What on-site commissioning time is actually spent on
If you instrument a conventional commissioning and categorise every hour, the distribution is remarkably consistent across projects. Roughly a fifth goes to mechanical fitment and alignment — genuinely physical work that no simulation avoids. Another fifth goes to process tuning: weld parameters, adhesive flow, grip force, the things governed by material behaviour. The remaining three-fifths goes to logic. Interlocks that deadlock. Sequences that work in isolation and fail when the upstream buffer is full. Fault recovery paths that nobody exercised because exercising them meant deliberately breaking a running line.
That last category is what virtual commissioning removes, and it is why the savings are large but bounded. A twin cannot align a fixture or find your optimal weld current. It can, and reliably does, find the deadlock.
The three failure modes it catches
In our own projects, faults found during virtual commissioning cluster into three groups.
- Handshake errors between stations. Station A releases before Station B has confirmed clamp. In simulation this appears within minutes of running the sequence. On site it appears the first time the line runs at rate with a real part.
- Recovery paths that were never written. The happy path works. What happens when an operator opens a guard mid-cycle with a part half-transferred is frequently undefined until someone does it.
- Buffer and blocking logic. Any line with a buffer has states where it can lock itself. Discrete-event simulation finds these deterministically; on-site testing finds them by accident, at 2 a.m.
What it does not save
It is worth being blunt about this, because overselling virtual commissioning is how the practice loses credibility with plant managers who were promised too much.
Mechanical fitment is unaffected. Process parameter development is unaffected — if anything a twin can mislead here, because a simulated weld looks perfect. Sensor commissioning, alignment and calibration all still take the time they take. And building the twin itself costs engineering hours: typically 8 to 15 percent of the control engineering budget for a line of moderate complexity.
When the economics work
The decision comes down to what an hour of on-site time costs relative to an hour of desk time. Three conditions push it strongly positive:
- The line is being installed into a plant that is running, so on-site time is constrained to shutdown windows that cannot be extended.
- The control logic has genuine complexity — multiple stations, buffers, shared resources, or a robot coordinating with a machine it does not own.
- The equipment is being shipped abroad, where an extra week of commissioning carries travel, accommodation and visa cost per engineer.
Conversely, a standalone cell with simple sequencing, installed on a greenfield floor with unlimited access, rarely justifies the model. We say so when asked.
A practical starting point
The cheapest useful entry is controller-in-the-loop emulation of one station — no photorealistic scene, no full plant model. Connect the real PLC program to an emulated I/O model and exercise the sequence including faults. Most of the logic-error value is captured there, at a fraction of the cost of a full Omniverse twin. Scale up only when the line complexity justifies it.
Talk to us about this
If this is relevant to a line you are working on, our engineers are happy to look at the specifics. Related capability: Digital Twin. You can also browse delivered projects, the technology stack or contact the engineering team.