Fiber Between Buildings, Every Time

The pump building and the control building are sixty meters apart. There is a spare conduit. Copper Ethernet is rated to a hundred meters. The electrician pulls a shielded cable, the switches link at a gigabit, and the job is done by lunch. It works for a year. Then a summer storm puts a surge on the ground grid at the pump building, a few hundred volts appear between the two buildings for a millisecond, and the cheapest path between them is the shield of that cable and the Ethernet port of the controller at the far end. The controller is replaced. The cable is not, because nobody connected the two events.

Two buildings are two grounds

Every building has its own grounding electrode, its own bonding, and its own share of the fault and lightning current flowing through the earth. The two are never at exactly the same potential, and during a fault or a strike the difference can be hundreds or thousands of volts for a moment. A conductor between the buildings, any conductor, becomes a path for that difference. A copper network cable is such a conductor twice over: its pairs and its shield. Transformers inside Ethernet ports isolate a few hundred volts of common-mode difference in normal service, which is why the link works on a quiet day. They do not survive the storm, and a shield bonded at both ends does not even wait for the storm; it carries a circulating current every day and injects noise into the pairs as it does.

What fiber removes

The point is not that fiber is more reliable in degree. It is that fiber removes the failure mechanisms entirely. There is no ground loop because there is no loop, no surge path because there is no path, and no coupled noise because nothing couples into glass. What remains are the connector cleanliness and the bend radius, which are handled at installation and stay handled.

What it costs

A fiber link between two panels needs a cable, two patch panels or fiber terminals, patch cords, and either fiber ports on the switches or a media converter at each end. On a plant network with managed switches, the switches usually have fiber ports or transceiver slots already. The cable itself costs little more than a good shielded copper cable and often less than the armored copper that a wet conduit would need. The premium over copper is typically a few hundred dollars per link for the terminations and transceivers, and the labor of a competent termination. Against that: a controller Ethernet port, a switch, or a network module destroyed by one surge, a day of downtime, and the week that someone spent chasing analog noise that turned out to be a shield current.

How to specify it

• Single-mode fiber for anything outdoors or between buildings; the transceivers cost a little more than multimode and the cable will never be the limit.
• An outdoor-rated, gel-free, loose-tube or armored cable for the conduit or the direct burial, with the strand count doubled for spares; twelve strands is the common minimum.
• A patch panel or a fiber terminal box at each end, in the panel or a separate enclosure, with the strands terminated on connectors of one type and one polish throughout the plant.
• Fiber ports on the switches where possible; industrial media converters where not, on the rail and the panel supply.
• Acceptance testing: inspection, loss with a light source and power meter in both directions, and a reflectometer trace, recorded in a fiber schedule.
• Labels on every strand at both ends and a fiber schedule in the engineering library.

The exceptions

Copper is fine for short runs inside one building, from a switch to a device in the same room or the same panel lineup on one ground, kept away from power. It is fine for a patch cord. It is fine for the last few meters to a camera or an access point where power over Ethernet is wanted, from a switch in the same building. What copper is not fine for is anything that leaves the building, crosses between grounds, runs beside drive cables, or approaches a hundred meters. Those links are fiber, every time, and a plant standard that says so removes the lunchtime decision that costs a controller.