Short answer: terminate I/O straight to the controller for equipment within about 15 m [50 ft] of the cabinet, and put everything farther out on a remote drop supervised by an alive bit. The master alarms on a dead drop within a few update intervals — under 5 s end to end on a typical network. Leave that alarm out and a failed drop freezes SCADA on its last values: a stopped pump displays as running, no process alarm fires, and nobody knows for hours. That frozen screen is an alarm-design failure, not proof that remote I/O is unusable. Here is how the two architectures split, what each costs in cable and cabinet space, and the watchdog logic that makes remote I/O safe.
The two architectures side by side
Both architectures read the same points into the same scan. They differ in where field wiring lands and what sits between the contact and the input image.
| Local I/O | Remote I/O | |
|---|---|---|
| Field wiring | Every field core home-runs to the main cabinet | Field cores land at a drop box near the machines; one network cable runs home |
| Planning distance | Up to about 15 m [50 ft] | 30-200 m and more, wherever signals cluster |
| Cabinet space | Terminal rail grows with each signal: 48 signals take 288 mm of 6 mm-pitch rail | Main cabinet holds adapters and local I/O; the drop carries the terminations |
| Signal path | Copper only | Adapter and network sit in the loop |
| Failure pattern | One broken wire loses one input | A dead drop freezes every point behind it, often 8-64 at once |
| Alarm design | Input state is the whole truth | Each drop needs an alive bit and a communication alarm |
| Extra hardware | Cable, conduit, terminal rail | Drop enclosure, adapter, drop power supply, network cable |
The 15 m [50 ft] figure is shop practice, not physics: cable cost and terminal count set the split. Between 15 and 30 m, let the signal count decide — 8 points stay local, 40 go remote.
Where local I/O wins
Keep terminations in the main cabinet when the equipment is close and the count is small. A machine cell 10 m away with 16 signals is one multicore pull — 32 field cores over 10 m and 32 terminations — nowhere near enough hardware to justify an adapter, a drop power supply and a second enclosure. Direct wiring also removes the network from the signal path: if the contact changed, the input image changed. Alarm design stays simple because there is no separate health state to supervise.
The price is cabinet space, and it grows signal by signal. NEC 312.6 sets the wire bending space at terminals in cabinets and cutout boxes; where a conductor enters the wall opposite its terminal, Table 312.6(B) applies and the space grows again. Duct absorbs some of the bulk, but one wiring-duct maker caps every fill table at 50% of the duct's interior cross-section, citing NFPA 79-2012 section 13.5.2 — round cables leave air gaps, so half full is all a duct holds. Plan rail, duct and fill with the backplate and DIN rail layout guide, and size the cabinet for the terminations you have plus the ones you expect; the electrical enclosure sizing guide walks that arithmetic. If the cabinet also houses breakers or a disconnect, NEC 312.8 limits how much of it may be filled with conductors, splices and taps. A cabinet that is already tight stays tight; the cramped PLC panel wiring cleanup shows what recovering one takes.
Where remote I/O wins
Past about 15 m, count the cable before you choose. Worked example: a washdown skid 120 m from the cabinet carries 24 digital inputs. Local I/O pulls 24 × 2 = 48 field cores as two 24-core cables — 2 × 120 m = 240 m of heavy multiconductor — and lands them on 48 terminals inside the cabinet. At the usual 6 mm terminal pitch, that field row alone takes 48 × 6 mm = 288 mm of rail, with a duct run above it. Remote I/O lands the same 24 signals in a drop box at the skid and runs one shielded 2-core network cable home: 120 m of light cable, and 48 field terminations the main cabinet never sees. You halve the cable metres and free about 290 mm of rail; at a 100-signal skid the local option stops being a conversation.
Analog points tip the balance sooner still. A 4-20 mA loop pulled 120 m needs shielded cable, shield terminations at both ends and a noise path to manage; at a drop, the loop is 5 m and the network carries the value home.
House each drop in a wall-mount enclosure sized for the adapter, its power supply, the terminal blocks and about 30% spare DIN rail. In washdown or outdoor service that is a 316L stainless box rated NEMA 4X / IP66 by the manufacturer; not UL listed. An outdoor drop in full sun runs well above shaded ambient — how hot an enclosure gets in direct sun — so mount the west wall last and mind the adapter's temperature rating. We cut every gland and network-cable cutout from one piece and deliver in 10 business days from your drawing approval. The remote robot I/O sizing walkthrough runs this same arithmetic, and controller enclosure cable entry for machine wiring covers gland selection and cutout placement at the drop.
The alive bit that catches a dead drop
A remote drop fails as a block. Lost power, an adapter fault or a cut network cable takes every point behind it out at once, and the controller holds the last input image it received. SCADA reads the controller and shows those values as healthy: the level switch that changed stays in its old state, the stopped motor stays green, and process alarms keyed to those inputs never fire.
