We read straight off the machine, without touching its program.
To run services on real hours, you have to know how long each machine worked. Where the machine can report on its own, we connect it through the usual industrial protocols. Where it cannot, we fit sensors.
How we connect the machine
Industrial protocols
An agent installed in the shop-floor network reads Modbus TCP and OPC-UA, without writing anything back. The machine runs exactly as before; we do not change its program.
Every reading, along the same path
All readings enter the same message bus, with a versioned schema. When you add a new machine, nothing that already reads from it has to be touched.
Sensors for the older machines
Machines with no digital interface get sensors that measure how they run: current draw, vibration, temperature, or simply whether they are running or standing.
If the link drops
The agent on the floor holds the readings and sends them when the link comes back. Not an hour is lost off the counter.
What we read
- The hours actually worked and how many cycles it has done
- When it started, when it stopped and how long it stood, to the minute
- Temperatures, pressures, flows and consumption, where the machine gives them
- Threshold breaches, announced the moment they happen
- Whether the machine is running right now or standing
It also works without a connection
If you do not want to connect the machines in the first stage, the platform runs in full on time-interval services and on the hours recorded from production. Connection can be added later, without redoing the setup.
What the sensor reads goes straight to the model.
The hours, the cycles and the values taken off the machine are the raw material of prediction. Nobody keys them in by hand and no time is lost on the way.
Actual hours, not estimated
The counter read off the machine takes the place of a guess from the calendar, and services and estimates start from how much it really ran.
Deviations become signal
When a value leaves the normal range, the platform notes it as an event and puts it into the risk calculation for that machine.
Continuous history
Reading without a break builds the run of data the model uses to tell normal wear from a real departure.