PSI - Issue 62

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

1062

2

1. Introduction In recent decades, Europe has seen an increase in bridge collapses, such as the I-35 W Mississippi bridge in 2007 (Shanjiang Zhu, 2010), the Genoa bridge in 2018 (GM Calvi, 2019), and the Tuojiang bridge in 2016 (Fu You Xu, 2016). These events highlight the critical problem of ageing civil infrastructure, especially for railway bridges as critical assets of the rail infrastructure, which is crucial for the sustainability of the European transport (Comission, 2011). By 2030 and 2050, the goal is to shift 30 per cent and 50 per cent of long-distance freight from road to modes less dependent on fossil fuels, such as rail or waterways. With over 300,000 rail bridges on more than 200,000 km of railways in Europe, 35% of which are more than 100 years old (Paulsson B., 2010), there is a significant risk to the integrity of the rail network due to degradation and anthropogenic factors. Despite the obvious advantages of Structural Health Monitoring (SHM) (HAQ G., 2019), its integration into civil engineering is limited. The lack of simple damage identification algorithms and real-world demonstration cases hinders the technology transfer of SHM (Cawley, 2018). This explains the circumstance that periodic maintenance policies based on visual inspections remain widely adopted (Rosmaini Ahmad, 2012), despite their well-known limitations in detecting early-stage damage and risks derived from incorrect assessments (Andrea Meoni, 2023). In contrast, the paradigm of Structural Health Monitoring (SHM) advocates the use of continuous monitoring to identify sudden or progressive damage (F. Magalhães A. C., 2012) (Reynders, 2012). The main steps in the SHM process include operational definition, data collection, extraction of damage-sensitive features and damage classification (Alessandro Cabboi, 2016). Approaches in structural monitoring include vibration-based and static-based identification. Operational Modal Analysis (OMA) techniques, such as stochastic subspace identification (SSI) (P Van Overschee, 1996) and frequency domain decomposition (FDD) (R Brincker, 2000), are commonly used for modal identification. These techniques are efficient for global damage, but they may encounter serious difficulties in identifying local defects with limited influence on the overall stiffness. It is thus necessary to include different detection solutions with superior local damage identification capabilities. Damage-sensitive characteristics, such as those in the time and frequency domain, can be extracted from monitoring signals (Vahid Reza Gharehbaghi, 2022). The influence of operational/environmental impacts on the structure response has been also widely reported in the literature (Chunwei Zhang, 2022). Such effects, which typically induce daily and seasonal oscillations, may mask the presence of damage and must be eliminated through proper statistical pattern recognition and data normalization techniques (CR Farrar, 2012) (C. Cremona, 2018). In this context, this study focuses on the analysis of the monitoring data recorded in the Quisi Viaduct in Alicante, Spain, a six spans stainless steel railway bridge built at the beginning of the 20th century. The bridge is equipped with highly sensitive accelerometers and a dense network of deformation sensors, collecting data for approximately three years. The acceleration data is processed using an automated Operational Modal Analysis (OMA) approach based on the Cov-SSI approach. Afterwards, discussion on the influence of environmental factors and their removal is presented, followed by the generation of control charts for two-class (damaged, undamaged) damage classification. 2. Theoretical background The Continuous SHM methodology adopted comprises four key steps: signal processing, OMA reference definition, modal identification and frequency tracking through acceleration histories. In the absence of directly measurable environmental variables, but observable through modal parameter variation, Principal Component Analysis (PCA) is used as a flexible solution. Feature extraction is based on modal data, anomaly detection through statistical pattern recognition and control chart definition. The protocol for Continuous SHM covers dynamic identification, modal tracking, filtering and statistical pattern recognition.

2.1. Covariance-Driven Stochastic Subspace Identification (Cov-SSI)

The steps of COV-SSI are (i) the computation of the covariance matrix; (ii) the construction of the block Toeplitz matrix; (iii) the SVD of the Toeplitz matrix; and (iv) modal identification and stability checking. The discrete-time state-space model of a linear time-invariant 2 degree-of-freedom (DOF) system under the assumption of white noise excitation can be expressed as follows: