The phrase "digital twin" is used loosely, but in SHM it has a precise and demanding meaning: a computational model of a specific structure that is continuously updated by that structure's own sensor data, so the model and the asset stay in step throughout the asset's life.
More than a 3D model
A static CAD model or a one-off finite-element analysis is not a twin. What makes a digital twin is the <strong>live link</strong>: measurements flow in, the model adjusts its parameters to match reality, and the synchronised model then returns quantities you cannot measure directly — internal stresses, fatigue accumulation, remaining useful life.
The feedback loop
A working twin runs a loop. Sensors report the structure's response; the model is updated so its predicted response matches the measured one; the updated model is interrogated for condition and forecasts; and those forecasts inform decisions that may even change how the structure is operated. The loop repeats continuously, and the twin's fidelity improves as it accumulates history.
What a twin makes possible
By bridging sparse measurements with physics, a digital twin reaches toward the higher levels of damage identification:
- estimating stress and fatigue at locations with no sensor, by interpolating through the physics;
- running "what-if" scenarios — a heavier load, a future storm — on the model before they happen;
- forecasting remaining useful life and the optimal time to intervene;
- carrying knowledge across the whole lifecycle, from commissioning to decommissioning.
The most capable twins are hybrid: physics-based models supply interpretability and extrapolation, while data-driven components absorb the messy reality the physics cannot fully capture. Neither alone is enough.
The hard parts
Twins are demanding. Model updating can be ill-posed — many parameter sets may explain the same data — so uncertainty must be quantified, not hidden. Keeping a high-fidelity model running fast enough to stay "live" is a real computational challenge, often met with reduced-order or surrogate models. And a twin is only as trustworthy as the sensor data feeding it.
Where it is heading
As sensing gets cheaper, computing gets faster and machine learning matures, digital twins are moving from research demonstrators toward standard practice for high-value infrastructure. The promise is compelling: not just knowing a structure's condition today, but credibly predicting its future — and managing it accordingly.