// Platform · Edge Devices

Two cables in. The whole plant out.

A fanless industrial PC that mounts in the panel next to your PLCs. One network port faces the plant: it speaks ControlLogix, SIMATIC, and groov EPIC, and through them every 4–20 mA loop, drive, and flowmeter on the floor. The other port faces the enterprise: an outbound-only encrypted tunnel that streams the plant's data, already cleaned, named, and contextualized, to your central platform. Local SCADA included. No firewall holes, no site visits, no six-month IT project.

SENSORS PLC EDGE TUNNEL ENTERPRISE
// The data path01

From the milliamp loop to the boardroom, one hop at a time.

Plant data already flows into your PLCs; that part works. The gap is everything above: getting that data off the floor, into shape, and up to the people and systems that need it, without exposing the control network to the internet. The edge device closes that gap from the middle.

CENTRAL · CORPORATE PLATFORM analytics · dashboards · ERP · quality · ML IT corporate + plant firewalls zero inbound ports · no NAT · no VPN box Cloudflare Tunnel / private link outbound-only 443 · mTLS · encrypted EDGE DEVICE · FANLESS INDUSTRIAL PC SCADA / HMI screens · alarms · trends HISTORIAN store & forward MODEL ENGINE contextualize · enrich BROKER MQTT · OPC UA · REST WAN OT LAN EtherNet/IP PROFINET · S7 OPC UA · MQTT CONTROLLOGIX Rockwell · 1756 · CIP SIMATIC S7 Siemens · TIA Portal GROOV EPIC Opto 22 · edge I/O 4–20 mA · HART DeviceNet · ControlNet Modbus RTU/TCP · IO-Link TRANSMITTERS temp · pressure · level · flow DRIVES & MOTORS VFDs · soft starters · energy VALVES & I/O solenoids · limit switches · scales // nothing below the edge device changes: your PLCs, networks, and field wiring stay exactly as they are
// field → PLC → edge → tunnel → enterprise · data climbs, nothing reaches back down
tap a component for the technical detail
hardware fanless · solid-state · DIN rail environment panel-rated · wide temp networking dual GbE · OT/WAN separated protocols EtherNet/IP · PROFINET · OPC UA · Modbus · MQTT install 2 cables · same day
// Duplicate IPs, solved at layer 202

Every OEM skid ships as 192.168.1.x. The edge still sees them all.

Machine builders deliver every skid with the same address plan, so a plant with ten packaging lines owns ten copies of 192.168.1.10. Re-addressing a validated OEM program to fix that is risky, and it breaks the builder's remote support along with more than a few warranties. So the skids stay untouched: an industrial NAT switch on each cell (Cisco IE-3400 / Rockwell Stratix class) translates the machine's private subnet to a unique block on the plant network, and the edge device ingests every PLC as if the conflict never existed.

EDGE DEVICE polls 10.10.21.10 · 10.10.22.10 · 10.10.23.10 plant OT network · one routable space 10.10.0.0/16 L2 NAT 192.168.1.x ⇆ 10.10.21.x 10.10.21.0/24 L2 NAT 192.168.1.x ⇆ 10.10.22.x 10.10.22.0/24 L2 NAT 192.168.1.x ⇆ 10.10.23.x 10.10.23.0/24 IE-3400 / Stratix one per cell LINE 1 · OEM SKID PLC .10 HMI .11 VFD .12 192.168.1.0/24 LINE 2 · OEM SKID PLC .10 HMI .11 VFD .12 192.168.1.0/24 LINE 3 · OEM SKID PLC .10 HMI .11 VFD .12 192.168.1.0/24 // three skids, one identical address map on all of them: duplication by design, from the machine builder // ten lines can own ten copies of 192.168.1.10: translated once per cell, unique everywhere above
// same skid · same program · same IPs → L2 NAT gives each cell a unique identity on the plant network

The skid keeps its image

The OEM program, its backups, its HMI addressing, and the builder's remote-support setup all stay byte-for-byte as delivered. Commissioning spares, cloning a line, or restoring after a failure works from the standard OEM image, because the plant network adapts to the machine instead of the reverse.

skid config · untouched

One small table per cell

The NAT switch holds a short static map: each inside address pairs with one outside address, translated in hardware at line rate. EtherNet/IP, PROFINET, and plain TCP flow through it transparently, so the PLC never knows it has a second identity.

192.168.1.10 ⇆ 10.10.21.10

A clean namespace upstream

The edge device polls each PLC at a unique, stable plant address, so the model engine can bind Line 1's filler and Line 2's filler as distinct assets with identical models. Adding line eleven is a new NAT block and a copied model, finished before lunch.

one /24 per cell · zero conflicts
// SCADA + aggregation03

Raw tags are noise. The edge turns them into answers.

A PLC will happily give you ten thousand cryptically named tags at machine speed. Streaming that to a data lake produces a swamp. The edge device does the refining on site, the way an intelligence-hub architecture should: collect close to the source, model it into named assets with units and context, then publish clean payloads that analytics teams can use without a plant-floor translator.

