This article presents an innovative approach to structural health monitoring of historic masonry structures through the integration of multiphysics data into thermoelastic peridynamic modelling. The prevailing research trend increasingly follows the idea that monitoring existing structures should emulate the approach a physician adopts in diagnosing and monitoring a patient. This perspective implies the integration of a wide range of information representing different physical phenomena, which can be interconnected and mutually informative. In this paper, the authors propose a novel approach to the eigenvalue problem based on a novel continuum theory. Thanks to its direct formulation, this theory allows the derivation of a temperature-dependent system stiffness matrix, thus enabling the definition of temperature-dependent eigenvalue problems. Furthermore, to simulate behaviours difficult to model using classical theories, the authors introduce a hybrid structural health monitoring strategy combining this causal formulation with black-box correlational models. This approach is initially applied under operational conditions but is designed to be easily extended to scenarios involving high-intensity events, such as earthquakes. This is made possible by the integrodifferential formulation of peridynamics, which inherently leads to nonlinear equations of motion, regardless of the constitutive model adopted. More specifically, the proposed framework integrates various types of multiphysics data, including temperature measurements, displacement fields, and other environmental parameters. Correlational models are employed to estimate the behaviour of structural parameters, such as the variation of the thermal expansion coefficient, as a function of observable quantities. This allows the system to infer difficult-to-model mechanical responses from accessible environmental and operational information, while preserving the underlying physics of the problem, thanks to peridynamics. The framework is applied to the northwest bell tower of the Sanctuary of Vicoforte. The structure is equipped with a dense sensor network, including satellite corner reflectors, thermocouples, laser measure with inclinometers, strain gauges, and accelerometers. The study is part of the DPC/ReLUIS 20242026 project, which focuses on advanced satellite and in-situ monitoring techniques.

While neural networks represent a promising approach for evaluating sensor data to assess damage presence, location and severity, large amounts of data are required for training. However, the generation of experimental data is both labor-intensive and costly. Transfer learning is a well-established method in deep learning that enables the reuse of knowledge from pre-trained models to improve performance and reduce data requirements across various domains such as computer vision and natural language processing. Its application within the field of strain-based structural health monitoring (SHM) has received little attention from currently published literature.

This work develops a resource-efficient transfer learning approach for strain-based SHM. While the Convolutional Neural Network (CNN) model learns relationships between strain distribution and crack geometry in an aluminum beam from Finite Element (FE) data, fine-tuning adapts it to experimental conditions, accounting for factors such as measurement noise, increasing overall accuracy and robustness.

An encoder-decoder CNN (UNet) is initially trained with synthetic data from FE simulations. The model is then fine-tuned based on a smaller experimental Digital Image Correlation (DIC) dataset obtained from an aluminum beam subjected to a four-point bending fatigue test. For this purpose, the encoder part of the CNN is frozen, while parameters of the final layers of the decoder are updated. The approach is validated with respect to its accuracy, robustness and applicability for SHM systems.

The presented approach significantly reduces the experimental data requirements while improving damage detection performance for an aluminum beam under four-point bending. This study demonstrates that transfer learning enables efficient adaptation to real-world conditions, offering a cost-effective and scalable solution for data-driven SHM.

The identification of dynamic parameters in nonlinear systems is a challenging task, particularly when dealing with experimental measurements. If the nonlinearity is negligible for the response, one can rely on linear techniques without compromising the accuracy of the solution. However, if the nonlinear effects are significant, as frequently happens under strong excitations, these assumptions do not hold anymore, and one should refer to more sophisticated methods. This is the case of bistable systems, i.e., systems characterised by two stable equilibria and one unstable equilibrium point. In experimental settings, additional uncertainties may arise from the definition of boundary conditions, materials, and manufacturing. In this context, the need for a technique capable of instantaneously tracking the nonlinear parameters of systems exhibiting rapid changes, such as the bistable ones, is fundamental. This requirement is mainly linked to the ultimate purpose of these systems, namely the dissipation and/or absorption in terms of instantaneous energy, which is critical in real-world applications. This work describes a novel method for experimentally measuring and identifying dynamic parameters of bistable systems. An instantaneous probabilistic estimator, constructed using the unscented Kalman filter (UKF), is proposed for the identification of the parameters governing the bistable dynamics. The method is validated by employing experimental measurements achieved during a testing campaign carried out on a 3D-printed bistable sample, with the final goal of estimating the instantaneous energy dissipated under different external excitations. To this end, an ad hoc testing machine, composed by a pendulum actuator, was built to accurately simulate the single degree-of-freedom model employed in the identification task. The procedure proved robust and reliable, also compared to other classical nonlinear identification methods.

Temperature variations can significantly influence the dynamic response of bridges, masking structural changes and complicating vibration-based damage detection. To develop and validate damage detection algorithms under such conditions, it is essential to generate realistic data that capture these thermal effects. However, real damaged data are rare, thus numerical simulations using finite element (FE) models are often employed to test both damage detection and temperature compensation methods.

This study presents a data-informed FE modeling approach that reproduces thermal effects directly from experimental observations. Unlike conventional thermo-mechanical formulations, the proposed method learns how temperature modifies the effective mechanical properties of the structure using long-term frequency-temperature monitoring data.

