How to Get Production Data Out of Old Machines (Without Replacing Them)

You can get production data out of old machines without replacing them, and in most cases without touching their control wiring at all. The three standard approaches are external sensor retrofits (current transformers, proximity sensors, photo eyes), tapping signals that already exist (like the stack light), and pulling data from an onboard PLC through a protocol gateway. A typical retrofit runs $500 to $3,000 per machine in hardware and can be producing usable data within a week.

That 1987 Bridgeport, the press brake with a controller nobody makes parts for anymore, the injection molding machine that runs three shifts and owes you nothing. These machines still make money. What they don't do is tell you anything. No Ethernet port, no OPC server, no API. Just a motor, some relays, and an operator who knows its moods.

The good news: getting production data from old machines is a solved problem. It's mostly a matter of picking the cheapest signal that answers your actual question.

Start With the Question, Not the Sensor

Before anyone quotes you hardware, write down what you want to know. The retrofit for "is this machine running right now?" is radically cheaper than the retrofit for "what was the spindle load profile on part 4,412?"

Most small manufacturers start with one of three questions. Is the machine running or idle? How many parts did it make this shift? Why did it stop? Those three questions cover run/idle status, cycle counting, and downtime reasons, and they're the foundation of everything from machine monitoring to OEE. Answer them first. Fancier data can come later, on machines that have earned the investment.

A useful rule: your first data project should cost less than one day of the downtime it's meant to explain.

Can You Get Data From a Machine With No Electronics at All?

Yes. This surprises people, but a machine with zero digital anything is often the easiest retrofit, because there's nothing to integrate with and nothing to break.

The workhorse here is the split-core current transformer, or CT. You clamp it around one leg of the motor's power feed (no wires cut, no lockout drama beyond opening the panel safely), and it reports how much current the motor is drawing. Current draw is a shockingly good proxy for machine state. A saw idling pulls a few amps; a saw in the cut pulls a lot more. Set two thresholds and you've got off/idle/running states, timestamped, all day, every day. Split-core CTs typically cost $20 to $80 each, plus an edge device to read them.

For counting parts, the usual answer is a sensor watching something physical that happens once per cycle. An inductive proximity sensor picking up a ram, a turret, or a clamp closing. A photoelectric sensor watching parts drop down a chute. A magnetic reed switch on a guard door. These sensors cost $15 to $150 and have been surviving on factory floors for fifty years. The engineering work is mostly in picking a mounting spot where the sensor sees exactly one event per cycle and can't be knocked out of alignment by a broom.

Then there's the stack light tap, which is our favorite trick for machines that have a tower light but no accessible controller. That green/amber/red light is already a machine-state output, wired and maintained. A small interposing relay or an optocoupler board on each lamp circuit turns the light pattern into clean digital inputs. The machine's own logic did the hard work decades ago; you're just listening.

What If the Machine Has a PLC, Just an Old One?

A surprising number of "dumb" machines have a PLC in the panel, and even old PLCs usually talk. The catch is the dialect.

Anything with a serial port likely speaks Modbus RTU over RS-232 or RS-485, or a vendor protocol like Allen-Bradley DF1. A protocol gateway in the $200 to $700 range converts that to Modbus TCP or MQTT on your network. SLC-500s and PLC-5s with Ethernet or serial sidecars, old Siemens S5 and S7-300 racks, Mitsubishi FX bricks: there is almost always a documented path. Newer controllers speak Ethernet/IP or OPC UA natively and barely count as a retrofit.

Two honest warnings from the trenches. First, you need the tag or register map, and on a twenty-year-old machine the program may be undocumented, password-locked, or living on a floppy disk in a drawer. Budget a few hours for archaeology. Second, keep your polling gentle. Old CPUs have little spare capacity, and an aggressive poll rate can slow a scan time enough to change machine behavior. Read a handful of registers once a second and no one will ever notice you're there.

If the PLC route looks messy, don't force it. A CT and a prox sensor on the outside of the machine will answer the run-state and part-count questions without anyone opening the program. We've seen plenty of projects where the external retrofit shipped in a week while the "proper" PLC integration was still waiting on a password. For a deeper look at protocols and gateway options, see our guide to getting data out of PLCs.

Where the Data Goes Once You Have It

Sensors and gateways produce signals. Turning signals into production data takes three more pieces, and they're all commodity now.

An edge device reads the inputs. This is typically a small industrial I/O gateway or a compact edge PC with digital and analog input modules, mounted in or near the panel, usually $150 to $800. It debounces the signals, applies the thresholds ("above 6 amps means running"), and stamps everything with time.

A transport moves the data. MQTT has become the default for retrofit projects because it's lightweight, works fine over plant Wi-Fi or cellular, and every edge device on the market speaks it. The edge device publishes small messages (machine 14, state=running, count=1) to a broker on a server or in the cloud.

A database and a screen make it useful. Time-series storage, a dashboard showing current state and shift counts, and simple reports on stops. This can be an off-the-shelf monitoring platform or a lightweight custom stack; the IoT and smart systems approach we take depends on how many machines you have and what you already run. Resist the urge to start with a giant historian and a data lake. Start with one screen the shift lead actually looks at.

What Does This Typically Cost?

For a run-state and part-count retrofit on a machine with no usable controller: typically $500 to $1,500 per machine in hardware (sensors, edge device, enclosure, power), plus installation. Electrical installation is often a day or less per machine if the panel work is straightforward.

For a PLC gateway integration: typically $300 to $1,000 in hardware, but more engineering hours for the register mapping and testing, especially if documentation is thin.

Software is where the range widens. Subscription monitoring platforms often run $50 to $150 per machine per month. A custom stack costs more up front and much less per month, which starts to matter somewhere around ten to twenty machines. There's no universal right answer; there is a right answer for your machine count, your IT situation, and your appetite for subscriptions. Our industrial automation team spends a lot of time helping shops make exactly that call.

The wrong answer, almost always, is replacing a working machine to get data out of it. A $1,200 retrofit on a paid-off machine is one of the best returns on the floor.

FAQ

Can I get data from a machine without connecting to its controls at all?

Yes. Current transformers on the motor feed, proximity or photoelectric sensors watching the cycle, and taps on the stack light circuits all work without touching the machine's control logic. This is usually the fastest and lowest-risk path, and it sidesteps warranty and support concerns on machines that still have service contracts.

Will monitoring hardware slow down or interfere with an old machine?

External sensors can't; they only observe. PLC polling can, in principle, load an old CPU, but at sane poll rates (one poll per second for a handful of registers) the effect is negligible. A competent integrator will check scan times before and after connecting anything.

How accurate is current monitoring for run/idle detection?

For most machines, very. Motors draw distinctly different current when cutting, forming, or molding versus idling, so simple thresholds catch state changes reliably. It gets fuzzier on machines with highly variable loads or multiple motors, which is when you add a second signal like a cycle sensor to disambiguate.

Is it worth retrofitting a machine we might replace in two years?

Usually yes, because the sensors and edge hardware move to the next machine, and two years of downtime data is often what tells you whether replacement is actually justified. More than once, the data has shown that the "worn out" machine was mostly starved for material by an upstream problem.

Old iron with good data beats new iron with none, and the retrofit is cheaper than most people expect. If you want a second opinion on the right approach for your floor, our industrial automation engineers do this work across the boundary between machines and software every week. Get in touch and tell us about the machine that's driving you crazy.