Adapter-side detection is quick: one classic remote I/O adapter manual declares communication lost after 100 ms of network silence. Your job is the master side. Set a bit in each drop that toggles every scan the drop runs, count its transitions in the master, and raise a drop-specific communication alarm when transitions stop. Watch the transitions, never the level — a frozen image can hold the expected level forever.
Worked timeout, with one input to replace: say the drop's configured update period is 200 ms. Set the master watchdog at 3 × 200 ms = 600 ms. The adapter flags the silence at 100 ms, the master confirms it at 600 ms, and one controller scan plus one SCADA poll lands the banner in about 1-2 s — inside the 5 s target with margin. On SCADA, give every drop its own alarm, name the area in the message, and mark every tag behind the drop stale while the alarm is active. "Skid 3 drop: no communication" tells the operator to distrust the whole screen; normal-looking values invite trust in a dead plant. Configuring alarms on local and remote stations walks through the alarm-side setup. Then interrupt a drop during a planned outage and confirm three things: the banner, the stale marks and a clean recovery.
The decision by environment
Decide area by area, not project by project.
- Equipment within about 15 m [50 ft] of the cabinet: local I/O. Terminate to the controller and keep the network out of the signal path.
- Signal clusters 30-200 m out — pump rooms, washdown skids, packaging lines, roof units: remote I/O in a wall-mount drop box, one network cable home.
- Analog-heavy or still-growing areas: remote I/O. Each new point lands in the existing drop instead of becoming another 120 m home-run.
- A cabinet already at its duct and bending limits: remote I/O for new field points; the 50% duct fill cap and NEC 312.8 decide that one for you.
- Drives add heat to the same question; VFD in the main control enclosure or a remote box runs that version.
- Safety circuits stay hardwired to the cabinet: an E-stop loop must open with the network, the adapter and the controller all dead, and networked safety is only for a function designed and rated for it.
If your plant froze once and banned remote I/O the next day, the fix was the alive bit and the alarm — the ban just freezes your architecture too.
What to put on the order
Write the split into three documents.
Cabinet order: sized for the local terminations with NEC 312.6 bending space at every row, duct fill at or under 50%, and 20% spare DIN rail. Configure material, size and cutouts in the enclosure builder — 304 stainless indoors, 316L for washdown and outdoor.
Drop orders: one enclosure per signal cluster, about 30% spare rail, cutouts for the network cable and the power glands specified with the order, 316L wherever hoses or weather reach. Pick the size with the enclosure selector, then send each box through the builder as a custom order.
Controls spec: one alive bit per drop toggling every scan; a master watchdog at three to five times the update period; one named communication alarm per drop; stale marks on every tag behind an active drop alarm. Four lines, written before the panel is built. Retrofitting them after the first frozen screen is how plants end up banning remote I/O outright.
Catalog dimensions to compare
Compare these catalog dimensions with your component layout. Open each enclosure for its rating, material and usable space, then check those specifications against your application.
| Enclosure | Material | Outside H x W x D / display | Rating |
|---|---|---|---|
| 60 × 36 × 16 in | 304 stainless | 60 x 36 x 16 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
| 48 × 36 × 16 in | 304 stainless | 48 x 36 x 16 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
| 60 × 36 × 12 in | 304 stainless | 60 x 36 x 12 in | Rated NEMA 4X / IP66 by the manufacturer; not UL listed |
Questions
How fast should a remote IO drop raise its alarm?
Set the master watchdog at three to five times the connection's update period. A classic remote IO adapter declares communication lost after 100 ms of network silence; with a 200 ms update period the watchdog fires at 3 × 200 ms = 600 ms, and the alarm banner lands on SCADA in about 1-2 s, well inside the 5 s target.
Where is the distance cutoff between local and remote IO?
About 15 m [50 ft]. Below it, home-run cable is cheap and terminations stay in the main cabinet. Above it, cable metres and terminal rail grow together: 24 signals over 120 m mean 240 m of multiconductor cable for local IO against 120 m of network cable for one drop. More signals move the cutoff shorter; a dozen signals stretch it.
Why does SCADA look healthy after a remote IO drop fails?
The screen displays the last values received, so a stopped pump stays green and process alarms keyed to the frozen inputs never fire. Group the affected tags by drop, add a named communication alarm per drop, mark the dependent tags stale while it is active, and prove the set-up once by interrupting a drop during a planned outage: the banner should arrive inside 5 s.
Should E-stops and safety interlocks ride the remote IO network?
Keep them hardwired to the cabinet by default. A hardwired E-stop loop still opens when any network, adapter or controller has died, which is exactly the condition you worry about. Put safety signals on a network only where the safety function is designed for it and the safety components carry the rating for networked safety use.
What enclosure suits a remote IO drop in a washdown area?
A 316L stainless wall-mount box rated NEMA 4X / IP66 by the manufacturer; not UL listed. Size it for the adapter, its power supply and the terminal blocks with about 30% spare DIN rail, and order the gland and network-cable cutouts with the box so every penetration stays rated. Boxes are cut from one piece and delivered in 10 business days, custom work counted from your approval of the drawing.
