Industry 4.0 on the Line: Data, Sensors and Predictive Maintenance
I'm Iris Xu , a Sales Manager in ZTRFM.
A buyer in Poland asked me last quarter whether our roll forming lines could feed data to his plant system without a separate PC on the floor. That question used to be rare. Now it comes up in almost every technical call. The short answer is yes, and the longer answer is that the value depends less on the sensors than on what you do with the signal once it leaves the machine.
What actually gets sensed
A roll forming line is mostly rotating steel, so the signals that matter are the ones that warn you before something breaks. The usual set is small and cheap to fit.
| Signal | Sensor | What it tells you |
|---|---|---|
| Motor load | Drive current or torque readout | Binding rolls, worn tooling, jams |
| Encoder counts | Inline encoder on the shaft | Line speed, length, drift |
| Hydraulic pressure | Pressure transducer | Punch or shear wear, leaks |
| Temperature | RTD or infrared on bearings | Overheating, lubrication loss |
| Vibration | Accelerometer on the stand | Bearing wear, imbalance |
None of these needs a research budget. A current clamp on the main drive and an accelerometer on the first stand already catch most failures we see in the field.
OEE and downtime tracking
The metric plants care about is OEE, overall equipment effectiveness, which multiplies availability, performance, and quality. A line that runs 80 percent of the shift at 90 percent of rated speed with 2 percent scrap is sitting near 70 percent OEE. The point of logging the signals above is to turn that number from a guess into a daily fact on a dashboard.
Downtime is the easiest win. A timestamped stop event tells you whether the line stalled for a coil change, a jam, or a tool swap. Over a month the pattern is obvious, and the obvious pattern is usually the one worth fixing first. We set this up by reading the drive and the PLC status bits and pushing them to a local display, no cloud required.
Predictive maintenance, not just alarms
An alarm says something is wrong now. Predictive maintenance says something will be wrong next week, and that is the part worth paying for. Two cases come up most often on our lines.
- Bearing wear shows up as a rising vibration signature and a slow creep in temperature on one stand. Catch it early and you swap a bearing during a planned stop instead of after a shaft seizes.
- Roller wear shows up as a creeping scrap rate and a slow drift in the profile tolerance. Track the scrap per thousand metres and the trend tells you which station to regrind before the parts go out of spec.
Neither needs artificial intelligence. A simple trend line with a threshold does the job, and it is honest about what it knows.
The data standards that make it work
The trap in Industry 4.0 is a line that speaks only its own dialect. Two open standards keep that from happening. OPC UA, short for Open Platform Communications Unified Architecture, is the rich, secure option for the factory floor. It carries not just values but meaning, so a tag arrives with its unit, its range, and its alarm state. MQTT, Message Queuing Telemetry Transport, is the lightweight publish and subscribe protocol that ships those values to a broker and onward to the plant system or the cloud. A Sparkplug B profile adds state management so the network knows when a device is alive or dead.
In practice the line runs OPC UA at the edge, an edge gateway normalises the nodes, and MQTT carries the clean stream up to MES, the historian, or a dashboard. Control commands stay on OPC UA. MQTT only reports, it does not drive the machine, which keeps the safety case simple.
Security is part of the design, not a bolt on. We keep the machine network on its own segment, separate from the office network, and the gateway is the only bridge. Certificates authenticate the OPC UA server, and the MQTT broker uses TLS with topic level access control. That boundary is what lets a plant open the data to the cloud without opening the drive to the internet. A line that reports its health should never become a door into the line.
A third signal worth naming is the hydraulic seal. On a press or a flying shear, a slow pressure drop between cycles points to a weeping seal long before it fails. Logging the pressure trend turns a sudden leak into a planned gasket swap. The pattern across all three cases is the same: a slow trend beats a loud alarm, because a trend gives you a date and a parts list, while an alarm gives you a stoppage.
Why it matters for uninterrupted supply
A roll forming line feeding a just in time customer loses money the moment it stops. Predictive maintenance turns an unplanned stop into a scheduled one, and a dashboard turns a gut feel about output into a number the plant manager can defend. For a buyer, the question is not whether to collect data but which three signals to trust first, because a noisy dashboard gets ignored faster than a blank one.
The honest baseline is this: measure OEE for a month with no extra sensors, just the drive run signal and the stop counter. That single number shows you whether you have a speed problem, a stop problem, or a quality problem, and it tells you which of the sensors above to fit first. Most lines we visit are stopped more often than the manager believes, and seeing it on a screen is what finally funds the fix.
Practical takeaway
Start small. Fit a current sensor on the main drive and an accelerometer on stand one, log them to a local screen, and read the trend for a month before adding more. Pick OPC UA at the machine and MQTT to the network so your line stays readable when you change suppliers. Ask for the tag list and the units in the offer, not as an afterthought.
At ZTRFM we build roll forming lines that ship with these sensor and data options fitted and documented, and we are glad to help you choose the first signals that will actually change a maintenance plan. A line that reports its own health is easier to keep running than one that stays silent until it breaks.
