The first thing I do on a new shop floor isn't to install software. I walk the line with the owner and ask one question at each machine: how do you know what this one ran yesterday? Nine times out of ten the answer is an operator's memory and a paper log. The machine knows far more than the log does. It just has no way to tell anyone.
Most owners assume getting that data means buying new machines with screens and connectivity built in. For a plant full of perfectly good fifteen-year-old lathes and presses, that's a non-starter. Here's the good news from the field: you can put most older machines online for the cost of a few service visits, no replacement required.
What "machine data" really means on day one
Set the grand IIoT diagrams aside for a minute. At the start you want answers to three plain questions per machine: is it running right now, how long has it sat stopped, and how many parts has it made this shift. Those three signals already expose most of the hidden losses on a floor, and you can capture all of them without opening the machine's control panel.
The three ways I usually capture it
1. A clamp-on current sensor
A CT sensor clipped around the machine's power line tells you, cheaply and safely, whether it's drawing load. Running versus idle, logged every few seconds. No wiring into the control circuitry, no electrician shutting the line down for a day.
2. The signal lamp it already has
Most machines carry a green-amber-red tower light. A small sensor reading those lamps turns a signal the operators already trust into a digital feed. It's inexpensive, non-invasive, and the floor understands it the moment they see it.
3. The PLC, when it earns its keep
If a machine has a modern PLC with a spare port, reading cycle counts straight from it gives the cleanest data of all. I save this for the machines that matter most, since it takes more setup and sometimes the original vendor's cooperation.
An edge device, usually a small industrial gateway or a rugged single-board computer, sits on the floor, gathers these signals locally, and pushes them up. Local-first design matters in India: a power dip or a patchy connection shouldn't cost you a shift of data, so the device buffers everything and syncs once the link is back.
Start with one machine, not the whole plant
I never wire a full floor at once. We pick the one machine the owner argues about most, instrument it in an afternoon, and let a week of honest numbers settle the debate. That first little dashboard wins the floor over better than any slide deck. Once they can see one machine clearly, they want the rest. It's the same lesson as making the invisible visible, applied one machine at a time.
"We were certain the new press was our bottleneck. Two weeks of data showed it was sitting idle half the day, waiting on the old one."Plant manager at a sheet-metal unit, Pune
A few things to watch out for
Some lessons from the field, learned the hard way. Keep sensor cables away from high-voltage lines or shield them well, otherwise your data turns to noise. Agree with the operators up front on what counts as a "stop", so the numbers match the day they actually live. And resist the urge to capture forty signals at once. Three you trust beat forty nobody looks at.
Once the data is flowing, it feeds straight into the handful of numbers that decide your week: OEE, downtime, output per shift. The sensor work is only worth doing because of what those numbers let you go and fix.
