From raw sensor data to an operational decision.
We combine three disciplines that rarely exist under one roof: AI and time-series intelligence, optical and multi-vendor sensing, and structural dynamics. That combination is the edge, and it is the reason the same company can work on a blade, a bridge and a production line.
AI and Time-Series Intelligence
The core of what we do is turning streams of raw sensor data into a clear statement about the condition and likely future behaviour of an asset, with an honest measure of how confident that statement is.
- Machine learning for anomaly detection and early degradation identification
- Time-series analysis across many sensor channels at once, at different sample rates
- Uncertainty quantification, so a model states how confident you should be rather than implying certainty
- Predictive algorithms for fault progression and for the timing of an intervention
- Agents that express asset health in plain language, so a non-specialist can act on it
- Physics-informed machine learning that understands the mechanics underneath the data
- Bayesian inference for real-time probabilistic state estimation
Optical and Multi-Vendor Sensing
Demanding environments need sensors that survive them, avoid electromagnetic interference, and still deliver signal quality good enough for analysis. Inside a blade, on a rail bridge, or in a factory, the constraint is the same even though the setting is not.
- Photonic vibration and strain sensing, very high sensitivity, full immunity to electromagnetic interference
- Multi-axis measurement: vibration, strain, deflection and temperature
- Multi-vendor integration. The processing is hardware agnostic, so existing instrumentation can be used rather than replaced
- Edge computing architectures for data collection where connectivity is limited or interrupted
- Secure and ruggedised data pipelines for critical infrastructure
- Field deployment experience in harsh environments, offshore and onshore
Structural Dynamics and Digital Twins
We bring the physics. Understanding how a structure actually behaves, under wave loading, traffic, fatigue cycling or thermal stress, is what separates a meaningful insight from a pattern in noisy data.
- Structural dynamics modelling for complex assets, from blades to spans to machines
- Operational modal analysis and modal expansion, so a sparse sensor set carries the information of a dense one
- Fatigue and load analysis for critical components
- Digital twin development that updates from real sensor measurement rather than staying a design-time artefact
- Condition-based maintenance frameworks grounded in how the asset actually behaves
- Alignment with certification frameworks where an asset class has them
AI anchored in physics
Most industrial AI needs many examples of things going wrong before it can recognise the next one. Valuable machines and structures do not oblige: the events that matter most are, by definition, rare.
So we invert it. The physics supplies the fundamental behaviour and your data is asked only for what is specific to your situation. Less data is needed, and what comes out is more trustworthy.
Physics for the fundamentals. Your data for the specifics.
Structural dynamics, load paths, thermal behaviour. What is possible at all.
Sensors, control systems, environment, maintenance history. What is true here.
Realistic outside the measured range, explainable in engineering terms, and available from day one rather than after years of history.
Little failure data
The events that matter most are rare by definition. Physics fills the gap that missing examples leave behind.
Reliable beyond the data
Predictions stay realistic outside the range you happened to measure, because the physics constrains what can occur.
Explainable, not a black box
Results are grounded in physical terms, so your own engineers can check them, repeat them and defend them.
Value from day one
Monitoring, prediction and early warning without first building years of history you do not have.
FleetSense
FleetSense is the sensing and data platform underneath our structural monitoring work. It ingests sparse sensor data from large, dynamically excited structures, extracts modal parameters, fatigue indicators and change signatures on the device itself, and delivers clean outputs rather than raw waveforms. Buffering and store-and-forward run in firmware, so an intermittent link costs data delivery almost nothing. It is shipping code on live assets, not a pilot-phase build.
The same algorithms carry across asset classes. When we take on a new domain, the domain is new. The code is not.
From the sensor on the asset to the decision on the screen
Every piece is built to work with the ones either side of it. One party across the whole chain means no handover points between suppliers, which is usually where the data quality quietly goes.
Sensors
Multi-vendor sensor arrays measuring vibration, strain, motion, temperature and environmental conditions. Yours or ours.
Edge processing
Local collection with pre-processing on the device: filtering, feature extraction and quality checks, working through intermittent connectivity.
Secure data pipeline
Validated pipelines that ingest, clean and enrich time-series data, handling multi-channel and multi-rate feeds without silently losing any of it.
Measurement health
A judgement on whether the data can be believed at all, channel by channel, before anything is concluded from it.
Models
Anomaly detection, modal analysis, physics-informed degradation models and uncertainty-aware remaining life estimation.
Decision support
Dashboards, alerts and plain-language condition summaries that give the operator clear guidance with a confidence level attached.
Which capability powers what, and where
Seven offers across four industries can look like a list. It is not. Each one draws on the same three capabilities, in a different proportion, on a different asset.
| Solution | Sector | What the physics contributes | What the AI contributes |
|---|---|---|---|
| Blade Monitoring | Wind | Modal behaviour of a composite structure, and how it should differ between three blades that are nominally identical | Anomaly detection across channels, with operating point removed so that a change in behaviour is not confused with a change in conditions |
| BladeTrace | Wind | The sibling comparison, which is why no measurement history is needed to establish a reference | Scoring at consistent rotor speed, and attribution of a shock event to a phase of the handling operation |
| MOOR-LIFE | Wind | Fatigue and corrosion mechanics, turning platform motion into the loads a mooring line has actually carried | Bayesian inference on degradation, and remaining life with a stated uncertainty band |
| ZeeCAIbel | Wind | Electromagnetic field behaviour around a buried cable, given type, depth and orientation | Learned correction of systematic measurement error, generalising beyond the survey conditions it was trained on |
| Bridge Monitoring | Civil | Modal basis and load paths of a span, which is what lets a sparse sensor set reconstruct the response everywhere | Automated mode selection, influence-line extraction from traffic, and ranking across a portfolio |
| Tarucca Edge | Manufacturing | What a measurement should look like given speed, load and the sensors around it. This is what separates a sensor fault from a machine fault | Drift detection that finds slow degradation no fixed threshold will ever see |
| Plant Data Unlock | Manufacturing | The mechanism test. Without a physical explanation there is no conclusion, only a correlation | Pattern discovery across production, maintenance and process data, validated on periods the model has never seen |
| Defence applications | Defence | The same structural and measurement physics, applied where access is constrained and data is sparse by design | Edge computation under intermittent or denied links, and fleet-level prioritisation |
Read down the last two columns and the pattern is the point: it is one method, applied to eight problems, rather than eight products that happen to share a logo.
Curious whether this applies to your asset?
The quickest way to find out is to describe it. If our methods do not fit, we would rather tell you that in one conversation than find out three months into a project.
