PSI - Issue 84

Michele Morici et al. / Procedia Structural Integrity 84 (2026) 89–96

94

Fig. 4. Rotations data (column 1-3) and displacement data (fourth column) vs temperatures: a) – d) thermocouple TC1-s; e)-h) thermocouple TC5-s; i)-l) thermocouple TC2-b1; m)-p) thermocouple TC1-b2 and q)-t) thermocouple TC3-b1.

The selected training period spans from January 2025 to the end of May 2025 and includes the progressive temperature increase associated with seasonal variations. Incorporating this seasonal effect allows the control chart to better reflect the different rotation trends observed across temperature ranges, as previously discussed, have been shown to significantly influence the bridge’s structural behaviour,. The upper control limit is statistically defined using the cumulative distribution function of the residuals computed during the training phase, with the threshold set at 0.95. The results show that the number of detected outliers remains relatively stable throughout the analysed period, covering both the training and post-training phases. This outcome suggests that the regression model is capable of reproducing the three rotation responses measured on beam #2 with a reasonable level of reliability. Such reliability is expected to improve as the length of the training dataset increases. From a practical perspective, a training period covering an entire year, and thus a complete seasonal cycle, would likely lead to a more robust control chart and a further reduction in the number of spurious outliers.