Can old machines be connected to IIoT without a PLC?

Yes. Machines with no PLC at all can be monitored from the outside: a split-core current clamp on the power feed gives run, idle, and off states; photo eyes count parts; vibration and temperature sensors track health. None of these require any connection to the machine's controls.

Does retrofitting sensors void a machine's warranty or certification?

Non-invasive methods generally do not, because nothing on the machine is modified. A split-core CT clamps around an existing cable without cutting it, and magnet-mount sensors bolt to the frame. Always confirm with your OEM and follow lockout procedures during installation, but this is precisely why retrofit starts with non-invasive taps.

How much does it cost to connect an existing machine?

Ranges vary widely by method: current sensors run tens to a few hundred dollars per point, stack light taps similar, wireless vibration points from low hundreds to one to two thousand dollars at the industrial high end, and PLC gateways typically a few hundred to a couple thousand dollars covering several machines. A full line is usually far cheaper than one significant unplanned failure.

Do I need to shut down production to connect machines?

Mostly no. Clamps, taps, and magnet-mount sensors install in minutes and are commonly fitted during changeovers, breaks, or planned maintenance windows under normal lockout procedures. Gateway connections to a PLC are read-only network work that rarely touches the running program.

What data can you actually get from a machine built before 1990?

More than most people expect: run, idle, and off states plus load from current draw, part counts from a photo eye, status from a stack light tap, and bearing health from vibration sensors. What you cannot easily get without controls work is internal process values, which is why very old machines are usually monitored physically rather than digitally.

Is it better to replace an old machine or retrofit it with monitoring?

Retrofit first unless the machine is unsafe, out of tolerance, or unsupportable. Monitoring costs a fraction of replacement and the data it produces makes any later replacement decision far better informed. Many machines flagged for replacement turn out to need a repair and a schedule, not a purchase order.

Connecting Machines Without Replacing Them

You do not need new machines to get live machine data. A clamp on the power feed, a tap on the stack light, a bolt-on sensor, or a gateway on the existing PLC will pull run state, counts, and health signals from almost any powered machine, without modifying it.

That one sentence undoes the most expensive misconception in plant digitization: that connectivity is a property of the machine, so old machines must be replaced before the plant can see itself. It is not. Connectivity is a property of the layer around the machine, and that layer can be added. This post covers the practical ways to add it, what each one costs, and the order to do it in. It sits alongside our full machine connectivity guide and the deeper how-to on connecting legacy machines.

Why do plants assume they need new machines?

Because the pitch for smart manufacturing usually arrives wrapped around new equipment. Machine builders ship connectivity as a feature of the new model, so the story becomes: data comes from new iron. Meanwhile the controls engineer looks at a 1994 press with no Ethernet port and reasonably concludes it has nothing to say.

Both conclusions are wrong for the same reason. A machine does not need to speak a modern protocol to be observed. Everything a machine does leaves physical evidence: it draws current, it vibrates, it gets warm, it turns its stack light red, it moves parts past a fixed point. Every one of those is a signal you can capture from outside the machine, without touching its controls, its warranty, or its certification. The average age of equipment in American plants has been climbing for decades, which means the fleet that needs connecting is mostly the fleet that already exists. Replacing it to get data would be like replacing your workforce to get a phone list.

What are the ways to get data off an existing machine?

There are five, and they form a ladder from cheapest to richest. Most plants end up using several at once, matched to each asset.

Watch the power. A split-core current transformer clamps around the motor feed without cutting a wire. Current draw tells you if the machine is off, idling, or working, and roughly how hard. For many machines that is 80 percent of what a supervisor actually wants to know, for tens of dollars per point.

Tap the lights. If the machine already has a stack light, its own controls are already broadcasting status in red, amber, and green. A small sensor or relay tap on the andon light turns that broadcast into a digital feed with zero controls work.

Add senses. Bolt-on smart sensors add signals the machine never had: wireless vibration sensors on bearings, temperature probes, photo eyes counting parts on the outfeed. These are magnet-mount or bracket-mount installs, wireless, battery powered for years.

Talk to the PLC. If the machine has a PLC from roughly the last three decades, an industrial gateway can usually read its registers over Modbus, a fieldbus, or a vendor protocol, then translate to something modern. This is read-only in most deployments: the gateway listens, it does not command.

Use the native tongue. Newer equipment speaks OPC UA or MTConnect directly. Those machines just need a network path and a subscription, no hardware at all.

Each rung costs more and tells you more. Most floors mix all five, matched asset by asset.

How do you connect a floor without replacing anything?

The order matters more than the hardware. This is the sequence we use and recommend:

Walk the floor and list every asset. Record make, age, controls, existing ports, stack lights, and what question each machine needs to answer. This takes a day, not a quarter.
Decide what each machine needs to say. A bottleneck line needs counts and cycle times. A critical compressor needs vibration and temperature. A rarely used saw might need only on and off. Do not buy uniform instrumentation for non-uniform questions.
Pick the cheapest rung that answers the question. Start with clamps and taps. Reserve gateways for machines whose PLC data is genuinely needed, and add sensors only where the physics matter.
Land everything in one layer, not five apps. Sensor vendors each offer a dashboard. Resist. Feed every signal into one operational layer with one naming scheme, or you are building new data silos with fresher paint.
Put the data to work in week one. Downtime reasons, shift reports, alerts to the right person. Data that changes no decision will not survive its first budget review.

Note what is absent: no machine replacement, no controls rewrite, no production shutdown. Steps one through three are days of work per line. The machine monitoring practice this enables is covered in depth in its own guide.

What does it cost compared with replacement?

Ranges, because every floor differs, but the gap is not subtle:

Split-core current sensors run from tens of dollars to a few hundred per monitored point, install in minutes, and require no electrical modification, per vendors such as Milesight.
Industrial wireless vibration sensors commonly quote battery lives of roughly 3 to 7 years depending on sampling interval, per Emerson and KCF Technologies, with hardware from low hundreds to roughly one to two thousand dollars per point at the high end.
The equipment being connected is old and getting older: U.S. Bureau of Economic Analysis fixed-asset data shows the average age of manufacturing capital stock has risen substantially for decades.
The cost of staying blind is not subtle either: Siemens' True Cost of Downtime 2024 study put unplanned downtime at roughly 11 percent of revenue for the world's largest manufacturers.

A whole line can usually be given a voice for less than the cost of one significant unplanned failure on that line. Run your own numbers with our ROI calculators.

What does this look like when Harmony AI does it?

This retrofit-first approach is not a workaround for us. It is the model. Harmony AI connects machines, ERP, spreadsheets, and paperwork into one operational layer, and we deploy white-glove: our engineers come to your floor in person, walk the asset list with your team, and connect each machine the cheapest reliable way rather than shipping you a box and a wish. No rip-and-replace, ever. The machines you trust keep running; they just stop being silent.

Once connected, the same layer runs live dashboards, downtime capture, and AI workflows on top, so the data lands somewhere that acts on it. That last part matters most, and it is the subject of turning machine data into action. To see the whole pattern on a real multi-plant operation, read the CLS case study.

Nothing on the floor changed except the silence. The intelligence lives in the layer, not the iron.

When is replacement actually the right call?

Sometimes, and honesty requires saying so. If a machine is unsafe, can no longer hold tolerance, has no parts supply, or blocks a capability the business needs, replace it, and buy the new one with connectivity native. But replace it for those reasons, not for data. Connecting first often improves the replacement decision itself: three months of real utilization and health data tells you which machine deserves the capital and which just needed a bearing. Data before dollars, in that order.