Issue 72

X. Cao et alii, Frattura ed Integrità Strutturale, 72 (2025) 162-178; DOI: 10.3221/IGF-ESIS.72.12

Miner model ME (%)

Ye model ME (%)

Peng model ME (%)

KELM model ME (%)

SVM model ME (%)

RF model ME (%)

BP model ME (%)

Materials Butt joint

37.26 41.49 68.95 158.7 93.81

35.16 37.47 63.26

10.13

14.47 25.61 16.31

10.47 20.63 19.38

12.74 24.81 24.14

11.47 23.09 22.18

Corner joint Al-2024-T42 Al-7070-T7451

8.34

11.86 62.41 31.23

157.88

193.47

218.02

226.98

195.28

Ti-6Al-4V

82.79

20.11

18.49

23.99

20.82

Table 4: ME values of machine learning models with unaugmented data on the testing set.

Miner model ME (%)

Ye model ME (%)

Peng model ME (%)

KELM model ME (%)

SVM model ME (%)

RF model ME (%)

BP model ME (%)

Materials Butt joint

37.26 41.49 68.95 158.7 93.81

35.16 37.47 63.26

10.13

6.95

7.68 4.72 12.5

10.94

7.32 4.68

Corner joint Al-2024-T42 Al-7070-T7451

8.34

4.9

4.68

11.86 62.41 31.23

13.23 47.54 14.35

18.59 72.74 18.45

11.13 49.27 14.24

157.88

48.68 12.78

Ti-6Al-4V

82.79

Table 5: ME values of machine learning models with augmented data on the testing set.

For the four machine learning models with data augmentation, the ME for almost all materials are near 10%, with the exception of Al-7070-T7451, which is slightly higher. It is worth noting that the ME values for butt and corner joints are particularly low after data augmentation, with minimum errors of 6.82% and 4.65% on the KELM and RF 、 BP models, respectively. When compared to the conventional Miner, Ye, and Peng models, as well as the original machine learning models, the machine learning models that incorporate data augmentation exhibit a notable increase in prediction accuracy. It has been improved to some extent for the low fatigue data leading to poor accuracy of machine learning predictive models. Stability validation of the augmented model on machine learning models In order to avoid the uncertainty associated with the fatigue test data extracted above, five-fold cross-validation is added to this section. Thereby the stability and generalizability of the data augmentation model is verified on the machine learning models. ME is used here as an evaluation indicator to validate the generalization ability of the augmented model on four machine learning models. The results are shown in Tab. 6 and Tab. 7.

Miner model ME (%)

Ye model ME (%)

Peng model ME (%)

KELM model ME (%)

SVM model ME (%)

RF model ME (%)

BP model ME (%)

Materials Butt joint

37.26 41.49 68.95 158.7 93.81

35.16 37.47 63.26

10.13

18.86 18.17 24.51 96.81 45.81

18.56 18.88 30.75 97.72 40.96

26.92 27.67 42.27

24.81 25.62 31.09

Corner joint Al-2024-T42 Al-7070-T7451

8.34

11.86 62.41 31.23

157.88

137.68

102.01

Ti-6Al-4V

82.79

75.49

44.82

Table 6: ME values of machine learning models with unaugmented data on the testing set.

Miner model ME (%)

Ye model ME (%)

Peng model ME (%)

KELM model ME (%)

SVM model ME (%)

RF model ME (%)

BP model ME (%)

Materials Butt joint

37.26 41.49 68.95 158.7 93.81

35.16 37.47 63.26

10.13

5.11 5.33 5.74

5.14 5.34 5.32 7.66 6.91

5.68 5.38

5.19 5.43 5.82

Corner joint Al-2024-T42 Al-7070-T7451

8.34

11.86 62.41 31.23

13.76 18.21 12.71

157.88

10.32

10.66

Ti-6Al-4V

82.79

7.03

9.55

Table 7: ME values of machine learning models with augmented data on the testing set.

175