Page 188 - Biaxial Multiaxial Fatigue and Fracture

172 I? BONACUSE AND S. KALLURI

The stress range vs. cycle number plots for the higher amplitude, second load level (Fig. 2)
show some interesting behavior. The lower initial life fraction experiments continued to
cyclically harden during the second load level. The higher initial life fraction experiments
tended to cyclically soften (with exception of the axial-axial where all hardened to failure).
The material in each type of load level combination tended to converge to a similar stress range
as the cycles accumulated in the second load level, independent of the number of lower
amplitude cycles previously imposed. The axiaYaxial and torsiodtorsion experiments tended
to stabilize to the same stress levels as the baseline (constant amplitude fatigue tests performed
at the second load level) experiments (also plotted in Fig. 2), whereas the mixed loading
experiments, axial/torsion and torsionlaxial, stabilized at stress levels 6.5% and 12.0% above
the baseline experiments, respectively. The extra hardening observed in the mixed loading
experiments may be attributable to the same mechanism that causes additional hardening in
mechanically out-of-phase multiaxial loading [9].
Experiments ending with torsional loading always failed on the maximum shear strain plane,
i.e. the final failure cracks were parallel to the specimen axis. Experiments completed under
axial loading all failed at or near the plane of maximum tensile stress; perpendicular to the
specimen axis. In the tests completed with axial loading, the 10% load drop failure criteria
corresponded to an average crack length of 20.5 mm. Somewhat longer surface cracks (27.4
mm average length) occurred in the tests completed with torsional loading. This result is not
unexpected in that shear cracks (cracks that form and propagate on maximum shear
stress’strain planes) tend to be longer at failure than cracks that form and propagate on
maximum normal stress/strain planes [lo].
Five of the eighteen tests were terminated due to controller interlocks. Controller interlocks
are preprogrammed limits on: load, strain, and displacement. These limits were typically set to
approximately 10% above or below the expected maximum values of the measured variables.
An interlock can also occur when the difference between the command and feedback signal in
the control loop reaches a preprogrammed threshold, which was 15% of the commanded strain
for these experiments. A controller interlock shuts off hydraulic pressure and sends a signal to
the control software to indicate that the test has been terminated. Larger final cracks with
significant ductile tearing and specimen distortion were associated with the controller
interlocks. The length of the cracks that propagated in fatigue, as identified on the fracture
surfaces, were of the same order as those where the experiments ended at the 10% load drop.
The difference in the number of cycles to failure associated with the interlock events, as
compared to those terminated at a 10% load drop, is believed to be small.
The results of the axiaYaxial, torsion/torsion, axial/torsion, and torsion/axial experiments are
compared with the predictions of the LDR and the DCA in Fig. 3. It’s clear that neither
cumulative damage model appears to predict the observed behavior adequately. In most cases,
the LDR seems to more closely approximate the observed behavior for lower initially applied
life fractions (nl/NI 5 0.4), whereas the DCA model seems to do better with the higher (nl/N1 >
0.4) initially applied life fractions.
Figure 4 displays this cumulctive fatigue data with the applied life fraction in the first load
level on the horizontal axis and the observed sum of life fractions on the vertical axis. The
horizontal line at 1.0 depicts where the data would fall if the LDR perfectly predicted the
damage interaction. Again, the deviation from the LDR as the imposed life fraction increases
is readily apparent.
Figure 5 shows a comparison between the predicted and observed remaining cycle life, n2.
for (a) the LDR and (b) the DCA. Both models, in general, predicted fatigue life within the
expected experimental scatter-band (factors of two of the average life) for LCF. However, the