Digital twinning approaches in structural health monitoring (SHM) are applied in conventional cable-based SHM systems, relying on the collocation of sensor data and physics-based models on a single server (“single source of truth”). This “centralized” digital twinning rationale increases the risk of data loss in the event of server malfunctions, thus converting the single source of truth into a “single point of failure”. To mitigate this risk, we introduce a decentralized digital twinning approach for SHM. Building upon advances in wireless “smart” SHM, we propose distributing physics-based models into processing units of wireless sensor nodes, thus transferring large parts of SHM tasks to the sensor nodes, which collaboratively analyze the structural behavior via limited wireless communication. The relative independence of wireless sensor nodes grants SHM systems redundancy, which enhances the robustness of digital twins.

Physics-based models are herein represented by finite element (FE) models, which are reduced via model-order reduction (MOR) (dynamic condensation) in order to (i) be computationally manageable by the wireless sensor nodes and (ii) be compatible with the topology b of the sensor nodes (Fig. 1). Changes in structural conditions are modeled as perturbations of structural parameters θ, which result in families of L reduced-order models. SHM tasks involve segmenting the degrees of freedom (DOFs) of the reduced-order FE models into partial models, encompassing internal DOFs (w₁, w₂) and interface DOFs (v). Thereupon, wireless sensor nodes located at internal DOFs are tasked to estimate the structural responses (e.g. accelerations) at interface DOFs using the respective families of partial models. Finally, the fittest partial model θ_w is retrieved upon minimizing the differences between estimates of interface-DOF responses Ẍ_v and the actual interface-DOF responses Ÿ_v, for a range of Ω frequency components.

A case study of the decentralized digital twinning approach showcases the capability of the digital twins in detecting structural damage on a 4-story laboratory shear-frame structure. First, FE models of the structure are created, corresponding to the intact structure as well as to damage scenarios, and reduced into so-called “super-elements” (Fig. 2). Next, the super-elements are segmented into partial models PM1 & PM2, which are embedded into wireless sensor nodes. Laboratory tests are conducted both on the intact structure and after inflicting damage on the 3^rd story.

The results, tabulated in Table 1, demonstrate the capability of the wireless sensor nodes in retrieving the super-element that represents the correct structural behavior in both tests. These preliminary findings are expected to pave the way towards developing fully-fledged decentralized digital twins into real-world smart monitoring systems, thus advancing structural maintenance.