DIGITAL TWIN
Testing a change on the live line costs a shift of production
Every significant question about a production line — reschedule the sequence, add a machine, change a setpoint, reroute around a stoppage — has an answer that costs real output to find out. A digital twin is simply a model of the line accurate enough to answer some of those questions without running the experiment. The qualifier matters: a twin is only useful for the questions it was validated against, and vendors who imply otherwise are selling a visualisation.
What the system does
Scoped to the questions it must answer
We agree up front which decisions the twin exists to support, and build to that. A twin scoped to answer scheduling questions well is worth more than a general one that answers everything approximately.
Physics and process logic, not animation
The model encodes the real behaviour — cycle times, buffer capacities, changeover rules, failure and repair distributions, material flow constraints. Visual representation is useful for communication but it is not what makes a twin valid.
Fed by live plant data
The twin is parameterised from the MES record of what the line actually does, and re-parameterised as that changes. A twin built once from design specification diverges from reality within months.
Validated before it is trusted
We validate the twin’s output against measured production over a period you choose, and publish the discrepancy. Until it reproduces history within an agreed tolerance, it does not get used to make decisions.
Bottleneck and scenario analysis
Where the constraint really sits under different mixes, what a buffer change would do, what happens to throughput when a machine is down. Questions that are expensive to test on the line and cheap to test on the model.
Training without consequences
New operators and planners can see the effect of decisions on a model of their own line. Mistakes made there cost nothing, which makes it a far better teacher than a line that cannot be allowed to fail.
How a digital twin project runs
Define the decisions
Which questions must this answer, and to what accuracy for the answer to be actionable? This single step decides the scope, the cost and whether the project is worth doing at all.
Characterise the line
Measured cycle times, changeover durations, buffer capacities, failure and repair distributions, routing rules. Where the MES is already in place much of this is available; where it is not, this phase is also an instrumentation phase.
Build and validate
The model is built and then run against a historical period and compared with what actually happened. We report where it matches and where it does not, and we are explicit about which questions the discrepancy makes it unfit to answer.
Operate and maintain
The twin is kept current as the line changes, and re-validated periodically. Like the MES, it is maintained under the monthly service agreement rather than delivered once and left to drift.
What this is built on
Discrete-event simulation for line and flow behaviour, with process-physics models where the step is continuous rather than discrete. Parameterised from MES production history so the model reflects measured behaviour instead of nameplate specification. Scenario runs are accessible through a web interface alongside the MES, so planners and process engineers can test a question themselves rather than filing a request with us.
What would you test if testing were free?
Tell us the change you have been putting off because you cannot afford to try it on the line. That question is the right starting point for deciding whether a twin is worth building — and sometimes the honest answer is that a simpler analysis would do.