PSI - Issue 84

Michele Morici et al. / Procedia Structural Integrity 84 (2026) 89–96

90

Keywords: Structural Health Monitoring; Bridge Static Response; Temperature Effects; Static Monitoring

1. Introduction Bridges are essential assets within transportation networks, and maintaining their structural integrity is a primary requirement for ensuring safety, serviceability, and economic sustainability. Since the 1990s, bridge condition assessment has mainly relied on two complementary approaches. Bridge Management Systems (BMSs) are based on periodic visual inspections and provide a systematic framework for asset-level decision-making (Dayan et al. 2022; Ling et al. 2022; Natali et al. 2023). Structural Health Monitoring (SHM), on the other hand, employs sensor-based systems to continuously measure structural and environmental parameters, enabling performance evaluation and damage detection over the operational life of bridges (Figueiredo and Brownjohn 2022; Deng et al. 2023; Sohn et al. 2003; Farrar and Worden 2012). In addition to damage identification, SHM data support the interpretation of structural response under traffic, wind, and thermal actions and allows their respective effects to be distinguished. Typical monitoring systems include accelerometers, strain gauges, fiber optic sensors, displacement transducers, thermistors, anemometers, and vision-based devices (Lynch 2006; Mukhopadhyay and Ihara 2011; Micozzi et al. 2023; Micozzi et al. 2025). Within SHM frameworks, damage detection methods are generally classified into numerical model updating and data-driven approaches. Model updating techniques aim to calibrate numerical models using measured data (Alvandi and Cremona 2006; Xiong et al. 2009; Ierimonti et al. 2023; Giri et al. 2024), but their application to real bridges is often limited by computational cost and by uncertainties related to material properties, boundary conditions, and environmental influences. Data-driven methods overcome these limitations by directly extracting damage-sensitive features from measured responses and environmental variables, without requiring an explicit structural model (Azimi et al. 2020; Yang et al. 2021; Scozzese and Dall’Asta 2024; Meoni et al. 2024; Carbonari et al. 2024). Among environmental actions, temperature has been consistently identified as the dominant factor influencing bridge response, often exceeding the effects induced by traffic loads (Kromanis and Kripakaran 2017). Seasonal temperature variations can induce strains that are an order of magnitude larger than those associated with vehicular loading (Hoult et al. 2010) and significantly affect global stiffness and modal properties (Kromanis and Kripakaran 2016; Luo et al. 2022). From an engineering perspective, the accurate interpretation of temperature-induced response variations is therefore essential for reliable condition assessment and anomaly detection. Bridges are subjected to complex thermal actions arising from daily and seasonal temperature cycles (Krkoška and Moravčík 2015; Ngo and Nguyen 2024; Lorenz et al. 2024). These actions are mainly governed by solar radiation and convective heat exchange, while the resulting internal temperature distribution depends on material thermal properties, structural geometry, and thermal inertia. As a result, temperature fields within bridge components are generally non uniform and may exhibit pronounced non-linear gradients. In practical monitoring applications, temperature measurements are often available only at a limited number of locations, although complementary techniques such as thermal imaging can improve spatial coverage. Furthermore, due to the delayed response of structural deformations to temperature changes, measurable time lags between thermal inputs and structural responses are commonly observed (Jiang et al. 2024). Recent SHM studies have increasingly focused on exploiting temperature-driven variations in static structural responses—such as displacements, rotations, and strains—for condition assessment and anomaly detection. This process typically involves two key steps: the development of response models that account for environmental effects and the definition of appropriate detection thresholds. Regression-based models are widely adopted to predict structural responses using environmental variables, thereby reducing the risk of misinterpreting benign temperature induced variations as damage (Kromanis and Kripakaran 2014). Commonly used techniques include Principal Component Analysis (PCA) (Wang et al. 2020), Support Vector Regression (SVR) (Li et al. 2003; Smola and Schölkopf 2004; Vapnik 2008), Multiple Linear Regression (MLR), and autoregressive models applied to temperature-dependent variations of modal properties (Kromanis and Kripakaran 2014). More recent contributions have explored machine-learning methods, such as Long Short-Term Memory (LSTM) networks and robust PCA, to improve prediction accuracy and environmental effect removal (Zhang et al. 2022).