Understanding the influence of temperature on the natural frequencies of concrete bridges is essential for reliable vibration-based structural health monitoring. Generally, temperature shows a negative correlation with bridge modal frequencies. However, comparative investigations of temperature sensitivity between healthy and damaged bridges remain scarce. In this study, two in-service prestressed concrete bridges of identical design, located at the same site, are analyzed—one in healthy condition and the other with three confirmed tendon breaks caused by a truck impact. Both bridges are instrumented with accelerometers and temperature sensors. Operational Modal Analysis (OMA) is conducted using Frequency Domain Decomposition (FDD) and Stochastic Subspace Identification (SSI) methods. The time lag between temperature variation and modal frequency change is then examined, followed by sensitivity and correlation analyses. Furthermore, the influence of temperature gradients on modal frequencies is evaluated. Results reveal that the damaged bridge exhibits higher temperature sensitivity in its modal frequencies compared with the healthy one. A higher temperature gradient also leads to greater frequency variation. Possible physical mechanisms underlying these observations are discussed. The findings emphasize the importance of damage-state-dependent temperature normalization for vibration-based bridge monitoring and provide valuable insights for long-term modal analysis under varying environmental conditions.

Structural health monitoring is based on vibration response measurements under random ambient excitation. Sometimes, the structure is subjected to combined harmonic and random excitation, for example if the structure is located in the vicinity of a power plant. Harmonic loading is often a nuisance in damage detection if broadband excitation is desired. An attempt is often made to mitigate the harmonic component of the response from the measurement data. In this study, both stochastic and harmonic responses are retained, and damage detection is performed automatically in the time domain using autocovariance functions (ACFs). For random excitation, the ACFs have the same mathematical form as the free decay of the structure, while for harmonic excitation, the ACFs are harmonic functions. Both mathematical forms are advantageous, because they exhibit spatiotemporal correlation. Since the ACFs of the sensor network are correlated, it is possible to achieve redundant measurement data. The data space can then be divided into two subspaces, a signal space and the noise space. The information lies in the signal space, while the noise space consists mostly of measurement errors. If the structure is damaged, the new data also enter the noise space, where damage can be detected. A numerical experiment was performed to study damage detection under harmonic, random, and combined harmonic and random loading. The same algorithm could be used for all cases without modifications.

Data‑driven Structural Health Monitoring (SHM) enables damage detection without requiring a pristine baseline, making it suitable for ageing or already‑deteriorated structures. However, performance depends strongly on the informativeness of the selected feature‑extraction methods. This study evaluates the robustness of several feature‑extraction strategies through a cross‑system, cross‑domain comparison under consistent conditions, highlighting trade‑offs between detectability and localisation in both output‑only unsupervised and supervised settings. Two structures are examined—a small‑scale wind‑turbine blade and a full‑scale concrete bridge—each subjected to different damage scenarios to assess the generality and resilience of the approaches. Results show that autoregressive‑based features provide highly robust global damage detection, while frequency‑domain features excel in local sensitivity and reveal stronger sensor dependence.

Both structural degradations and environmental variations are known to influence the dynamic behavior of large and flexible structures, yet inferring the conditions that give rise to a specific response remains challenging in a monitoring context.

This study investigates how the background, resonant, and inertial components of the response react to structural versus environmental changes. These components have never been analyzed in such a context before, though they are expected to offer valuable insights into the sensitivity of each regime to specific sources of variation, as they are indeed associated with different physical mechanisms. The background component represents quasi-static or slowly varying effects, the resonant component isolates the amplifications due to natural modes of vibration, and the inertial component reflects high-frequency oscillations.

These components are derived using the recent extension of the Multiple Timescale Spectral Analysis (MTSA) framework to wave-loaded floating structures. As usual, this extension builds on the classical background-resonant decomposition, which is commonly used in wind engineering, but includes features that are needed to model the behavior of floating structures, such as the frequency dependency of the structural matrices and the inertial nature of the wave load excitations. In the end, each component is formulated through closed-form expressions, that explicitly depend on structural and environmental parameters.

Building on this framework, the present paper is organized to first introduce the methodology and analyze the sensitivity of each response component mathematically, leveraging the simple analytical formulas provided by MTSA. These closed-form expressions are particularly valuable because they allow the effects of structural and environmental changes to be assessed clearly and directly, making the sensitivity analysis both efficient and interpretable.

The approach is then applied to a numerical model of a floating tunnel, which will also be reproduced at a 1:20 scale and installed at sea next year. Structural degradations are simulated by adding concentrated masses or reducing the stiffness of anchorage points, allowing the study to capture both highly localized changes and broader environmental variations, such as shifts in wind, waves, or currents.

The primary aim of this work is to understand how each MTSA-derived component responds to these influences and to assess whether the decomposition can serve as a meaningful diagnostic tool in complex marine environments. This work represents an initial step toward developing interpretative methodologies for monitoring large-scale floating tunnels, which are essential for ensuring that such innovative structures can be safely operated and maintained over their service life.