Machine monitoring from zero: what you actually need
Published 09/29/2026
Most monitoring projects do not fail on technology. They fail because they collect values nobody uses to make a decision, and in the end nobody opens the dashboard.
Start from the question, not the sensor
Write down what you want to answer first — for example, "how many times did the line stop last night, and why each time" — then work backwards to the signals you need in order to answer it.
Where the signals can come from
If a PLC is already in place, values can usually be read straight from its registers over Modbus TCP without touching the existing program. On older machines with no controller at all, external sensors can pick up run status and count parts.
How often is often enough
Temperature every 5–10 seconds is more than enough. Run status and alarms should be recorded on every change rather than on a timer, otherwise short events disappear between samples.
Alerts people still read
Require a condition to persist before it raises an alert, and give alerts clear severity levels. A system that sends ten alerts a day is muted within a week.
All of this can start with a single machine. Prove that someone actually uses the data, then expand — far better than instrumenting the whole plant at once and discovering you measured the wrong thing.
