Smartphone Application for Structural Health Monitoring of Bridges · Figueiredo E, Moldovan I, Alves P et al.

4. Case Study #2: Twin Bridges over Itacaiúnas River

Two side by side “twin” bridges (Figure 15) over the Itacaiúnas River in Brazil and integrated in the Transamazônica Road were used to test the App4SHM under real-world conditions. One of the bridges (old bridge) was built in the 1980s and the other (new bridge) was inaugured in 2010 due to the increase of traffic. Both bridges are post-tensioned concrete box-girder bridges with a main span of 120 m and two side spans of 70 m each. Both intermediate supports are concrete piers resting on concrete piles and clamped into the girder. Small structural differences exist between the bridges, most notably the height of the piers (from the girder to the cap of the piles) being roughly 1 m larger on the old bridge. As was the case for the lab beam presented in the previous section, the tests were designed to compare the readings obtained with conventional accelerometers and App4SHM and to test the damage detection feature offered by the latter.

4.2. Ambient Vibration Tests

Ambient vibration tests were carried out in October 2021 on the two bridges to estimate the operational mode shapes, using a conventional sensing system composed of three-axis strong motion force balance accelerometers from GeoSIG. In previous tests carried out in 2018, nine checkpoints were used as shown in Figure 16. This time, additional checkpoints were added to increase the resolution of the mode shapes’ configuration, resulting in a total of 19 checkpoints. Checkpoints between sections 4 and 6 (see Figure 16) were designed to find possible cracks in that area. The checkpoint placed at section 4 was kept as a reference point throughout the ambient vibration tests. Each checkpoint is composed of one fixed accelerometer (reference) and two moving accelerometers.

The modal properties (natural frequencies, mode shapes, and damping ratios) were estimated through ARTeMIS Modal using the enhanced frequency domain decomposition (EFDD). Table 3 shows the operational mode shapes and natural frequencies for both bridges through two ambient vibration tests. The natural frequencies correspond to the average of the 19 checkpoints. As those two bridges are similar, the mode numbering is based on the direct comparison of the mode shapes. For the new bridge, the mode shape with natural frequency of 1.416 Hz is left behind, as there is no direct correspondence with the old bridge. Table 4 summarizes the estimated natural frequencies. In general, assuming both bridges have similar mass, the new bridge is more flexible than the old one.

4.3. Feature Extration and Comparative Study

A comparative study was conducted resorting to measurements in section 2 (see Figure 16) of the new bridge using both sensing systems. As the smartphone measurements are obtained from a section where frequencies F2, F3, and F4 are best read, these frequencies are used for comparison. The smartphone was placed directly on the sidewalk as shown in Figure 17.

As reference, three time series with 15 min duration each were obtained using the traditional sensing system (GeoSIG). For later comparison, three natural frequencies were estimated and are summarized in Table 5. The difference between the mean and the measured values is also summarized (see the values in the brackets). A maximum variation of 1.08% around the mean was observed in the estimated frequencies over one hour of testing, roughly.

The natural frequencies estimated using App4SHM are based on six time series of 20 s duration each obtained in the same one-hour time window (Table 6). The acquisition time was limited to ensure efficient communication between the mobile device and the server, under the rather modest 3G wireless mobile network available on the site. According to Table 6, the maximum variation of the App4SHM measurements as compared to the GeoSIG mean values (taken from Table 5) was −2.53%. The absolute values of the differences in the mean values are smaller, namely between 0.55% and 1.43%.

Comparing the means of both sensing systems and taking GeoSIG as reference, one can conclude that the App4SHM outputs consistent results in terms of natural frequencies (differences less than 1.43%, as shown in bold in Table 6), even using a short time-series duration.

Another study was carried out on the old bridge to study the variability of the measurements throughout the day. Twenty-five observations were made with the App4SHM at section 5, using the same amount of time series, during the ambient vibration test performed on the 15th of October 2021 using the traditional sensing system (GeoSIG). Table 7 summarizes the results from both sensing systems, which shows a maximum variation of 2.42% compared to the values obtained from the 6-h ambient vibration test (GeoSIG) at 19 checkpoints.

4.4. Damage Detection Performance

Finally, to test the damage detection performance of the App4SHM for long-term monitoring, 165 observations obtained at section 5 from the new bridge over the course of one year (120 measurements in October, 2021 and 45 weekly observations ever since) were included in the training data set of the Mahalanobis-based machine learning algorithm. Each observation is composed of three natural frequencies (F1, F3, and F5, see the corresponding modes in Table 3).

After the training step, two tests were performed independently on both bridges, using one observation from each bridge at section 5 and the training database from the new bridge only. The markers in Figure 18 show the DI for each observation obtained from the new and old bridges against the threshold defined for a level of significance of 5%. As expected, the results show that the observation from the old bridge is not statistically similar to the training data from the new bridge, giving an indication of some anomaly and recommending further analysis to discover the source of such outlier (the observation in red is highlighted inside a red circle, in Figure 18b). Conversely, the observation from the new bridge ranks below the threshold and no damage warning is issued (the observation in green is highlighted inside a green circle in Figure 18a).