7 Anomaly Detection Through Long-Term SHM: Some Interesting Cases on Bridges 59 Fig. 7.1 Bridge elevation view with identification of the structural joint position Fig. 7.2 Plan view of the bridge: Cn identifies the nth span, blue dots are two-axis clinometers, and green dots identify a node made of a two-axis clinometer and a three-axis accelerometer collected, pre-preprocessed, and filtered at the gateway level, to send a selection of significant information to the cloud for further analyses and long-term storage. The biaxial MEMS clinometers installed on the viaduct have two measuring axes x and y, aligned to the transversal and longitudinal directions of the bridge, respectively, while the triaxial MEMS accelerometers have the additional z-axis oriented along the vertical direction as shown in Fig. 7.2. Both clinometers and accelerometers are equipped with a 32-bit microcontroller for data processing at the sensor level. The clinometer data acquisition is executed by the gateway following a “polling cycle,” a predetermined time interval interrogation under a sequential node order, adaptable with remote control. Each clinometer records a window of 1 second length with a sampling frequency of 208 Hz, providing synthetic statistical parameters like the mean value, standard deviation, maximum and minimum rotation, internal temperature, and relative humidity. The chosen sequential approach allows to spare network resources, missing the chance to have contemporary sampling: this just provides the bridge average behavior over a certain time lag in the order of minutes. To optimize the cloud resources and to guarantee some redundancy, the gateway stores the data received locally in a buffer memory and forwards it via a mobile network to the cloud database. The accelerometers perform a continuous data stream, with a sampling frequency of 100 Hz on the three measurement axes: as anti-aliasing filters are not provided, the real sampling frequency is 56 kHz, then followed by proper filtering and downsampling at the microcontroller level to get to the mentioned 100 Hz sampling frequency, without aliasing. The data are pre-selected and filtered at the gateway level to record and send to the cloud only those data selected from continuous acquisitions that have the most relevant energy content in terms of vibrational response. In addition, the gateway can further perform some pre-analysis based on defined threshold levels and on relevant groups of sensors, presenting the opportunity to have a prompt alerting system in case of anomalies. Anomaly conditions detected by different sensors correlated in space and time and the possibility to perform multi-parameter cross-checks help increase the reliability offered by the SHM system, avoiding false positives as much as possible. In the case of threshold exceedance, alerts are propagated to the cloud with attached an estimated severity metadata, to be handled with the appropriate urgency.
RkJQdWJsaXNoZXIy MTMzNzEzMQ==