// what the PLC says… Line4_EVP_TT101.Val = 154.2 Line4_EVP_PP210_VFD.Freq = 52.5 B3:6/12 = 1 // …and what the enterprise receives: modeled, in context, joined with the running order { "site": "plant-04", "area": "evaporation", "line": "Line 4", "asset": "Evaporator-1", "measures": { "product_temp": { "value": 154.2, "uom": "degF", "quality": "GOOD" }, "supply_pump_speed": { "value": 52.5, "uom": "Hz" }, "cip_active": true }, "context": { "order": "00550123", "product": "PT00089518", "mode": "CIP · caustic wash" }, "ts": "2026-07-22T14:03:11Z" }

SCADA that lives on site

Operator screens, alarming, and trending run locally on the box. When the WAN drops, the plant keeps its eyes: screens stay live, alarms still fire, and history keeps recording. The cloud consumes the plant's data without ever becoming a dependency for running it.

local screens · local alarms

Store and forward

Every value is buffered on solid-state storage before it's shipped. An outage of minutes or days replays in order when the link returns, with quality codes intact. Enterprise dashboards see a gap close itself; nobody re-keys anything and no history is lost.

buffer → replay → no gaps

Models instead of tag dumps

An intelligence-hub layer (we deploy HighByte, or the modeling engine your stack prefers) maps raw tags into reusable asset models: an "Evaporator" looks the same in every plant, whatever the PLC underneath calls its tags. Change a tag address and the model absorbs it; the enterprise never notices.

tag → model → payload

Every dialect, both directions

Native drivers for EtherNet/IP, PROFINET, OPC UA, and Modbus on the plant side; MQTT with Sparkplug B, REST, and SQL on the enterprise side. Legacy serial gear rides in over gateways. One box retires the drawer full of one-off protocol converters.

EtherNet/IP ↔ MQTT/Sparkplug B
// Secure by design04

The plant network never touches the internet. The data still arrives.

Every connection the edge device makes is outbound: it dials home over an encrypted tunnel (Cloudflare Tunnel, or a private link where policy requires) and keeps that session open. Nothing on the internet can dial in, because there is nothing to dial: no inbound firewall rules, no port forwards, no public IP, no VPN concentrator to patch. Your IT team approves one outbound HTTPS flow, and that's the whole conversation.

Outbound only

The tunnel is established from inside the plant, outward, on port 443. Attack surface from the internet: zero listening ports. The same session carries telemetry up and authorized management traffic back down the established, authenticated channel.

no inbound rules · port 443 out

Identity over addresses

Each device authenticates with its own certificate; every session is mutually authenticated TLS end to end. Access to the device's services is governed centrally by identity and policy, so "who can see plant 4" is a rule you set once, and revoke in one click.

per-device mTLS · central policy

OT and WAN, physically split

Two network interfaces, no routing between them. The control network stays flat, private, and invisible from the WAN side; the only thing that crosses the gap is data the model engine explicitly publishes. The edge device is a diode with a brain.

dual NIC · no forwarding

A managed fleet

Health, updates, and configuration ride the same secure channel. We watch every device centrally, patch the OS and containers remotely, and roll out a new data model to twenty sites in an afternoon. Site 21 gets the same treatment the day its box comes online.

monitored 24/7 · managed remotely
// The install05

Commissioned in a morning.

The device ships preconfigured for your site. The on-site work is electrician-simple; everything after the cables is done remotely by us, over the tunnel, with your team watching if they want to.

01Mount it and power itDIN rail or panel mount, 24 VDC, fanless and solid-state: it lives happily in the same enclosure as the PLC~15 min
02Connect two cablesLAN port to the industrial switch, WAN port to the enterprise network (or a cellular modem where there isn't one)~5 min
03It phones homethe outbound tunnel comes up and registers with central monitoring: no firewall tickets, no static IP, no port forwardsautomatic
04We browse the plant, remotelyPLC connections come online, tags are discovered, and the asset models for your site get applied and tunedsame day
05Data flowslocal SCADA live on site, contextualized streams landing on the central platform, dashboards filling inday one
// per additional site: repeat. the models, dashboards, and security posture are already built
// what the customer gets

The device is the delivery mechanism. What you're actually buying is the outcome on the enterprise side:

  • Analytics on data that explains itself. Every value arrives named, unit-tagged, and joined to its asset, line, and running order, ready for ERP, quality, and ML systems to consume.
  • A start measured in days. No plant IT project, no rip-and-replace: the box sits beside what you already own and starts talking to it.
  • History that survives outages. Store-and-forward buffering means the record at the center is complete even when the network wasn't.
  • A security posture IT will sign. One outbound HTTPS flow to approve, zero inbound exposure, and the OT network stays dark from the outside.
  • A pattern that scales. Each new site is the same box, the same models, the same tunnel: plant 12 costs a fraction of plant 1 and matches it exactly.
// Contact06

Want your plant's data on a dashboard by next month?

Tell us what's on your plant network: which PLCs, how many lines, where the data should land. We'll spec the box, the models, and the tunnel, and put a pilot on your bench.

Start a conversation