Mines and remote sites
Backhaul and site telemetry for mines and remote industrial sites, engineered per site survey rather than assumed coverage. Walk the pit, the stockpile and the haul road with a signal tool, then decide where a fixed link, a mast or a repeater is needed. Telemetry from pumps, conveyors and dewatering is small and steady; video from the weighbridge or the gate is bursty and upstream. Put them on separate profiles so a video session cannot starve the pump data, and keep a local buffer at the gateway for the minutes when the link is gone.
Solar and storage plants
Per-string and per-unit telemetry on narrow always-on links, with maintenance traffic on the same profile. The devices are data loggers behind inverters and battery management systems, and their traffic is a poll every few minutes plus a backfill after an outage. Size the plan from the logger count and the polling interval, not from the megawatt rating. Where the site also has a camera or a weather station, give it its own profile: a 5 GB monthly budget shared between a logger and a video stream is a budget nobody can forecast.
EV charging networks
A charger is a payment terminal with a power stage. It boots, registers with the charging management system, sends a heartbeat and authorizes a session before the first kWh flows; metering values arrive during the session and a final meter reading closes it. That sequence is OCPP over a persistent connection, so the design question is what happens when the connection drops mid-session: the charger should finish locally and reconcile later, not refuse to charge. Charger telemetry and payment traffic can share a data estate, but authorization and firmware updates deserve separate budgets. The authorization path is the same problem as a payment terminal at a counter, with worse weather.
Industrial gateways
Gateways aggregating site devices onto private APNs, with drone-inspection video handled on the robots and drones page. A gateway polls Modbus registers or OPC UA nodes, publishes the result to the site platform and buffers when the link drops. Keep it outbound-only: no inbound port, no shared password, no route from the gateway into the office LAN. Where a site runs DNP3 or IEC 61850, tunnel the protocol through a device that understands its timing, and expect the polling pattern, not the bandwidth, to set the latency budget.
Safety and environmental monitoring
Slope movement, gas, dust, water level and weather sensors are the traffic nobody watches until an alarm fires, so the alarm path is the part to design first. Send routine readings on a slow cycle and alarms immediately, with a local rule that raises the alarm even when the uplink is down. Duplicate the path where the consequence is severe: a second SIM on a different network, a satellite backhaul or a wired line, chosen by what the site actually has. None of this replaces a safety case; it only makes the measurements available.
