TwinEdge / Platform · Working out what is wrongPlatform overview
Working out what is wrong
Explain the change behind the trend.
Combine physical relationships and operating history to identify anomalies, efficiency loss, and developing equipment problems.
Four rules
- 01Physics first, learning second
- 02It runs where the equipment is
- 03A finding ends in a job, or it did not matter
- 04You can see the working
§ 01
What it looks for.
- 01Something is behaving unlike itself
Learns what normal looks like for each individual machine and flags the moment it stops, rather than waiting for a fixed limit that was set for a different pump ten years ago.
- 02The equations, run continuously
Efficiency, head, power, and losses worked out from first principles every few seconds and compared against the curve the machine was sold on.
- 03How long it has left
An estimate of remaining life for the parts that wear, so a repair can be planned into a window instead of arriving as an emergency.
- 04Work, drafted
A finding worth acting on becomes a proposed job with the readings, the likely cause, and the suggested repair already attached, for a planner to accept or reject.
- 05Energy you are paying for and not using
Plant running when nobody is there, machines fighting each other, and efficiency lost to wear, each with a number attached.
- 06Condition, without sending anybody to look
A continuous health score from the readings, which removes most of the walk-round inspections that exist only to confirm nothing has changed.
§ 02
How it decides what to tell you.
The failure mode of every analytics product is crying wolf. Set the sensitivity high and operators stop reading it within a fortnight; set it low and it misses the thing you bought it for.
The way out is not a cleverer threshold. It is context: knowing which machine this is, what it has done before, what was repaired last month, and what the equations say it should be doing right now.
- 01Physics first, learning second
If an equation can explain it, the equation is used, because an engineer can argue with an equation. Learned models get the part that genuinely resists description.
- 02It runs where the equipment is
The analysis happens on the box at the site, so a site with a poor connection is not a site with no analysis.
- 03A finding ends in a job, or it did not matter
Analysis that stops at a chart is a hobby. Everything here is built to end in work that somebody approved, or to be dismissed with a reason.
- 04You can see the working
Every number can be opened up: which readings, which equation, which assumptions, and how confident it is. Nothing is asserted.
“Not how many findings it produced. How many people acted on.”
The only measure that counts
Run it against a year you already lived through.
Give us twelve months of history from one site. We run it forward and show you what would have been raised, when, and how often it would have been wrong. You already know what actually happened, which makes it a real test rather than a demonstration.