PSI - Issue 80

306 Mengke Zhuang et al. / Procedia Structural Integrity 80 (2026) 299–309 Author name / Structural Integrity Procedia 00 (2019) 000 – 000 where is the crack length range from 30−40 mm and can be evaluated from the DBEM and CSDE method. Multi-fidelity modelling is employed to efficiently predict the effective stress intensity factor K eff at the crack tip. 8

Fig. 2. Figure showing: (a) The geometry of the shallow shell window section with the crack located in the region A. (b) The detail illustration of the crack location with = 29.24◦ from the fillet centre and crack initiation angle . Two fidelity levels are established through mesh refinement strategies: the low-fidelity model utilizes a coarse discretization scheme illustrated in Figure 3(a), while the high-fidelity model employs a refined mesh configuration shown in Figure 3(b). The coarse mesh comprises 96 boundary elements with 80 DRM points, incorporating 8 elements around the crack tip region, requiring 24.01 seconds of CPU time per evaluation. In contrast, the fine mesh features 140 boundary elements and 212 DRM points, with resolution of 15 elements for the crack, at a computational cost of 102.76 seconds per evaluation. This results in a cost ratio of = / ≈ 4.3 . Table 2. Details of the shell structure parameters and the random variables Zhuang et al. (2025). Parameter Description Distribution Mean COV 2 Inner width Normal 0.468 m 0.01 2 Inner length Normal 0.273 m 0.01 2 Inner radius Normal 0.127 m 0.01 h Thickness Normal 0.01 m 0.01 Radius of curvature Normal 2.73 m 0.01 Crack initiation angle Normal 29.24◦ 0.05 P Domain pressure Normal 18 Psi 0.04 Fracture toughness Deterministic 26 MPa √m [2024-T3 Aluminum] Since the limit state function depends on SIF obtained from DBEM analysis, and given the substantial computational time for high-fidelity simulations, direct Monte Carlo simulation is computationally prohibitive. Therefore, a pre-trained high-fidelity Kriging surrogate model consisting of 267 training points was developed, achieving a root mean square error (RMSE) below 5%. This validated surrogate model enables Monte Carlo simulation with 107 samples to establish reference failure probabilities for comparison.