Process Monitoring: Real-Time Data Capture in Manufacturing

What is process monitoring?

Process monitoring is the continuous, automated observation of a manufacturing process — machines, parameters, states, and events — so that deviations are visible the moment they occur rather than at the end of the shift. It is not a report you read on Monday morning. It is the live data layer that tells you what is actually happening on the shopfloor right now.

After thirty years of connecting machines to higher-level systems, the single sentence that separates working process monitoring from theatre is this: if an event is visible only after it has ended, you are not monitoring — you are auditing. Real monitoring means the signal, the alarm, or the state change is captured within seconds of occurring and delivered to someone who can act on it before the next part is made.

Process monitoring vs. related terms

The English-language terminology around this topic overlaps heavily. A short disambiguation saves a lot of downstream confusion.

The practical boundary: monitoring tells you what is happening, control changes what is happening, condition monitoring predicts when the machine will break, and SPC tells you whether the output is in statistical control. A complete shopfloor architecture has all four, and each feeds the next.

What actually gets monitored

The list below is what I instrument in a typical rollout. Plants that monitor only two or three of these have huge blind spots without knowing it.

• Machine state: running, idle, setup, fault, planned stop. The backbone of any OEE measurement.
• Cycle count and cycle time: every completed part, timestamped at millisecond resolution.
• Process parameters: temperature, pressure, speed, torque, force — whichever variables drive the quality of the specific process.
• Alarms and fault codes: PLC-side events with their native error codes, not just "machine stopped".
• Setup and changeover events: start, end, who triggered them.
• Quality events: good parts, scrap, rework, downgraded parts.
• Material and tool changes: new coil, new batch, tool index.
• Operator login and intervention: who was on the machine, what they did, when.

The real problem: capturing the data from brownfield machines

Every vendor brochure shows a clean OPC UA pipeline from a modern PLC to a cloud dashboard. Real plants do not look like this. A typical German mid-size manufacturer I walk into has a mix: a few Siemens S7-1500s from 2022, a cluster of S7-300s from 2008, two lines running S5 from 1994, one injection-moulding machine with a proprietary control that exposes nothing, and a CNC from 2019 that in theory speaks OPC UA but in practice has it licensed separately.

Process monitoring that only works on the 20 % of modern machines is not process monitoring. It is a demo. The honest question in every project is how to connect the other 80 %, and the answer is almost always a layered gateway architecture:

• OPC UA where it exists, is licensed, and is stable. That is the preferred path and takes about an hour per machine when the stars align.
• S7 protocol or MQTT for older Siemens lines, reading directly from data blocks without touching the PLC program.
• Digital I/O gateway for machines with no network interface at all. A small box, 24 V signals tapped from the control cabinet, no PLC change, no CE re-certification. This is what gets the 1994 press online.
• File-based or database reads for older SCADA systems that log to CSV or SQL.

The point I hammer at every kickoff: most plants believe their old machines cannot deliver data. That is almost never true. With the right gateway and the right approach every machine can be connected — without PLC intervention and without a production stop. I have connected S5 controllers from the late 1980s that the OEM stopped supporting fifteen years ago. The data is in there; it just has to be lifted out the right way.

What separates real monitoring from theatre

Four design decisions determine whether a monitoring system delivers or becomes shelf-ware.

1. Sample rate per signal, not globally. Machine state wants sub-second resolution so short stops are not lost. A temperature signal that drifts over minutes wants 1 Hz. A setpoint that changes only at changeover wants event-based capture. Global sample rates produce either storage waste or blind spots.

2. Edge buffering for network outages. Factory networks drop. An edge gateway that buffers locally and catches up when the link returns is the difference between a monitoring record with integrity and one with holes that nobody trusts.

3. Timestamp at source. If timestamps are assigned when the data reaches the cloud, correlating events across machines is arithmetic noise. The timestamp has to be written at the gateway, synchronised to NTP or PTP, at the moment of capture.

4. Short stops — the hidden 80 %. Most legacy monitoring misses stops shorter than two minutes because the operator never entered them. In practice, on a typical line, 60–80 % of total downtime consists of micro-stops under 120 seconds. Automatic capture at sub-second resolution is the only way to make that time visible. The first week after honest monitoring goes live, OEE usually drops 15–20 percentage points — not because the plant got worse, but because the measurement finally tells the truth.

From data to action — the part that most projects skip

Capturing the data is the easy part. Turning it into action is where most projects stall. A monitoring system delivers value only when three layers are in place:

• Live visualisation on the shopfloor — not in a manager's office, on the line itself, so operators see the state of their own machine.
• Alerts triggered automatically when defined thresholds are crossed — not emails that arrive the next morning, but notifications on a smartphone or andon system within seconds.
• Historical analysis — Pareto analysis of stop reasons, correlation of alarms with quality defects (the Neoperl pattern), comparison across shifts and machines.

Plants that install the first layer alone get live dashboards that nobody looks at. Plants that install all three get a system that actually changes behaviour.

FAQ

Is process monitoring the same as an MES?
Process monitoring is one function of an MES — specifically the real-time data capture and visualisation layer. A full MES also handles order management, quality, traceability, and ERP integration. You can start with process monitoring alone and extend from there, which is how most SYMESTIC customers enter.

Do I need to modify my PLCs to enable monitoring?
Almost never. Modern gateways read from the control without changing the PLC program, and brownfield digital-I/O gateways tap 24 V signals in the cabinet without touching the control at all. No PLC change, no CE re-certification, no production stop.

How long does it take to connect a machine?
With OPC UA and a cooperative control, under an hour. With a brownfield digital-I/O gateway, two to four hours including cabinet wiring. Complex cases with proprietary protocols or special signal engineering can take a day. These numbers are from actual rollouts, not datasheets.

What's the hardest part of a monitoring rollout?
Not the technology — the signal definition. Deciding which machine state means "running", which means "fault", and which counts as "setup" is a conversation between production, maintenance, and IT that takes longer than the physical connection. Standardise these definitions across the plant before you connect machine number ten, or you will have ten incompatible datasets by machine number twenty.

Can process monitoring work without cloud infrastructure?
Yes, but increasingly it shouldn't. Cloud-native architectures give you elastic storage, cross-site comparison, automatic updates, and zero maintenance on the plant side. For small single-site plants an on-premise option still exists, but for multi-site operations cloud wins on every practical axis.

Related: Process Control · Production Time · OEE · BDE · MDE · MES · SYMESTIC Process Data

About the author

Martin Brandel

MES Consultant at SYMESTIC. Over 30 years in industrial automation — from Simatic S5 and COROS visualisation in the early 1990s to OPC UA, MQTT and IoT gateways today. Built and led SYMESTIC's automation department for eleven years; since 2019 project lead for MES implementations from first enquiry to go-live. Specialist in brownfield connectivity — connecting any machine, any age, without PLC intervention. · LinkedIn