Machine monitoring: comparing the four approaches

First, what are you actually trying to find out?

Worth being clear before you look at suppliers, because the answer changes which approach is right and most vendors will steer you toward the one they sell.

Most shops want one of three things. How much of the day the machines are actually cutting, because they suspect it's less than they'd say out loud. Why the gaps happen, so they know what to fix first. Or how many parts came off a machine in a shift, usually for costing rather than improvement.

Those want different things. The first two need machine state over time. The third needs a count. Several approaches give you one and not the other, and it's the commonest reason a monitoring project disappoints.

If you can't say which of the three you're after, do that before you get a quote. It's the question that decides everything else.

The four approaches

In rough order of how invasive they are.

Manual logging

Someone writes it down. Paper, whiteboard or a spreadsheet.

How it works: Operators or setters record start and stop times, and usually a reason. Someone types it up later.
What it tells you: Reasons, in the operators' own words. That's genuinely valuable and no sensor gives you it.
What it doesn't: Anything reliable about time. It records what someone remembered at the end of a shift, and short stops go unrecorded almost entirely. It also stops the moment things get busy, which is exactly when you need it.
When it's the right choice: As a short exercise to understand a specific problem. Not as a permanent measurement system.

Stack light sensing

A sensor watches the machine's tower light and reports the colour.

How it works: A photo-sensor or clip reads which lamp is lit and sends it on. No electrical work on the power side.
What it tells you: What the machine thinks it's doing, cheaply, on any machine with a working stack light.
What it doesn't: The truth, necessarily. Green often means powered and ready rather than cutting, and plenty of shops have machines with a lamp out or wired oddly. You're measuring a signal set by whoever programmed the machine, not the work.
When it's the right choice: Where machines have a well-configured stack light you already trust, and you want coverage fast and cheap.

Current sensing (what we build)

Clamps round the power cables read how much current the machine is drawing.

How it works: Split-core clamps clip round existing cables, typically one on the incomer and one on the main drive. An edge unit reads them and sends state to a dashboard.
What it tells you: When the machine was cutting, powered but idle, or off. Works on anything that draws current, whatever its age or brand, with no access to the controller.
What it doesn't: Part counts, on its own. It sees load, not pieces. It also can't see inside the control, so no programme names, feeds and speeds or alarm codes. Adding a proximity switch gets you a count; nothing gets you the controller data.
When it's the right choice: Mixed fleets, older machines, or anywhere you can't or won't touch the CNC control. It's the broadest coverage for the least disruption.

Controller integration

Software talks directly to the machine's own control.

How it works: Connects over MTConnect, OPC-UA, Fanuc FOCAS or a vendor protocol and reads what the control already knows.
What it tells you: By far the most: state, programme, part counts, alarms, overrides, sometimes tool life. If the control knows it, you can have it.
What it doesn't: Anything on machines that don't support it, which in a typical UK subcontract shop is most of them. Older machines have no port, some need a licence from the OEM to unlock, and support varies by model even within one brand.
When it's the right choice: Newer, homogeneous fleets from one or two makers, where you need part counts and programme-level detail and the machines can provide it.

Side by side

Generalisations. Any specific supplier may differ, so use this to frame questions rather than to decide.

	Manual	Stack light	Current sensing	Controller
Works on old machines	Yes	If it has a lamp	Yes	Rarely
Touches the control	No	No	No	Yes
Electrical work needed	None	Little	Yes, an electrician	Usually none
Gives part counts	Yes, if recorded	No	Only with an extra sensor	Yes
Captures short stops	No	Yes	Yes	Yes
Gives reasons	Yes	No	Only if operators tap one	Alarm codes only
Effort to keep running	High, forever	Low	Low	Low once working
Main risk	It quietly stops	Measuring a lie	No part counts	Machines can't support it

What to ask any supplier

Including us. If an answer is vague, that's the answer.

Which of my machines won't this work on?Every approach has exclusions. A supplier who says "all of them" hasn't looked at your list.
What exactly counts as running?Get the definition in writing. The gap between "powered" and "cutting" is where most of the disputed hours sit.
Who does the install and what do I have to provide?Electrician, network, ladders, downtime. Find out whose job each is before the date is booked.
How are thresholds set, and by whom?Shipped defaults across a mixed fleet produce numbers nobody believes. Ask whether they calibrate on your machines.
What happens when the network drops?Does the unit buffer and catch up, or is that shift's data gone?
Can I get my data out?Export or API. If the answer is a PDF report, you don't own your data.
What does it cost in year three?Hardware, subscription, support, and what happens if you add ten machines.
What happens if I stop paying?Does the hardware become yours, a brick, or something you have to send back?