PSI - Issue 57

David Mellé et al. / Procedia Structural Integrity 57 (2024) 61–72

67

David Melle´ / Structural Integrity Procedia 00 (2023) 000–000

7

Fig. 6. Example of an interferometry based measurement of the gauge length of a 30 min etched coupon (upper image) and the surface porosities detected by the in-house python program (lower image).

√ area parameter, which will be designated by’ size ’ in the following, was determined by considering the area of the convex domain needed to enclose the projection of the points defining the porosity in a plane orthogonal to the loading direction. This has been done so as to be comparable to measurements on a fatigue failure surface imaging (on a SEM). Figure 7 shows for the selected samples (the ones for which SEM and interferometry measurements were roughly consistent) the size population of the detected surface features. The killer feature is designated by the red diamond marker and its size measured on SEM images is also drawn as a red triangle. This graph confirms that the √ area parameter is not su ffi cient to predict the killer-feature in a given population. This has also been observed by several other authors. The nominal stress amplitude is also included which shows that, even if the killer surface feature is not the largest one, the evolution between the nominal stress and the killer-feature size exhibits a trend (that is visible in Figure 5). The same graphical representation is generated in terms of the surface micro-geometric features maximal depth (see Figure 8). It appears to be slightly better even if not entirely conclusive. Due to the coupon geometry and bending loading mode, each surface micro-geometric feature may not have been submitted the exact same stress level. The comparison of areas or depths of the surface features may be biased by their positions on the coupon gauge surface. To be able to compare these surface features independently from their position on the gauge surface, corrective factors were determined to take into account the longitudinal stress gradient ( K longi ) and the stress at the maximal depth of the surface features ( K depth ). The previous graphs are then drawn using the population of surface features ∆ K or SCF. The formula used to compute the ∆ K value is given by Equation 2. The SCF was determined for each porosity using a FFT based methodology proposed by Kim et al. (Kim (2020)) which is faster than FEA on extended surfaces. The SCF indicator is only the SCF of a surface feature multiplied by the two correcting factors already mentionned ( K longi and K depth ). These graphs are presented on Figure 9 and Figure 10. ∆ K = 2 × σ amp × 0 . 65 × K longi × K depth × π × √ area (2) Neither of these two indicators works perfectly as the killer surface feature does not exhibit the highest value of these parameters in the population, on all coupons. While the ∆ K populations are consistent with the nominal stress levels achieved, despite the fact that they do not predict the killer feature, SCF populations do not present such a consistency (for example, in Figure 10, coupon 4 max SCF is higher than coupon 6 ones and so do the nominal