Robot Software
Digital Twins for Robot Cells: What They Actually Save
A digital twin pays through commissioning time, not through simulation glamour. Here is where the hours actually come from and the two levels most projects should stop at.
The saving is commissioning time. A validated virtual model of a robot cell typically removes 30 % to 60 % of on-site commissioning, which on a cell with three weeks of planned commissioning is 5 to 9 working days of a production line not running. Everything beyond that, live synchronisation and predictive maintenance, is a separate business case that most cells never need.
Four levels, and where to stop
| Level | What it is | Payback | Effort |
|---|---|---|---|
| 1 | Geometric model: reach, collision, layout | Prevents layout mistakes before steel is cut | 1 to 3 days |
| 2 | Kinematic and program model: real controller motion, cycle time | Programs written and validated before delivery | 1 to 3 weeks |
| 3 | Virtual commissioning: emulated PLC and I/O against the model | Control logic debugged without hardware | 3 to 8 weeks |
| 4 | Live twin: synchronised with the running cell | Monitoring, what-if analysis, predictive maintenance | ongoing |
Levels 1 and 2 are where the return is clear and the effort is bounded, and they are what vendor offline programming tools deliver as standard. Level 3 pays on complex cells with substantial control logic, particularly where several machines interlock. Level 4 is an operations investment that needs its own justification, and cells bought as one-off installations rarely produce it.
Where the commissioning hours actually go
Understanding what a twin removes requires knowing what commissioning consists of. On a typical machine tending cell with 15 planned days on site, the distribution looks roughly like this.
| Activity | Without | With | Why it changes |
|---|---|---|---|
| Mechanical install and alignment | 2.0 d | 2.0 d | Physical work, unaffected |
| Electrical and network | 1.5 d | 1.5 d | Physical work, unaffected |
| Robot program creation | 4.0 d | 1.0 d | Written offline, only touch-up remains |
| PLC and interlock debugging | 3.5 d | 1.0 d | Logic already exercised against the model |
| Cycle time optimisation | 2.0 d | 0.5 d | Path already optimised offline |
| Fault and recovery testing | 1.5 d | 0.5 d | Edge cases exercised virtually |
| Safety validation | 0.5 d | 0.5 d | Must be measured on hardware |
| Total | 15.0 d | 7.0 d | 53 % reduction |
Note which rows do not move. Mechanical, electrical and safety validation are physical work, and no model removes them. The savings concentrate entirely in software and logic, which is exactly where a twin can substitute for hardware.
Who should own the model
Ownership decides whether a twin survives its first year. Three arrangements are common and they produce very different outcomes.
- Integrator owns it, hands over a copy. Cheapest, and the copy diverges within months because nobody on site can update it. Suitable for levels 1 and 2 only.
- Operator owns it, integrator delivers into it. The model becomes a plant asset with a maintainer, which is the only arrangement in which level 3 or 4 pays.
- Nobody owns it. The common default, and the reason so many organisations have a folder of stale simulation files.
A practical test: name the person who will update the model when a fixture moves. If that name does not exist, build level 1 or 2 and stop there.
The honest cost side
- Software licences for offline programming and virtual commissioning are a real annual cost, and per-seat pricing means the tool is often available to one engineer rather than the team.
- Model building. Machine models, fixture geometry and I/O mapping have to be created and maintained. A cell whose fixture changed but whose model did not is a trap.
- Skills. Virtual commissioning is a distinct competence. A team that does it twice a year will be slow at it every time.
- Maintenance. A twin that is not updated diverges from the cell and quietly becomes misinformation.
Those costs are why level 4 rarely pays for a single cell and often does pay for a line of twenty similar cells, where the model is built once and the commissioning saving multiplies.
Frequently asked questions
What does a digital twin of a robot cell actually save?
Commissioning time, typically 30 % to 60 % of the on-site days. The savings concentrate in robot programming, control logic debugging and cycle optimisation. Mechanical installation, wiring and safety validation are unaffected.
Which level of twin should a project aim for?
Levels 1 and 2, geometric and kinematic modelling, for almost every cell, since vendor offline programming tools deliver them. Level 3 virtual commissioning pays on cells with substantial interlocked control logic. Level 4 live synchronisation needs its own business case.
Why do twin projects disappoint?
Usually because the model was never calibrated against the physical cell, so it produces confident wrong answers, or because it was not maintained after a fixture or program change and quietly diverged.
Can a twin replace safety validation?
No. Force, pressure and stopping performance have to be measured on the real hardware with calibrated instruments. A model can plan the validation and predict problem areas, but it cannot substitute for the measurement.
When is a live twin worth it?
When the model is reused across many similar cells, or where the operation genuinely acts on the data through what-if scheduling or condition monitoring. For a single one-off cell the ongoing maintenance usually exceeds the benefit.
Sources
- ISO 9283, manipulating industrial robots, performance criteria and test methodsDefines the accuracy terms that determine how well a model can match a real robot
- ISO 10218-1:2025, Robotics, Safety requirements, Part 1International Organization for Standardization, validation obligations that remain physical
- Robotics at NISTNational Institute of Standards and Technology, robot performance measurement and test methods