Bridge monitoring that scales to a portfolio.
Monitoring one bridge well is a solved problem. Monitoring hundreds affordably is not, and the portfolio is where the question actually sits: which bridge needs attention this quarter?
Three things drive what bridge monitoring costs at scale
Not one of them is the algorithm. We address all three, and the first is the one that compounds into the other two.
Portfolio size
The number of structures under management. Anything priced per bridge is multiplied by a number you cannot reduce, so the per-bridge cost is the only lever available.
Sensors per bridge
Hardware, installation, traffic disruption and maintenance across the asset life. Every sensor is also a cable, a channel and a fault point. This is the lever with the most room in it.
Proprietary lock-in
Non-standard hardware and closed data formats tie an operator to one supplier, complicate every tender, and leave the inventory data fragmented across incompatible systems.
Four principles
Take the fleet viewpoint
Monitoring is about a whole inventory, not the optimisation of one structure. The economics, the priorities and the real question only make sense at portfolio level.
Apply structure, data science and AI together
Same-category data across the portfolio reduces the sensors each bridge needs. Cross-bridge learning turns sparse instrumentation into trustworthy structural insight.
Use standard, off-the-shelf sensors
Reliable, durable and multi-functional. Standard hardware makes tendering simple, ends vendor lock-in, and keeps the data layer uniform across the inventory.
Deliver value now, with headroom ahead
Optical sensing, edge AI and, on a longer horizon, chip-scale interrogators. Value on day one, and a cost curve that bends downward rather than flat.
Measure at five, know at twenty
Virtual sensing is what lets a sparse instrumentation plan carry the information of a dense one. Three complementary algorithms do the work, and none of them is new theory for us. They are production code, running on offshore structures today.
Operational modal analysis using covariance-driven stochastic subspace identification extracts the structure's modes from ambient vibration, with automated mode selection.
Modal expansion reconstructs the response at every original sensor location from a sparse subset. This is the step that turns five sensors into knowledge of twenty.
Influence-line extraction uses known load passages, with Kalman filtering to separate the quasi-static structural response. The shape that emerges is a direct measure of how stiffness is distributed along the span.
A small number of well-placed standard sensors, on hardware you can tender.
Modal basis, load paths and the physics of the structural type.
Plus a defensible minimum sensor count per structural type, and a ranked list of which bridges need attention first.
Every train passage is a known-axle-load measurement event. Engineers run controlled load tests periodically. Traffic lets us run them continuously, at no operational cost.Which is why the excitation problem on a busy structure solves itself
Rail
Known axle configuration, known weight, known speed and a measurable structural response. Train passages are strong and discrete, and discrete is exactly what influence-line extraction wants.
Highway
Random traffic is perfectly acceptable ambient excitation for operational modal analysis, and heavy vehicle events are separable from the strain signal.
Both at once
Dual-deck structures carrying rail and highway have been instrumented and studied for decades, and the same method holds across every combination of structure type and traffic type.
Optical sensing is why the signal stays clean
Rail bridges are one of the least friendly electrical environments a sensor can sit in: traction current, overhead lines and radio traffic all around the structure. Fibre optic transduction has no electrical pickup at the sensor, so this is a property of the measurement chain rather than a mitigation applied per bridge. We already run optical in production rather than planning to.
Interference
No electrical signal at the sensor and no metallic loop, so traction current and radio traffic have nothing to couple into.
Noise
Modal extraction and filtering run on the bridge edge device. Only modal parameters, influence-line signatures and health flags leave the structure, so noise is handled before the wireless link is touched.
Transmission and timing
Store-and-forward in firmware raises effective data delivery to 99 percent or better across operational windows. Time alignment across units uses GPS-disciplined clocks, with precision time protocol inside cavities where GPS cannot reach.
What we bring, and what we deliberately do not
What we bring
- Algorithms, not hardware. Modal analysis, Kalman filtering, spectral methods and anomaly detection, already deployed and running.
- A sensor-reduction analysis that produces a defensible minimum sensor count and optimal placement, per structural type.
- Sensor-agnostic processing. Fibre Bragg grating, MEMS, piezo and vibrating-wire data all fit.
- A first engagement that needs no field work at all: you provide data from structures you have already instrumented.
What we do not bring
- No sensors, no interrogators and no installation. You deploy your own instrumentation and we process the data.
- No replacement for your inspection regime. We measure what inspection cannot see, alongside it.
- No verdict on whether a structure is safe. That stays with the people who inspect and certify it.
- Methods that need their own separate sensor network, such as guided-wave or acoustic emission for tendon fracture, sit outside a low-cable scope, and we say so rather than bundling them in.
Data only, across your structural types
No hardware, no field installation and no procurement. You provide data from bridges that are already instrumented, and we deliver the sensor-reduction analysis and what generalises from it.
Data scoping
Confirm the structures, one per structural type where possible. Lock data formats, sample rates, channel inventories and metadata.
Ingestion and baseline
Ingest into the FleetSense stack. Run modal identification on the full sensor set per structure, extract the modal basis, capture thermal and seasonal effects.
Sensor-reduction analysis
Iteratively remove sensors and reconstruct via modal expansion. Output per structure: minimum sensor count, optimal placement, retention scores.
Generalisation
Compare the reductions. Separate what generalises by structural type from what is specific to one bridge, and feed that into your portfolio planning.
Start with data you have already collected.
If you have instrumented structures and a monitoring budget you are trying to defend, the first engagement needs nothing new in the field. Send us what those bridges already produce and we will tell you what could be measured with fewer sensors, and what could not.
