Digital twin & BIM
A model of an asset is not worth what it cost to build. It is worth whether it still designates the same things as the site does, three years and two renovations later. That is where a twin holds or falls apart, long before anyone runs a simulation on it.
A model is only worth the reference data feeding it.
Projects rarely fail because the simulation was wrong. They fail because the model and the field stopped designating the same equipment, and nobody noticed until a decision was taken on the wrong one.
Our work therefore starts with the object reference: what is modelled, at what granularity, and under which identifier that survives a replacement. We then measure the gap with what was actually built rather than assuming it, and only then do we close the loop shown below. A twin that observes is a dashboard; a twin that steers back is a control system, and it carries the rules of one.
What we do
Four workstreams, from the object reference to the loop that steers back to the asset.
Build the object reference
What is modelled, at what granularity, and under which identifier. The functional tag survives the three replacements the equipment will go through; the serial number does not, and neither does a name matched by hand.
two services naming one machine differently cannot share a model
- One identifier that survives a replacement
- What is not modelled, written down
- IFC to CMMS correspondence maintained, never guessed
The model, and the gap with the site
Federated models to ISO 19650, clash detection between disciplines, and a point cloud survey confronted with the record drawings. On a fifteen-year-old site the gap is not an anomaly, it is the normal state.
an object known to be wrong beats a wrong object nobody flagged
- Federated IFC models and clash reports
- Survey confronted with the record drawings
- What cannot be surveyed marked as uncertain
Keep the model alive
Every intervention that changes the geometry updates the model before the work order closes. Without that rule an as-built model is wrong again within eighteen months, and everyone goes back to the plans on the wall.
an as-built handed over and never updated is a photograph
- Update tied to closing the work order
- Telemetry, maintenance history and geometry joined
- Handover to the CMMS, not to a viewer
Close the loop, and hold it
Simulating scenarios before committing an intervention, and writing what may go back down to the asset. A loop that writes to the asset is a path to the asset: separated write direction, bounds set on the controller, every command signed.
a simulation whose assumptions are lost can no longer be contested
- Scenarios ranked with their assumptions written
- Nothing automatic on a safety device
- Bounds enforced on the controller, not in the model
The loop, from the field back to the field
A closed loop between the physical and the digital
Sensor data feeds a living model; simulation and optimisation turn it into decisions that act on the physical asset.
What you get
One project runs through the four deliverables below: taking over an existing industrial site. Each line states what is actually handed over, in the order it is handed over.
An object reference
What is modelled, what is not, and the identifier that survives a replacement. Written before the model, because a correspondence guessed from names works for four objects then produces false matches.
The measured gap with what was built
A point cloud survey confronted with the record drawings, corrections made where they are worth making, and what cannot be surveyed marked as uncertain rather than quietly kept as true.
The rules of the return loop
What may go back down to the asset and what never may, with the write direction separated from the read direction and bounds enforced on the controller rather than in the model.
A confrontation after six months
Predicted against actual consumption holds at 3.1% across the site, and reaches 19% on the one unit whose maintenance history was never digitised. The model answered with what it was given.
How we deliver
Discover
depending on scope, sector and the level of compliance required
- Audit of use cases and pain points
- Value / feasibility matrix
- Executable specification (ASDD)
MVP
depending on system complexity and integrations
- An agent in a real environment
- Generated tests, measured coverage
- Go / no-go before industrialisation
Scale
depending on the number of agents and connected systems
- Multi-agent orchestration on a shared foundation
- CI/CD and MLOps integration
- Team upskilling
Run
service commitment defined with you
- LLMOps observability
- FinOps optimisation of AI costs
- Continuous compliance audit
Representations that hold up in operations
A critical system mapped before anyone touched it
82 services instrumented · 12 TB of traces a day
A programme carrying national rail traffic, where nobody held a complete picture of how the system actually behaved, and where the cost of finding out by changing something was measured in cancelled trains.
An architecture audit across the data, application and infrastructure layers, then instrumentation of the eighty-two services so the real behaviour could be observed rather than assumed, with root cause analysis moving from days to hours.
A picture of the network that stays current
42 dashboards · data under 15 minutes old
Eight regional divisions steering field workload, incidents and service levels on a national network, where a monthly report describes a situation that has already changed twice.
A data architecture feeding forty-two dashboards refreshed every quarter of an hour, so managers arbitrate on the current state rather than on last month's, with the practice handed over to more than a hundred and fifty of them.
Insights & Perspectives
From black box to blueprint: a rapid legacy system assessment
How a fifteen-year-old application of 2.2 million lines was assessed in a month, to know what it actually contained before taking over its support.
How to build and use a product map inside your organisation
A six-step method for mapping what exists: scoping, customer value, inductive and deductive mapping, then rollout and governance.
Guiding the unknown: a compass for complex initiatives
A three-stage framework and a compass of questions to align strategy and execution, and to test a tangible first version before committing.
Represent the asset before committing anything physical
An object reference that survives replacements, a measured gap with the site, a model kept alive by the work orders, and a loop whose rules are written.
Frequently asked questions
With the object reference, not with the model. If maintenance calls a machine by its functional tag and the model calls it by its serial number, the two cannot be joined, and every later layer inherits that break. The identifier chosen has to survive the replacements the equipment will go through over the site's life.
It is what lets an operator find the right object without going down to the plant room. A federated model to ISO 19650, handed over as structured IFC data into the maintenance system rather than as a viewer, is what turns a design deliverable into something operations actually open.
No, and it is the normal state of an existing site. The gap is surveyed rather than assumed, corrections are made where they are worth making, and what cannot be surveyed is marked as uncertain. An object known to be wrong is more useful than a wrong object nobody has flagged.
By tying its update to closing the work order. Any intervention that changes the geometry updates the model before the order can be closed. Without that rule the model is wrong again within eighteen months, and the teams go back to the drawings on the wall, which is where most twins quietly die.
Within written bounds, and never on a safety device. A loop that writes to the asset is a path to the asset, so the write direction is separated from the read direction, the value range is enforced on the controller rather than in the model, and every command is signed and replayable.
Its assumptions being written down, with their date and their author. A scenario compared without them cannot be contested, and something that cannot be contested stops being used to decide. What we hand over is a ranking of scenarios with their assumptions, not a single recommended answer.
Framing takes 2 to 6 weeks depending on scope, sector and the level of compliance required, and produces the object reference, the modelling scope and the measured gap with the site. A first perimeter connected to live data, with one scenario simulated then executed, follows in 4 to 10 weeks.