PSI - Issue 75

Marco Bonato et al. / Procedia Structural Integrity 75 (2025) 677–690

688

12

/ Structural Integrity Procedia (2025)

4rd Iteration:

Figure 10: Correlation results of the 4th iteration of FEA simulations (blue curve for measurements, black for simulations).

4. Analysis and Conclusion The statistical analysis of the measured and simulated time-to-failure is used to assess the level of correlation of the stochastic simulations. The criteria to assess the level of correlation are the same as in [Halfpenny (2019)]. If the two populations under comparison (experimental vs FEA data) lie within the same 90% confidence interval, the two populations show a good level of correlation. The table below shows a comparison of the extrapolated statistics for all populations, based on the parameters of the Weibull analysis, the time-to-failure for 50% of the population (R50), and the B10 (i.e. the time corresponding to a probability of failure of 90%, a typical requested value for automotive components). A preliminary analysis focuses on the experimental data, in particular on the shape parameter, the Weibull beta. Its value is b = 7.3 which corresponds to the coefficient of variation of 16%. This value is higher than expected for real automotive components undergoing vibration fatigue failure. Previous investigations had shown that the typical variability is 40-50% , equivalent to beta values between 2 and 3. In this present case study, the small variability of the experimental results is probably to be associated with the high amplitude of the input signal. Indeed, this vibration test is considered very severe. On the other hand the selected signal was considered to obtain the average time to failure after 1 hour of testing, due to limitations in the use of the testing facilities. Moreover, also the excellent level of control of the experimental set up (no variability associated with the fixture, or the mounting of the prototypes) can explain why the scatter of the vibration time to failure is narrow. Finally, as aforementioned, the geometrical variability of the specimens were non-significant, therefore reducing the variability of the dynamical response and the time of the crack initiation.