A uniformly distributed dataset of temperature-sensitive (TS) parameters, such as Young's modulus, is first generated within realistic bounds. The updated FE model of the structure is then used to compute the corresponding modal frequencies, forming the training data for a surrogate model that links dynamic response to TS parameters. This surrogate is subsequently applied to field monitoring data, using measured modal frequencies to infer experimental TS parameter variations. A regression model is finally trained to map temperature to these inferred parameters, capturing their temperature-dependent evolution.

Once established, this model enables realistic simulation of thermal effects by adjusting FE parameters according to any prescribed temperature variation, and it can be extended to simulate damaged conditions. The approach is demonstrated on the KW51 railway bridge, accurately reproducing observed thermal trends and enabling the generation of realistic synthetic data.

Structural health monitoring (SHM) is an essential method for evaluating the performance of aging civil infrastructure. Earlier research relied on heavy dynamic shakers to examine structural dynamics, but these are cumbersome, prompting a shift toward using ambient vibrations to extract dynamic features. This study highlights the effectiveness of recent developments in structural health monitoring utilizing sensing, virtualization, and machine learning (ML)-enabled schemes.

First, it introduces a “virtual shaker” concept that effectively replaces physical shakers, offering all the necessary features for SHM. The concept of the virtual dynamic shaker (VDS) is analogous to a secondary system, such as a tuned mass damper (TMD), but it is a virtual device that is attached virtually to a primary structure. The system identification (SI) technique using the VDS is based on the dynamics of the combined primary structure and the VDS for the estimation of structural modal parameters. The dynamic response of the VDS is amplified around the natural frequency of the primary structure and depends on the level of damping ratio of the VDS. In this manner, the natural frequency of the primary structure can be determined as it represents the frequency at which the variance of the VDS response reaches its maximum (Fig. 1). The structural damping ratio can be identified through the derived ratio of the response variances of VDSs attached individually to the primary structure with different damping ratios. The proposed VDS scheme is validated in detail through extensive examples of building structures under wind loads using numerical simulations and full-scale records obtained through full-scale measurements, with emphasis on the damping estimation. In addition, a comparison with other popular output-only SI schemes is made to examine the efficacy of the proposed VDS-based scheme. On the basis of the extensive examples, the damping estimates by the VDS are overall accurate, comparable with those by other popular schemes, which corroborates the efficacy of the proposed VDS as an easy-to-use approach for the output-only SI.

Second, traditional SHM systems typically require wired connections, often called “hub and spoke,” where sensors are placed throughout the structure and connected to a central data acquisition server. To overcome issues associated with long cables, a unique prototype system based on a virtual wiring system called “SmartSync” has been developed. It is an “IoT” system with “edge computing” that leverages the building's existing Internet backbone as a network of “virtual” instrumentation cables, performing limited computations at sensor locations (Fig. 2). Since the system is modular and mostly “plug-and-play,” units can be quickly deployed anywhere with power and Internet access, as initially demonstrated in the Burj Khalifa, the world's tallest building.

Finally, in an era of unprecedented sensing technology that enables collecting and analyzing large volumes of climate and infrastructure data, new demands have arisen. This surge in data has led to the development of data-driven models to better assess the condition of infrastructure and implement solutions focused on sustainability and resilience. The paper addresses new advancements in system identification involving non-stationary observations and their real-time monitoring.

Nowadays the integration of advanced sensors, pervasive connectivity, and large-scale data analytics is fundamentally transforming Structural Health Monitoring (SHM) and structural identification. The importance of the state-of-the-art methodologies and technologies, emphasizing knowledge transfer, data integration and advanced data-driven paradigms for remote, contactless, and cost-effective monitoring are relevant for the static and dynamic structural characterization of the bridges and viaducts both short and long-term. Combining Finite Element (FE) models and large quantities of data generated by modern monitoring systems (i.e. proof load test and continuous monitoring), due to the growth of the sensing technology, high-speed data transfer, and cloud-based computation, allow Model Driven (MD), Data Driven (DD) and hybrid methodologies to analyse the global structural behaviour and to identify locals defects. In both MD and hybrid strategies, a model updating process (MUP) is required to align FE models with the sensors data. Next to the deterministic inverse model updating methods such as Genetic Algorithms (GAs), used to minimize the difference between sensors data and the FE model, probabilistic approaches are rising in the SHM field to explicitly take into account uncertainties and provide confidence levels. Static updating involves calibrating a model to match rotation measurements of MEMS biaxial clinometers focusing on the stiffness of the system, while dynamic updating focuses on refining the model based on vibration data, including natural frequencies and mode shapes. In the dynamic MUP, the MEMS accelerometers sensors signals are used to identify the structure modal parameters under real-world operational conditions, clustering the data by means of the Operational Modal Analysis (OMA). In this paper, both deterministic and probabilistic approaches are implemented and compared in terms of their performance and cost-benefit when applied to several real-world bridge case studies within the Sacertis SHM diagnostic industrial framework. The methodologies are first validated on a theoretical benchmark case and subsequently applied to a selected real bridge. The results obtained from the two optimization strategies enable the assessment of the strengths and limitations of the integrated framework developed by Sacertis, which has been deployed on more than 90 bridges for automated static and dynamic model updating within the SHM process.