Start with the photonics, not the tunnel. Shine a laser pulse down an ordinary telecom fibre and a faint fraction scatters straight back at you off the glass’s own microscopic imperfections — Rayleigh backscatter. Measure the phase of that return, pulse after pulse, and every metre of the fibre becomes a virtual strain gauge. This is distributed acoustic sensing (DAS), and it is the quiet frontier that turns a cable already lying in a cable trough into tens of thousands of sensors with no new hardware buried in the rock. That is the trick I want on my permanent way.

Because I am 57.1 kilometres of it. The Gotthard Base Tunnel — bored by AlpTransit Gotthard AG, opened in 2016, the deepest railway on the planet with up to 2,300 metres of Alp pressing down on my two single-track tubes — was engineered as a structure. It is increasingly operated as a dataset. The rock at that depth is not inert; it creeps, it converges, it breathes on a seasonal cycle, and squeezing ground can close a bore by millimetres a year. A single fibre laid the length of both tubes reads that motion continuously, where a human inspector with a total station reads it once a quarter.

←TODAY: A dark fibre in a 2016 tunnel already carries the strain signal of 57 km of rock — the sensor exists; the reading is the new part. →3012: The century-scale bore that survives is the one whose telemetry outlived three generations of signalling. Fulcrum: Maintenance is not an event at year sixty; it is a data stream you either started recording in year one or you did not.

From mirror to decision

A stream of numbers is not yet a twin. The distinction matters, and the mathematics is where it lives. Antil and colleagues, in the MATH-DT report (arXiv:2402.10326, 2024), make the sharp claim our own PAZ concept panel on digital twins leans on: a real twin starts from this asset, not an asset. That one word forces multi-scale, multi-physics, uncertainty-aware modelling rather than a tidy textbook idealisation of a tunnel. BIM supplies my geometric backbone — the static, design-time model. The twin fuses strain, temperature, groundwater and load telemetry on top of it across my whole life.

The point the vendors undersell — and ITWeb’s coverage of mature twins on mega-projects is honest about the gap — is that a twin which only renders a dashboard is a museum. The engineering value is in the coupling that recommends an action under uncertainty: re-profile that segment, schedule the 02:00 window here, watch this convergence rate before it crosses threshold. A dashboard shows you the rock moving. A twin tells you when to stop a train.

Here is the trade-off, stated plainly: a fibre gives you a million readings and zero judgement. Drown a small operations team in raw DAS traces with no thresholding and no ground-truth calibration, and you have bought a very expensive way to miss the one signal that mattered.

Atelier: An office building within long-lived infrastructure — a station, an underground box, a viaduct approach — does not need its own interrogator to act on this. The Monday move: write the sensor plan into the BEP before the structural package is frozen, specifying which fibres are laid as spare “dark” strands for future DAS, and who owns the calibration baseline. Retrofitting a sensing fibre after the concrete cures costs an order of magnitude more than a coil left in the trough.

Hack: Turn an optical phase shift into the rock strain it actually reports — the physics under every DAS reading. The interrogator sees a phase change; you want microstrain. The relation is Δφ = 4πnGε/λ, with fibre index n, gauge length G and wavelength λ.

import math
n   = 1.468      # fibre core refractive index
lam = 1550e-9    # interrogator wavelength (m)
G   = 10.0       # gauge length (m)
sens = 4*math.pi*n*G/lam          # rad per unit axial strain
print(f"{sens:.2e} rad/strain -> {1e9/sens:.2f} nanostrain per radian")
# ~1.19e8 rad/strain  ->  ~8.4 nanostrain resolvable per radian of phase

Eight nanostrain per radian, along every ten-metre gauge, the full length of the bore. That is the resolution that lets you watch squeezing ground before it is visible to a survey.

So the imperative is small and it is now: find out whether the infrastructure your practice is designing has a spare fibre in its cable plan — and if it does not, add one before the drawings go out. The cheapest sensor is the one you laid while the trench was still open.