Evaluated at the machine: only the result travels

Detailed measurement data from a spindle is large. Sending all of it across the plant network just to calculate one figure from it is wasteful — and if two computers ever calculate the same figure, the numbers no longer agree. The calculation had to run where the data is, exactly once, proven on a real machine.

The laboratory's Hanwha XD38II Swiss-type lathe with a Sinumerik 828D. A sensor measures the current the spindle draws, in full detail; the EdgeBox at the machine records it.

• The detailed measurement data, directly on the EdgeBox.
• A calculation defined once in Toleron and assigned to this machine.
• The assignment, sent from the central server to the EdgeBox.

The EdgeBox read the data locally and ran the same calculation that otherwise runs on the central server. A calculation runs only where it is assigned — so exactly one device produces each value, by design. Every result carries a running count, so a missing or doubled value would show up immediately.

• Every value arrived exactly once — no gaps, no duplicates, no corrupted data.
• The central server only stored the results; it calculated nothing of its own.
• The new value is available in the charts, live, like any other measurement.

The detailed raw data never left the machine — only the result did. That keeps load off the plant network and keeps the data where it was measured. The same approach later put a model for grinding-wheel condition on the AI box. Where a calculation runs is a setting, not a project.