PSI - Issue 57

Mohammad F. Tamimi et al. / Procedia Structural Integrity 57 (2024) 121–132 Mohammad F. Tamimi & Mohamed Soliman/ Structural Integrity Procedia 00 (2023) 000 – 000 8

128

To ensure the trained ANN model is neither overfitting nor diverging when analyzing datasets outside of its training domain, a new testing dataset (not included in the training dataset) comprising 500 samples was randomly selected from the input parameter ranges. These samples were then tested against the FE results. Figure 6 illustrates the comparison between the FE outputs and the ANN predictions. The results indicate that the ANN offers a precise prediction of the crack driving parameter for these stiffened panels, with an average prediction error of less than 5%. Therefore, the ANN will be employed in subsequent sections for validating the crack growth prediction approach and conducting the sensitivity analyses.

Fig. 6. Probability plot illustrating the correlation between FE results and ANN predictions based on a dataset not included in the training data. The ability of the ANN-assisted simulation model to predict crack propagation in stiffened panels is validated using experimental results from Mahmoud and Dexter (2005). The trained ANN is used to predict the J -integral corresponding to the given load levels, crack sizes, and the cross-section of the stiffened panel for each load cycle. Then, the obtained J -integral using the ANN at specific input parameters for each load cycle, in conjunction with the analytical crack propagation rule (Equations (1)-(2)), are integrated to compute the crack size increment under the applied load. Figure 7 displays the crack growth profiles from the ANN-assisted approach in comparison to the experimental results. As shown, the adopted technique can capture the crack growth behavior of the tested specimen.

Fig. 7. Crack propagation profiles obtained using the proposed ANN-assisted simulation approach and experimental results reported in Mahmoud and Dexter (2005).