You connect legacy machines by reading them from the outside. A current clamp on the motor leads gives run state, a sensor on the stack light gives faults, a vibration sensor gives health, and a retrofit counter gives throughput. No network port, no controller access, and no machine modification required.
The machines that make the money in most plants are not new. A well-built press, grinder, molder, or filler runs for decades, and replacing a machine that holds tolerance just to get data out of it is a terrible trade. The trade you actually want is a few hundred to a few thousand dollars of retrofit instrumentation per machine, installed without touching the machine's guts, feeding the same data platform as your newest line. This post is the playbook: what each retrofit option tells you, what it costs in rough terms, and how to pick per machine.
Why not just replace the old machines?
Because the machines are not the problem; the visibility is. A 1985 press that holds tolerance is an asset. Capital replacement to gain a network port costs six or seven figures per machine, takes months, and carries process risk, while a retrofit costs a rounding error of that and is installed in an afternoon. Even the modernization argument mostly favors retrofit: the payoff people want from "smart" machines is knowing when they run, why they stop, and when they will fail, and every one of those is readable from the outside.
There is a second reason. Legacy equipment fails dark. When a machine with no data feed goes down, the downtime clock starts when someone notices and the cause is whatever the operator remembers at shift end. Plants that retrofit their old iron consistently discover that downtime on those machines was undercounted, sometimes badly, because the short stops never made the clipboard. You cannot fix what you cannot see, and the oldest machines are usually the least seen.
What can you learn without touching the controls?
More than most people expect. Every machine leaks information through its electrical draw, its indicator lights, its vibration, and its physical output. Retrofit connectivity is the discipline of catching those leaks with sensors instead of opening the panel.
The current clamp is the workhorse. A split-core current transformer snaps around one motor supply lead without disconnecting anything. Motor amp draw separates cleanly into off, idle, and under-load in most machines, which means run state and utilization come from a sensor that costs tens of dollars plus a small transmitter. For many plants this single retrofit, applied across the old fleet, is the fastest visibility win available.
The stack light is a free signal, already engineered. Someone already decided what red, amber, and green mean on that machine. An optical or wired sensor on the stack light inherits that logic: faults, warnings, and running states with no reverse engineering. Where lights are wired through a relay, a simple digital input does the same job even more reliably. If your floor uses andon conventions, the retrofit maps straight onto them.
Vibration tells you about the future. Bearing wear, imbalance, and misalignment show up in vibration signatures weeks before failure. Battery-powered wireless vibration sensors mount magnetically on a bearing housing and report over low-power radio, which makes them the rare predictive tool that works as well on a 1970s gearbox as on a new servo line. Picking the right type is its own topic, covered in vibration sensor types.
Counters and photoeyes measure truth at the discharge. A photoeye or proximity sensor at the outfeed counts actual parts, which converts directly into cycle time and throughput. Paired with run state from the clamp, you get the difference between "running" and "running well," which is the heart of machine monitoring.
The panel still speaks, if you listen at the right layer. Many "unconnected" machines actually have a controller from the 1990s with a serial port speaking Modbus RTU or a vendor protocol. A protocol gateway converts that to modern transports for a few hundred dollars, and suddenly the machine is Tier 3, not Tier 4, in the connectivity tier model. Always check the panel before buying sensors; the cheapest retrofit is the port someone forgot was there.
Two smaller options round out the kit. Panel power meters measure energy per part, which doubles as a process anomaly detector: a cycle that draws more energy than its neighbors usually has a story. And modern smart sensors increasingly combine sensing, thresholds, and radio in one housing, which shrinks the install to a magnet and an app. Neither is where you start, but both extend the same principle: measure the machine's side effects, and the machine does not need to cooperate.
How do retrofit signals get into your data platform?
The same way everything else does. Retrofit sensors report to a nearby gateway, wired as 4-20 mA or discrete inputs into a small I/O module, or wireless to a radio receiver. The gateway timestamps, names, and publishes the signals over MQTT exactly as it does for PLC tags, so downstream systems cannot tell a 1985 press from a new line. The plumbing details are the same ones covered in connecting PLC data to the cloud: outbound-only, buffered, report by exception.
The naming step matters more on retrofits than anywhere else. A PLC tag at least has a name; a current clamp has nothing until you give it one. Name signals by machine and meaning, not by sensor serial number, and fold them into the same hierarchy as the rest of the floor so the old machines appear next to the new ones in every view.
How do you run a legacy retrofit project?
- Audit the fleet. For each machine: controller and ports if any, stack light, motor size, mechanical output. Sort into "has a hidden port" and "true retrofit."
- Convert the hidden ports first. Serial and Modbus conversions are the cheapest wins and need no new sensors.
- Fit current clamps across the true retrofits. Run state and utilization on the whole old fleet is the fastest broad payoff.
- Add stack light and counter sensing on the machines where downtime hurts most. Now you have state, faults, and counts on the machines that matter.
- Add vibration sensing on the assets that would hurt worst to lose. Gearboxes, main spindles, big fans: the failures that take weeks to recover from.
- Feed it all into one layer and put reason codes on top. Sensors say the machine stopped; operators say why. The pairing is what makes the data usable.
That last step is the honest limit of retrofits: a clamp knows the press stopped, but only the operator knows the die was being cleaned. Pairing automatic detection with quick human reason codes, on a tablet at the machine, is what separates plants that use their retrofit data from plants that collect it. The wider set of options for combining machine and human data is covered in machine data collection methods.
What does this cost, roughly?
Ranges, not quotes, and hardware only: split-core current clamps run from tens of dollars, plus a transmitter or I/O module typically in the low hundreds. Wireless vibration sensors typically land in the low-to-mid hundreds per point. Photoeyes and stack light sensors are tens to low hundreds. Protocol gateways for serial-to-Ethernet conversion are typically a few hundred. A practical full-machine kit, state plus counts plus faults, usually lands in the hundreds to low thousands per machine, which is two to four orders of magnitude below replacement. Installation is commonly an afternoon per machine, and clamp-on devices do not require modifying the machine's electrical system.
Where does Harmony AI fit?
Retrofitting mixed-vintage floors is exactly the situation Harmony AI was built for. Harmony AI connects the 1985 press and the new line into the same operational layer, alongside your digitized paperwork and existing software, with no rip-and-replace anywhere. And because Harmony AI deploys in person, the fleet audit above is something Harmony AI engineers do on your floor, machine by machine, rather than leaving you with a parts list. The CLS case study shows the result on a real plant: operations that used to live on paper and in memory becoming live, visible, and queryable, on the equipment they already owned. If you want to size the prize first, the ROI calculators will put a number on what dark downtime is costing you.
What do the standards say?
Retrofit connectivity leans on standards that are older and more stable than most of the machines:
- The 4-20 mA analog signal that most industrial sensors and transmitters use is defined by ANSI/ISA-50.00.01, first published in 1975, which is why a new transmitter still wires cleanly into decades-old instrumentation practice.
- Modbus was introduced in 1979 and has been an open protocol under the Modbus Organization since 2004; it remains the most common language hiding behind old panel doors.
- On the publishing side, MQTT (an OASIS standard, ISO/IEC 20922) and Eclipse Sparkplug (ISO/IEC 20237, published 2023) mean retrofit signals ride the same modern infrastructure as everything else.
The pattern to notice: the standards span fifty years, and they interoperate. That is the entire thesis of legacy connectivity. Old machines are not a dead end; they are just machines whose data path has not been built yet.