PSI - Issue 62

S. Anastasia et al. / Procedia Structural Integrity 62 (2024) 1061–1068 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

1067

7

12.8047 Hz

6.6036 Hz 8.8067 Hz

4.5246 Hz

Fig. 3.(a) Tracking of the natural frequencies of Quisi Bridge, (b) and data normalization using PCA

This implies that the reference modal forms are kept constant throughout the monitoring period, while the reference natural frequencies vary over time as shown in the Fig. 3b. On this basis, the same four identified modes are tracked throughout the monitoring period. The quality of the pattern recognition was assessed by inspecting the statistical distribution of the residuals. Ideally, the residuals in the training period should contain only normally distributed errors resulting from limitations in the identification of the sound database of the monitored structure and marginal effects of the EOC. For the construction of the PCA, all four frequencies were used considering two principal components (PC) with 85.68% variance explained by the selected natural frequencies, i.e. for mode 1 as shown in Fig. 3b.

Out-of-control

In-of-control

Fig. 4. Hotelling's 2 control chart The residuals between the statistical models and the plotted frequencies were defined in order to calculate the Hotteling statistical distance 2 and define the control charts. The control charts referring to the entire structure evaluated in the training period are shown in Fig. 4. Two different upper control limits (UCL) were set in the control charts, corresponding to 95% and 99% confidence levels in the statistical distribution of the residuals. By defining a region under control, the appearance of out-of-control processes, possibly associated with damage, is detected in the form of data points that violate the region under control (presence of possible damage).The region in control can be defined by an interval [0, UCL] , with the upper control limit relating to the distance associated with a certain confidence level for the distribution of data within the training period. The identification of the appearance of an abnormal condition through persistent accumulations of data points that violate the region under control. 4. Conclusions The primary aim of this study was to showcase the efficacy of automatic frequency tracking for continuous Structural Health Monitoring (SHM) processing in facilities with extensive datasets. Additionally, the research aimed to explore the capability of identifying progressive damage to enhance computational efficiency. The presented results