PSI - Issue 75

Mohamed El Yazrhi et al. / Procedia Structural Integrity 75 (2025) 262–275 Mohamed El Yazrhi , Jean-Yves Disson / Structural Integrity Procedia (2025)

272

11

5.2. Results and Comparative Analysis of Prediction Methods Figure 4 presents a direct comparison between the actual cumulative FDS monitored on the highway dataset and the predictions obtained from the three tested methods: linear extrapolation (rule-of-three), Monte Carlo simulation using a Gaussian Mixture Model, and an analytical method based on the Central Limit Theorem (CLT) as recommended in AFNOR XP X50-144. Overall, both the Monte Carlo and CLT-based approaches closely follow the ground truth, whereas the linear extrapolation method consistently overestimates the accumulated FDS across nearly the entire frequency range.

Figure 4 : Prediction Accuracy FDS calculation Methods

To quantify these differences, Figure 5 shows the relative prediction error as a function of frequency. The linear extrapolation method exhibits the highest error, with values ranging from approximately 6% to 8%, and peaking at lower frequencies. In main below 3% across the entire spectrum, with the Monte Carlo approach achieving the best accuracy overall. This highlights the importance of incorporating variability and distribution fitting when predicting future damage.

Figure 5 : Relative error (%) of each prediction method across the frequency spectrum