extrapolated to ±5500 lb/in.2 (±38
lb/in.2 (±76 MPa). MPa).
Equation 1. This was accomplished by 99% confidence bands are also shown. a maximum reflection of any channel of
multiplying the fatigue life of each point The data of Fig. 13 does not appear to be 13% is used, then the mean life at W =
by the ratio of the fatigue life at 5500 very log normal, i.e., the data deviates –12.72 is greater than 1.5 x 105 cycles.
lb/in.2 as predicted by Equation 1, by the considerably from a straight line. This de- (This W value corresponds to a probabil-
value Equation 1 predicts for the stress of viation does not correlate with the indi- ity of failure of 3 x 10–6 and is used to de-
that test. For instance, the fatigue life of vidual data sets. This is to say, the data for fine “six sigma” reliability.) At this W
each test run at ±8500 lb/in. 2 (±59 MPa) each set are interspersed and do not layer value, there is a 99% probability of hav-
is corrected by 3 x 109/6.74 x 106 = 445. as individual groups. The lack of normal ing a fatigue life of greater than 6.5 x 104
The tests run at the higher stresses are behavior has prompted the use of the cycles (based on the lower limit of the
corrected by larger amounts. Inherent in Weibull parameter to correlate this data. 99% probability curve instead of the
this approach is the assumption the dis- This is done in Figs. 14 and 15. mean curve). This compares to a design
tribution is independent of the stress The data of Fig. 14 are fit reasonably life of 104 cycles. It is clear, therefore, that
range (i.e., the standard deviation, or well by a straight line when the Weibull even if this very high degree of reliability
Weibull shape parameter, is independent parameter function is plotted vs. the log is desired, the inspection criteria can be
of the stress level used in the test group). of the fatigue life. The Weibull slopes of set to allow reflections of somewhat
While not a bad assumption for this data, 0.70 and 1.24 illustrate this data does not greater than an average of 10% and max-
this is not always true (Ref. 9), as the stan- conform to a log normal distribution (the imum of 13%, but an inspection should
dard deviation tends to increase as the distribution is very close to being normal not allow a maximum average reflection
stress level decreases. When this is the when the shape factor is 3.57). Note, as of 26%, or maximum reflection of 39%.
case, a maximum likelihood approach is shown in the appendix, the Weibull Graphs of the sort shown in Fig. 14 can
should be used (Ref. 9). slope is determined from a natural Ln be used to trade-off the reliability vs. the
Figure 13 shows the data correlated plot, while the data has been shown on a acceptance criteria.
for a maximum average reflection of 25% common log plot so that the fatigue lives While the use of the Weibull, parame-
and maximum reflection of 39%, and for can be read directly. The resulting curve ter, W (discussed in the appendix), allows
a maximum average reflection of 10% fit slope must be corrected by 2.3, or the standard graphing packages to be used, it
and a maximum reflection of 13%. These curve fitting should be done on a Ln plot. is not the most accurate approach. A bet-
criteria encompass all the criteria used Figure 14 shows when a criteria of a ter approach is to use a program that can
for the data that makes up this figure. The maximum average reflection of 10% and calculate a Weibull distribution function
Conclusions Acknowledgments