Industrial IoT

ISO 20816, the successor of ISO 10816, evaluates machine condition from vibration velocity in mm/s RMS, measured on non-rotating parts such as bearing housings over a broad band of roughly 10 to 1,000 Hz. The standard defines four evaluation zones: zone A is where newly commissioned machines normally sit, zone B is acceptable for unrestricted long-term operation, zone C is unsatisfactory for continuous operation and calls for a planned intervention, and zone D indicates severity high enough to cause damage. Boundaries depend on machine size and mounting; for a medium machine of 15 to 300 kW on flexible supports, ISO 20816-3 places them around 2.3, 4.5 and 7.1 mm/s RMS.

In an IoT deployment, a permanently mounted accelerometer on each critical bearing streams the RMS value continuously, so evaluation stops being a monthly route with a portable meter. The platform stores the trend and a rule fires when a machine drifts from zone B towards zone C, which is exactly the window where maintenance can still be scheduled instead of suffered.

OEE, overall equipment effectiveness, multiplies three factors: availability, the run time divided by planned production time; performance, the actual output against the ideal cycle time while running; and quality, the good parts against total parts. A line that runs 90% of its planned time, at 95% of its rated speed, with 99% good parts scores 0.90 x 0.95 x 0.99, about 84.6% OEE. As a reference, 85% is commonly cited as world class for discrete manufacturing, while typical plants measure closer to 60%.

The hard part has never been the formula but the data collection. Connected machines feed the three factors automatically: run and stop signals give availability, cycle counters give performance and reject counts give quality, so OEE becomes a live number per line and per shift instead of a spreadsheet built days later.

Yes, and retrofit is where industrial IoT usually pays back first, because the oldest machines are the least visible ones. The approach is non-invasive: a power analyzer with clamp-on current transformers at the cabinet feed reveals machine state, cycles and consumption from the electrical signature; magnet or stud mounted vibration and temperature sensors watch bearings and gearboxes; a photocell on the outfeed counts parts. None of it touches the machine's original controls or wiring.

Installation typically fits in a normal maintenance stop. The sensors talk 4-20 mA, pulse or Modbus to a small gateway, so a machine from the nineties reports to the same dashboards as a brand new line.

In parallel and read-only, which is why plants accept it. The built-in Node-RED plugin polls PLC registers over Modbus TCP or RTU and subscribes to OPC UA servers, the two interfaces that virtually every controller and SCADA already exposes. PLC programs are not modified and the SCADA keeps doing control; the platform adds what control systems lack: long-term storage in data buckets, dashboards you can share outside the control room, alarms and a REST API.

Integration also runs the other way. Stored and validated data can feed historians, MES or ERP systems through the API or webhooks, so plant data stops being locked to one supervision desk.

It depends on the number of machines, the sensors involved and the integration work, so most projects are quoted individually. As a reference, our Private Cloud plans are published on the pricing page and start at 49 euros per month for a dedicated instance, which is usually enough to run a pilot on one line or a handful of machines. Retrofit hardware is modest too: an industrial vibration sensor or a three-phase power analyzer costs a few hundred euros per point.

That is worth comparing with what downtime already costs. The Siemens True Cost of Downtime research puts a single unplanned event at around 2 million dollars on average for large plants, so a pilot pays for itself the first time it catches a failing bearing early.

In most deployments, yes. An industrial gateway, whether a DIN rail box, an industrial PC or a Raspberry Pi class device running the Thinger.io Linux client, sits on the machine network, polls Modbus and OPC UA locally at high rate, then filters and aggregates at the edge, sending one useful value instead of thousands of raw samples.

The gateway also answers the two objections IT and OT raise first. It buffers data locally during connectivity loss, so a WAN outage does not mean lost history, and it connects outbound over TLS to the platform, so no inbound firewall ports are opened towards the plant network.