A functional digital twin is a digital representation of a piece of equipment, in the form of a navigable graph. It connects its components, its diagrams, its failure histories and its documentation. It is used to diagnose failures faster and to retain field know-how, drawing on the plans and manuals already in place, with no sensors and no 3D model.
When a line stops, the technician first tries to understand the machine before understanding the failure. This knowledge is often scattered across a folder of diagrams, a CMMS filled in hastily and the memory of an expert close to retirement. Platforms such as Mimorian start precisely from this raw material to build a functional digital twin. Mimorian models equipment and supports technicians in their diagnostic reasoning through a multi-agent AI architecture. This guide explains what this twin covers, how it differs from the visual twin and the sensor-based twin, how it is built, and where to start.
What is a functional digital twin in maintenance?
A functional digital twin is a model that describes how a piece of equipment works: its components, the links between them, their functions and their failure modes, connected in a graph. It does not reproduce the shape of the machine, it reproduces its logic. The principle is simple: when you understand how a machine works, you understand what can make it break down.
This twin differs from three objects with which it is often confused. It is not a 3D model, which shows the geometry of the parts. It is not a sensor feed, which shows real-time curves. Nor is it a flat documentary database, where manuals pile up with no links between them. The functional twin connects these elements: a reference mark on an electrical diagram points to the real component, which points to its function, to its past failures and to the documentation page that describes it.
This graph structure makes the machine queryable. Instead of searching through a folder, the technician follows the cause-and-effect links, as on a map rather than in a book index.
Take a common fault: a fuse that blows every three weeks. On a flat documentary database, you find the fuse record and replace it, again and again. On a functional twin, you follow the links: this fuse protects a motor, which drives a conveyor. When a bearing wears, the motor strains and the fuse gives way. The graph then points to the bearing, the real cause, not to the symptom.
How does it differ from the visual digital twin (CAD/3D) and the sensor-based twin?
The three approaches answer different questions. The functional twin answers "how does it work and why does it break down". The visual twin answers "what does it look like". The sensor-based twin answers "how is it behaving right now". None replaces the others, they complement each other.
| Criterion | Functional twin | Visual twin (CAD / 3D) | Sensor-based twin (IoT) |
|---|---|---|---|
| Question it answers | How the equipment works and why it breaks down | What the equipment looks like in space | How the equipment behaves in real time |
| Raw material | Diagrams, documentation, failure histories | CAD model, 3D scans, geometric plans | Installed sensors, measurements, continuous data feed |
| What it shows | The links between components, functions and failures | The geometry and layout of the parts | Curves, thresholds, alerts |
| Initial investment | Your existing plans and manuals | CAD modelling or 3D digitisation | Sensor installation, network, IoT project |
| Main use in maintenance | Guided diagnosis and knowledge capture | Design, training, visualisation | Condition monitoring and predictive maintenance |
The international standard ISO 23247 frames the digital twin for manufacturing as a digital representation of observable manufacturing elements: equipment, materials, processes. Condition monitoring through sensors falls under another framework, ISO 17359, which sets out the guidelines for condition monitoring and diagnostics. These two angles, the visual and the sensor, dominate online research on the subject, even though they require a heavy investment before the first result.
A functional digital twin does not predict failure from signals. It helps to understand it from the structure of the machine.
The sensor route remains demanding for a measured gain. Among the manufacturers that have deployed advanced predictive maintenance, 60% report better availability, with an average gain of at least 9% (PwC and Mainnovation, 2018). This result comes at the cost of a sensor and data project that is often long to install. The functional twin, by contrast, starts with what you already have at hand.
How is the graph built from the diagrams and manuals?
The graph is built from the electrical, pneumatic and hydraulic diagrams of the equipment. The AI parses these plans, extracts each component and each link, then cross-references everything with the machine documentation and the intervention histories. The result is a model where each element is connected to its neighbours and to its known failures. For the detail of this extraction from the plans, read our article on modelling an industrial piece of equipment with AI. For the way the graph connects the electrical, pneumatic and hydraulic, read our article on the relational graph of an electrical diagram.
In practice, three layers come together. The diagrams provide the structure, the backbone of the graph. The documentation and procedures feed a knowledge base that the AI consults through augmented retrieval. The failure histories add the memory of incidents already seen. The strength of the model comes from the back-and-forth between these layers: the AI tests a hypothesis on the graph, checks it against the documentation, returns to the graph to refine it. The CMMS history enriches this reasoning, it does not condition it: the diagnosis starts as soon as the diagrams and documentation are extracted. It is specialised agents that exploit this graph, a multi-agent AI architecture where each one has a precise mission.
The human stays in control at every step. They validate the links in the graph, correct an extraction, add a component by voice. The model aims for rigour, the team remains in charge of the input data.
