PSI - Issue 78

Marco Pirrò et al. / Procedia Structural Integrity 78 (2026) 1641–1648

1647

some classification uncertainty due to the partial restoration of the undamaged state (Kim et al. 2021).

Fig. 5. Average MAE values (MAE av ) for each channel and scenario.

Table 1. Performance indicator for damage detection method computed with (a) and without (b) the datasets collected in RCV scenario. % A1 A1 A3 A4 A5 A6 A7 A8

(a) (b) (a) (b) (a) (b)

95.9 87.8 91.8 85.7 95.9 93.9 89.8 91.8

ACC

97.4 96.9 96.9

100 100 100

100 100 100

97.4 97.4 96.9 96.9 96.9 96.9

100 100 100

100 100 100

100 100 100

DR

5.9

35.3 23.5 35.3

5.9

17.6 29.4 23.5

FAR

0

0

0

0

0

0

0

0

Table 2. Average (MAE av ) and standard deviation (σ MAE ) of MAE for each test scenario.

Scenario Index

A1

A1

A3

A4

A5

A6

A7

A8

MAE av

0.015 0.014 0.014 0.016 0.014 0.012 0.009 0.011 0.004 0.003 0.005 0.004 0.004 0.003 0.002 0.003 0.037 0.045 0.044 0.046 0.036 0.036 0.032 0.034 0.010 0.010 0.016 0.017 0.008 0.005 0.007 0.007 0.074 0.097 0.123 0.099 0.066 0.073 0.080 0.074 0.020 0.023 0.033 0.026 0.012 0.021 0.023 0.018 0.021 0.025 0.024 0.027 0.017 0.017 0.014 0.018 0.008 0.009 0.008 0.011 0.008 0.008 0.009 0.010 0.071 0.087 0.086 0.119 0.067 0.080 0.083 0.069 0.020 0.022 0.016 0.024 0.019 0.024 0.014 0.020

INT

σ MAE

MAE av

DMG1

σ MAE

MAE av

DMG2

σ MAE

MAE av

RCV

σ MAE

MAE av

DMG3

σ MAE

The capability of the anomaly localization approach is further highlighted in Fig. 5, which presents the average MAE values (MAE av ) across different channels and damage conditions. As shown in the figure, sensor A3 registers the highest MAE av under damage scenario DMG2, while sensor A4 shows the peak value for DMG3. To provide a clearer evaluation of the anomaly detection method’s effectiveness, Table 1 lists the results for Accuracy (ACC), Detection Rate (DR), and False Alarm Rate (FAR), calculated both with (a) and without (b) the inclusion of datasets obtained under RCV conditions. It is important to note that the trained model consistently demonstrates strong performance in terms of Accuracy and Detection Rate, achieving values exceeding 85% and 96%, respectively. However, while the False Alarm Rate remains significant when RCV data is included, it drops to 0% across all channels once the “uncertain” datasets are excluded . Moreover, the MAE average (MAE av ) and its standard deviation ( σ MAE ), calculated for each sensor channel and damage case (see Table 2), provide a clear indication of the damaged section of the bridge across different scenarios