PSI - Issue 76

Robin Motte et al. / Procedia Structural Integrity 76 (2026) 74–81

79

Table 3. S-N curve parameters determined from OLSLR analysis. Dataset m

R 2

log( C )

CONV

-2.03 -2.79

10.14 11.95 46.82 71.24 32.40

0.823 0.902 0.833 0.919 0.657

CMT

S355J2 (Ulewicz et al. (2014)) EMK8 X (Motte et al. (2024)) EMK8 Z (Motte et al. (2024))

-15.78 -25.02 -10.46

Fig. 4. Determining the moment of crack initiation based on the thermoelastic amplitude T 1 for specimen CMT-1.

3.2. Crack initiation

As described in Section 2.2, detecting a decrease in the thermoelastic amplitude T 1 can serve to determine the moment of crack initiation. A Fast Fourier Transform was applied to the experimentally measured temperature signal in a region of 3 × 3 (pixels)² (approximately 0.606 × 0.606 mm²) around the location of failure, which allowed identifying the thermoelastic amplitude T 1 at the fatigue loading frequency. Figure 4 shows the evolution of T 1 over the course of a fatigue test. To identify the onset of the decreasing phase, a threshold based on the measurement noise was implemented. Based on measurements of a specimen not subjected to a fatigue load, a noise level of T 1 was determined as 0.0014 K. Starting from the first two points on the curve, the average of points on the T 1 curve was incrementally calculated. From this average, 0.0014 K was subtracted to obtain a threshold value. If the value for T 1 of the next point fell below the threshold, it was considered as the moment of crack initiation. Otherwise, the point was incorporated into an updated average, and a new threshold was calculated. This process was repeated iteratively until a drop below the threshold was observed, as illustrated in Figure 4. The result of this analysis is summarised in Table 2 for each specimen, indicating that between approximately 6% and 50% of the total fatigue life was spent in the crack initiation stage. Neither a relation between the applied stress range and the cycles till crack initiation, nor clear di ff erences between the CMT and CONV series of specimens can be observed. From Table 2, outliers seem to be present for specimens CONV-2 and CMT-3 with longer and shorter crack initiation stages, respectively. However, the limited number of tested specimens does not allow evaluating whether these specimens are also outliers in terms of total fatigue life, or whether they fall within a normal scatter band.

3.3. Crack growth

The crack length was determined by detecting a discontinuity in the horizontal displacement field U obtained from the DIC measurements, as schematically illustrated in Figure 5 for data obtained near the end of a fatigue test. By comparing the di ff erence in the values for U (Figure 5(a)) of two adjacent subsets (denoted as U di f f , see Figure 5(b)) with a threshold value ( U di f f , threshold ), a binary image was constructed (Figure 5(c)) to indicate the presence of the crack in a certain subset. The crack tip was located at the highest row where U di f f > U di f f , threshold and the crack length