Edge Computing for Industrial IoT

Edge computing keeps a machine running and a cloud bill predictable by processing data where it is produced, on or beside the equipment, instead of shipping every reading to a distant data center first. The edge is the factory floor, the remote pump station, the haul truck, the wind turbine. The computer that does the work is a gateway or industrial PC sitting on the same panel as the PLC.

It is not the opposite of the cloud. It is a division of labor. The edge handles what has to be immediate or local; the cloud handles what benefits from scale and central storage. Almost every system that survives contact with the field uses both.

Four pressures decide how much computation belongs near the machine:

• Bandwidth and cost. A vibration sensor sampled at 10 kHz produces megabytes a minute. Sending that raw over a metered cellular link is expensive and pointless. Compute an RMS value, an FFT peak, or a threshold crossing at the edge, and you send a few bytes instead of a stream.
• Latency. A round trip to a cloud region is tens to hundreds of milliseconds, and that is when the network is healthy. If a reading has to trip an interlock or stop a motor, the decision has to be made on the gateway, not over a WAN.
• Offline resilience. Cellular coverage drops in tunnels, basements, and rural sites. An edge node with store-and-forward buffering keeps acquiring, timestamping, and queuing data through the outage, then replays it in order when the link returns. Nothing goes blind, and nothing is lost.
• Data gravity. Some data is large, regulated, or simply not worth moving. Filtering it in place and sending only the result is often the only option that holds up under a real connectivity and compliance budget.

Decide at the edge, learn in the cloud. That single line settles most of the architecture.

On the gateway: reading field protocols (Modbus on an auxiliary pump, J1939 on a drivetrain, OPC-UA off a newer PLC), normalizing tag names and units, applying local thresholds and alarms, buffering through outages, and any control loop that cannot tolerate network delay. In the cloud: long-term storage, cross-fleet analytics, dashboards spanning many sites, model training, and integration with ERP and CMMS. The edge acts on the moment; the cloud finds the pattern across thousands of moments and assets, then pushes refined logic back down as an over-the-air update. Where you draw the line is a cost and risk decision, not a fixed rule, and it usually shifts toward the edge as data rates climb and links get worse.

For most industrial deployments the edge node is a gateway, not a server rack. An open Linux gateway reads the machine's protocols, runs the filtering and local logic as ordinary software (often in containers), buffers through outages, and publishes the distilled result upward over MQTT with store-and-forward so a dropped link costs nothing.

Because it is a general-purpose computer, the same hardware grows from simple data collection to on-device analytics by changing the software, not the box. You add an inference model or a new alarm rule with an OTA update, and the gateway in the field starts doing more without a truck roll. Explore the hardware on routers and gateways, and the platform side on IIoT platform and data acquisition pipelines.

The pattern is consistent across deployments: field devices on their native protocols, an edge gateway that reads and processes them, and a platform that receives the distilled stream for storage, dashboards, and analytics.

What is edge computing? Processing data near its source, on a gateway or local computer, instead of sending all of it to a central cloud. The edge handles immediate and local work; the cloud handles storage and scale.

What is the difference between edge and cloud computing? Location and role. The edge runs next to the machine for low latency and to cut what travels over the link. The cloud runs centrally for storage, cross-site analytics, and heavy processing. They complement each other, and most real deployments use both.

What is an edge gateway? A gateway that does not only move data but processes it locally: filtering, normalizing, applying logic, and buffering with store-and-forward before publishing results to a platform over MQTT.

Do I need edge computing? If your assets are remote, your data rates are high, your decisions are time-critical, or your connectivity drops, yes. If everything is online and low-volume, sending raw data to the cloud may be enough.