Blog
Working notes.
Notes on modeling physical assets, handling industrial telemetry, and the parts of this problem that are harder than they look. Written by the people building the platform.
- · Data
Three inverters send 143. They mean three different things.
The same integer arrives from three manufacturers and means 14.3 amps, 14.3 amps and 1.43 amps. Nothing in the number tells you which — and a plausible wrong number is worse than an obviously wrong one.
4 min read - · Modeling
A specification is a better prior than a year of operating history
A model trained directly on how a plant has behaved learns its existing faults as normal, then defends them. Building the physical expectation first is what stops that.
4 min read - · Data
Resolution discarded on ingestion cannot be recovered at any price
Averaging telemetry as it arrives makes storage cheap and destroys the signal that identifies the fault. The five-minute question you cannot ask in two years is the one that matters.
3 min read - · Operations
An alarm is a threshold crossing. A finding is a conclusion.
Monitoring systems generate alarms faster than any team can act on them, so teams stop acting on them. The fix is not better thresholds — it is changing what the system is asked to produce.
3 min read
Start with how it works.
The architecture, the modeling approach, and where each part of the platform actually stands today.