What makes this twin different from a plant twin
Facility digital twins are well established, and so is computational-fluid-dynamics dispersion modelling. Both normally begin from a release point that somebody assumed or typed in - a scenario, not an observation.
This twin begins from a source position the system itself physically measured, by rotating a sensing head and reading a peak sector. That single difference is what changes the output from a simulation of a hypothetical to a forecast about the event actually in progress.
Three inputs, loaded at different rates
The twin is assembled from three things, and they change on completely different timescales - which is why they are separate inputs rather than one model.
- A facility mesh — floor extents, racking, plant blocks — loaded once and rarely changed.
- A registration of node positions and a named asset register, both in the model's own coordinate frame, updated when the plant is re-laid-out.
- The live measurement state streaming from every installed node, changing every ten seconds.
Why the growth model is deliberately lightweight
Spread is computed as a set of expanding radii at successive horizons — thirty, sixty and one hundred twenty seconds in the working embodiment — using a closed-form radial growth model rather than a fluid simulation.
That is a deliberate choice, not a shortcut taken for want of compute. The available input is one bearing and one magnitude, carrying a confidence figure that is honest about how loosely the range is constrained. A heavyweight dispersion model fed by that input would produce an output whose apparent precision the underlying measurement cannot support - a plausible-looking concentration field that is mostly an artefact of the solver.
A coarse forecast that is honest about being coarse is more useful on a shift than a fine one that is quietly wrong.
The intersection that produces the answer
Each forecast radius is intersected with the facility's asset register, and the assets and areas falling inside it are named. The output of the system is therefore a list of physical places, at each horizon, that a person can act on without further interpretation.
This is the step where sensor units stop being sensor units. Every stage before it - the channels, the genome, the bearing, the coordinate - exists to make this list possible, and the specification submits that translating a measurement into named threatened places is a capability absent from every prior-art category it surveys.
Footprints, and a site learning its own hotspots
Every located source is stored as a footprint: its bearing, its peak magnitude, its estimated coordinate, and the complete angular sweep profile that produced it.
Footprints accumulate. A corner that repeatedly leaks and a machine that repeatedly overheats stop being separate incidents and start being a pattern the site knows about - which is a different and more valuable thing than handling each occurrence afresh.
