No controller, no network port, no problem.
The signal-level engineering: which sources a conventional machine offers, how each is acquired, what accuracy to expect, and exactly where the approach runs out.
Every monitoring vendor can read a networked CNC. The question that decides whether a project covers your whole floor or a third of it is what happens at the machines that have no controller to read — and on most Indian shop floors that is a large minority, sometimes a majority.
This page is the engineering detail: which signals exist on a machine that predates connectivity, how each one is acquired, what accuracy to expect from it, and where the approach genuinely runs out.
What is available on a machine with no control, ranked by usefulness.
| Signal source | What it establishes | How it is acquired | Accuracy |
|---|---|---|---|
| Main drive contactor | Machine on / off | Dry contact to an isolated digital input | Exact — it is the machine's own logic |
| Supply current (CT) | Idling vs working vs stopped | Split-core current transformer, no panel modification | Good, once thresholds are learned per machine |
| Stack light | Running / attention / stopped | Three digital inputs from existing lamp circuits | Exact, and already understood by the floor |
| Existing stroke or part counter | Output count | Pulse input from the counter's own output | Exact |
| Proximity or limit switch | Cycle completion on presses, shears, SPMs | Sensor added at the ram or slide, wired to a digital input | Exact, once debounce is set correctly |
| Hydraulic pressure switch | Working vs pressurised-idle | Existing switch or an added pressure transducer | Good on machines where load is hydraulic |
| Operator kiosk or QR scan | Job, operator, downtime reason | Human input at the machine, two taps | Depends on discipline — interlocking makes it complete |
The technique that does most of the work.
A split-core CT clamped around the supply cable is the single most useful addition to a machine with nothing else to offer, because it separates three states that a contactor alone cannot: powered but idle, running unloaded, and actually cutting or forming. It requires no panel modification and no downtime to fit.
The subtlety is that raw current means nothing without a per-machine reference. A 7.5 kW lathe idling draws a different current from a 30 kW press idling, and the same machine draws differently with a worn belt. So thresholds are learned from the machine's own behaviour over its first days of running rather than set from a specification sheet. This is also why current-based state detection improves over the first fortnight and why we would rather show you a fortnight of data than a demonstration.
Where a machine's load varies enormously by job — a VMC roughing versus finishing — current alone will mis-state short transitions. On those machines we combine it with a discrete signal or accept that sub-minute resolution is not reliable, and we say so rather than presenting a smooth-looking chart that is partly fiction.
Specification that matters in an Indian plant.
Optically isolated inputs
Welding sets, VFDs and contactor coils sit metres away from the panel. Non-isolated inputs on a shop floor fail, and they fail intermittently, which is worse than failing outright.
Real-time clock with backup
A power cut must not corrupt the timeline. Without an onboard clock, a device that reboots produces a record that looks plausible and is wrong.
4G, Wi-Fi or Ethernet
Machines in sheds, yards and annexes have no structured cabling. A monitoring plan that assumes an Ethernet drop at every machine will not cover the whole floor.
Field-serviceable
Screw terminals rather than crimped connectors, and a service port. A device that needs to be returned to a vendor for reconfiguration will eventually be bypassed.
The honest boundary.
Discrete-signal monitoring gives you an accurate availability picture: when the machine ran, when it stopped, for how long, and — with operator input — why. On most unmonitored floors that is where the recoverable loss sits, which is why it is worth doing first.
What it cannot give you is anything that lives inside a control: the active program, feed override, spindle load, alarm codes, or cycle-level performance against an ideal cycle time. If a supplier claims otherwise for a machine with no control, ask precisely which wire carries that information.
Part counting is the one grey area worth being careful about. Where a machine has a counter or a repeatable cycle-completion event, counts are exact. Where output is inferred from current or cycle time, counts are estimates, and they should be labelled as estimates in any report that leaves the plant.
What a first machine looks like.
Half an hour at the panel
We identify which of the signals above exist. Most machines offer two or three; a few offer only the contactor, which is still enough for a timeline.
An afternoon, at a planned stop
Gateway mounted, inputs wired, CT clamped. No control modification, nothing written to the machine.
One shift, watched
Machine and dashboard compared side by side. States that disagree with reality are corrected before anyone is asked to trust the data.