Why does a functional twin make guided diagnosis possible?
Guided diagnosis needs a model that connects symptoms to causes. This is exactly what the graph provides: it knows the components, their functions and their links, so it can trace back from a visible fault to its possible causes, in order of probability.
In the field, the technician describes what they observe. The twin proposes targeted hypotheses according to the context and guides them towards the test that confirms or rules out each one. It gives the exact reference mark and component number to check on the diagram, with no back-and-forth to the folder. It distinguishes the immediate cause, identified at the moment of resolution, from the root cause, traced back through analysis. This logic rests on the structure of the machine, not on a statistical prediction drawn from past failures. To go further on this mechanism, read our guide to AI-guided diagnosis in maintenance.
The gain is direct: a junior supported by the graph reasons like a veteran, because they rely on the same map of the machine.
How does it capture field know-how?
Every diagnosis carried out on the twin leaves a clean, reusable trace. When the technician closes an intervention, the report is built from their voice dictation, linked to the component and the cause concerned in the graph. The knowledge no longer stays in a single head, it joins the memory of the plant.
The stakes are high. Inefficient sharing of internal knowledge costs a large American company around 47 million dollars a year (Panopto, 2018). The skills gap could leave up to 2.4 million industrial jobs unfilled in the United States between 2018 and 2028 (Deloitte and The Manufacturing Institute, 2018). Every retirement takes away years of knowledge of the machines. The functional twin turns this knowledge into collective, queryable know-how that serves the next diagnosis. For the full framework, see our guide to knowledge capture in maintenance.
This memory also feeds the reliability engineers and the methods teams: structured closures instead of "part replaced, machine restarted", and recurring weak points finally visible by equipment and by cause. It also serves training. Learning sheets are built from real diagnoses, and a newcomer upskills by following the reasoning of their elders rather than searching on their own.
Do you need sensors or an IoT project to start?
No. A functional digital twin is built from your diagrams and your documentation, with no sensor installation and no prior IoT project. This is its main difference from the sensor approach and from the sensors and IoT or SCADA approach, which require instrumenting the machines before getting the first result.
This point changes the start-up. The pilot runs in a closed loop, in parallel with your information system: an export is enough, the IT connection comes afterwards, at your own pace. The first value lands within a few weeks, on real failures, because the modelling is done in advance by the vendor. Sensors keep their full place for condition monitoring, and the functional twin complements this data rather than replacing it. For the detailed comparison between the functional twin and sensor-based predictive maintenance, read our article digital twin without sensors versus predictive maintenance.
The functional twin also complements the CMMS. The CMMS plans and tracks interventions, the twin brings the diagnostic reasoning and the memory of failures. The two work side by side.
Frequently asked questions
How long does it take to build a functional digital twin?
Allow a few weeks for a first scope. You isolate a few problematic pieces of equipment, extract their diagrams, capture the documentation and history, then guided diagnosis starts. The knowledge capture is done in advance by the vendor, the team gains value from the first intervention rather than after months of invisible set-up.
Does the functional digital twin replace the CMMS?
No, it complements it. The CMMS handles planning, work orders and administrative traceability. The functional twin brings the modelling of the machine, guided diagnosis and the memory of failures. The twin sits alongside your CMMS, not in its place, and enriches its history with structured reports.
Which diagrams and documents do you need to provide?
The electrical, pneumatic and hydraulic diagrams of the equipment form the backbone of the graph. The machine documentation, the manufacturer manuals, the procedures and the intervention histories enrich it. The richer the context provided, the more relevant the reasoning, and the work starts as soon as the diagrams and documentation are available, without waiting for a perfect history.
Conclusion
Three key points to remember.
- The functional digital twin models the logic of the equipment, not its shape or its signals: it connects components, diagrams, functions and failures in a navigable graph.
- It starts with your plans and your manuals, with no sensors and no IoT project: the first value lands within a few weeks, on real failures.
- It serves two uses at once: guided diagnosis for the field and knowledge capture for the whole plant, alongside the CMMS and the existing sensors.
Next step for a maintenance manager: choose two or three pieces of equipment that break down often, gather their diagrams, and test the modelling on this scope before extending it.
To go further, read our guide to AI-guided diagnosis in maintenance or our guide to knowledge capture in maintenance.
📚 Sources :
- ISO 23247-1:2021 : Automation systems and integration, Digital twin framework for manufacturing, Part 1: Overview and general principles
- ISO 17359:2018 : Condition monitoring and diagnostics of machines, General guidelines
- PwC et Mainnovation, 2018 : Predictive Maintenance 4.0, Beyond the Hype
- Panopto, 2018 : Workplace Knowledge and Productivity Report
- Deloitte et The Manufacturing Institute, 2018 : Skills Gap and Future of Work Study