PSI - Issue 75

Philippe AMUZUGA et al. / Procedia Structural Integrity 75 (2025) 53–64 Author name / Structural Integrity Procedia 00 (2025) 000–000

61

9

Table 1: Model Stability : e ff ect of noise on variable selection and GLM structure.

Case P [%] A [% σ (log 10 ( N cycles )) ] Nb. vars

Selected variables

Coe ffi cients c i (constant last)

[ F 2 , F , FAT 2 , FAT , t 2

[5 . 13 × 10 − [ − 9 . 45 × 10 − 7 , − 0 . 00252 , − 9 . 71 × 10 − 5 , 0 . 03372 , − 0 . 00533 , 0 . 26127] , 3 . 833472 5 , 0 . 03355 , 46 . 60 , − 24 . 19 , − 0 . 00067 , 486 . 16] , 6 . 088244 2 ] [ − 0 . 00044 , − 8 . 23 × 10 − 5 , 0 . 03242 , − 0 . 00368 , 0 . 20526 , − 4 . 76 , 1 . 10 , 1 . 31 × 10 5 ] , 1 . 737665 2 ] [ − 0 . 00046 , − 8 . 28 × 10 − 5 , 0 . 03148 , 0 . 01699 , − 5 . 66 , − 1 . 10 × 10 2 , 1002 . 87 , − 1 . 09 × 10 5 ] , 3 . 474485 [ − 0 . 00066 , − 8 . 30 × 10 − 5 , 0 . 03116 , 0 . 06226 , − 5 . 66 , 507 . 02] , 4 . 065796

2 , t 2 ]

0 1 2 3 4

0

0

6 6 8 6 8

[FAT 2 , FAT , 1 / t 2

2 2 , F , 1 / F ]

10 10 30 30

10 30 10 30

2 , t

[ F , FAT 2 , FAT , t 2

2 2 , 1 / F , 1 / F

2 , t 2 , 1 / t

[ F , FAT 2 , FAT , 1 / t

2 2 , 1 / F ]

2 , 1 / t

[ F , FAT 2 , FAT , t 2

2 2 , 1 / F , 1 / F

2 , t 2 , 1 / t

Table 2: Predictive Performance of the GLM for di ff erent noise configurations.

Case P [%] A [% σ (log 10 ( N cycles )) ] RMSE / MAE ± 1% ± 3% ± 5% 0 0 0 0.075 / 0.058 56 % 95 % 100 % 1 10 10 0.085 / 0.070 45 % 95 % 100 % 2 10 30 0.041 / 0.034 80 % 100 % 100 % 3 30 10 0.062 / 0.052 62 % 100 % 100 % 4 30 30 0.052 / 0.041 72 % 100 % 100 %

(a) Case 1: P = 10%, A = 10%

(b)Case 2: P = 10%, A = 30%

(c) Case 3: P = 30%, A = 10%

(d)Case 4: P = 30%, A = 30%

Fig. 8: Correlation plots between predicted and target fatigue life for four noisy cases ( P , A ). Gray bands indicate ± 3 σ intervals; colors reflect accuracy within ± 1%, ± 3%, and ± 5 % margins.