Shop-floor network requirements: what the building actually needs.

None of those four jobs is exotic and none of it is specific to a vendor. It is also the part that quietly adds months to a project when nobody looks at it first.

Below is what we would check, in the order we would check it. It is deliberately written without a product in it.

Start with a survey, not a drawing

Walk the floor with the racking loaded and the machines running. An empty warehouse surveys beautifully and then fills up with steel, and a wireless plan built in a shutdown is a plan for a building that does not exist during production. Note where people stand, where tablets and scanners get used, where the machines you care about actually are, and where there is already a live network port within reach.

The output you want is boring: a list of locations that need a connection, a list of areas that need coverage, and a note of what is already there. That list is what turns three quotes into three comparable quotes.

A sheet with one row per location is enough. These are the columns worth filling in; the row below is an example of the kind of note to make, not a recommendation.

Coverage at the far press

Factory wireless is not office wireless with more access points in it. The failure mode is specific and familiar: everything works near the office, and the tablet drops out at the far end of the shop, mid-inspection, every time.

• Place access points for what is in the building — steel racking, mezzanines, roller shutters, a moving forklift fleet and a row of presses that absorb signal.
• Set channels and transmit power deliberately rather than leaving everything on automatic, so your own access points stop competing for the same air.
• Tune roaming so a device walking between cells hands over instead of clinging to a distant access point. A device holding on at two bars generates a support call a week.
• Put shop-floor devices on their own SSID, separate from office and guest traffic.
• Get a post-install survey and heat map, and keep it. When the racking moves, you want to see what changed rather than guess.

One rule worth adopting outright: anything fixed and important gets a cable. Label printers in particular. A printer on marginal wireless produces support calls out of all proportion to what it cost.

Ports near the machines

Most factory networks were designed around desks. Metering, gateways, machine interfaces and andon screens all want connections out where the machines are, and the honest answer is usually that there is nothing within sixty metres.

The failure pattern is predictable: the nearest switch is full, the run is long, so somebody puts a five-port switch on top of a cabinet on a shared extension lead. It works for two years and then becomes the thing nobody can find at seven on a Monday morning. Plan ports the same way you plan sockets — where the work is, with spares, on containment, going back to a real switch.

We wrote one such job up as a field note, Instrumenting a press with no OPC UA, where the instrument was the cheap part and the network route was not.

• Pull spare cores and spare drops while the containment is open. The expensive part is access, and you only want to pay for it once.
• Check the PoE budget of the switch, not just the port count, if it is feeding access points, cameras or screens.
• Label both ends against a port schedule you keep, so the next person in that cabinet is not stood there with a toner.

What each device needs

Different devices ask different things of the network. Check each one on the survey sheet against this list.

Cabling that survives a welder

Office cabling installed in a factory fails slowly and confusingly — a port that works until the forklift traffic starts, a run that is fine until the welder beside it strikes, data that goes strange on exactly the shift you needed the numbers from.

• Decide the route before the cable is pulled. Routing away from drives, welders and motor cable — and crossing them at right angles where you must — is the cheapest fix for electrical noise there is.
• Use steel or galvanised containment rather than clipping to a purlin and hoping. Where a run crosses a traffic route it goes in trunking rated for what drives over it.
• Use shielded cable and shielded modules where the environment justifies it, and earth the screen at one end only. A screen earthed at both ends can behave like an aerial.
• Consider fibre for any route that is long, hostile or between buildings. Glass carries no electrical noise, so a route past the press bay stops being an argument.
• Insist on per-port test results in the handover pack — the results themselves, port by port, not a summary sheet saying it passed.

Keep plant traffic apart from office traffic

A flat network is fine right up until it is not. Machines, meters, scanners, printers and screens belong on their own VLANs, separated from the office network, with what routes between them written down rather than remembered.

Three principles are worth holding to whatever you are connecting. Read-only where reading is enough — nothing should be writing to a PLC unless somebody has consciously decided it should. Per-device credentials rather than one shared password on a sticker, so a device that goes missing is a revocation rather than a rebuild. And no inbound path from the internet into the plant network: anything that needs the cloud should make an outbound connection, not wait for one.

A comms room that is not a shelf

A great deal of factory IT lives above a compressor. It works, right up until it is the reason the line is stopped and nobody can say what is plugged into what.

• A proper cabinet with patch panels, cable management, labels front and back, and spare U for the next three years.
• Ventilation and dust treated as part of the job. A cupboard sharing air with the compressor kills switch fans.
• A UPS sized to ride out short dips and shut things down cleanly when it cannot — and tested on the day rather than assumed.
• A rack diagram and photographs, so a support call can start with “which port has the amber light” instead of a site visit.

How to compare quotes

Three quotes are only comparable if they price the same job. A quote worth comparing itemises:

• the containment, and where it runs;
• the number of drops, and where each one ends up;
• testing, with per-port results handed over;
• a survey before the work, and a coverage survey after it;
• the handover pack.

Then ask every installer the same questions:

• Will you survey with production running?
• When will the work be done, and what needs a shutdown?
• What test results will I receive, port by port?
• Who holds the credentials afterwards?
• What happens to coverage when the racking moves?

What to insist on before you sign anything

Whoever does the work — your own team, your usual contractor, or somebody else — ask for the same five things, and treat a reluctance to provide them as the answer to a different question.

• A device and port inventory: what was fitted, where, with serial numbers and addresses.
• Network notes: addresses, VLANs, what talks to what, and which ports matter.
• Per-port test results.
• Credentials in your vault, not theirs, and revocable by you.
• Runbooks for the handful of things that realistically break.

An install only the installer understands is a liability that was sold to you. You should end up owning the estate, not renting an understanding of it.

If it would be useful to have somebody survey the building and tell you what the job really involves, that is what our factory network infrastructure service is for — and it is quoted on its own terms, with no software attached.