A CNC machine runs on a control loop of 1–10ms. A collaborative robot on an assembly line is similar. At those tolerances, a cloud round-trip on a fast corporate network adds an order of magnitude too much delay. The control logic runs at the machine because nothing else is fast enough.
The multi-site problem
A single-plant manufacturer manages one set of control systems. A manufacturer running 20 plants manages 20 sets, each with different network conditions, different equipment vintages, and — without a governed deployment model — different versions of the same control logic on nominally identical machines.
Firmware version drift is how unpatched security exposure accumulates. It is also how a failure pattern at one plant goes unnoticed at the other 19 until a machine stops.
Edge-resident predictive maintenance catches these patterns fleet-wide. Failure signatures at one plant get flagged across every plant running the same equipment, before any of those machines stop.
What a governed deployment model looks like
At each plant, an edge box connects to the production equipment, runs control logic and monitoring applications locally, and sends structured data upstream to a central platform. The central platform manages versions, deploys updates, and tracks which machines are running what.
When a bearing on a specific press type fails at plant three after 14 months, the fleet-wide model identifies that the same press type runs at 17 other plants and schedules preventive maintenance across all of them before any fail. That is the operational difference between a machine monitoring system and a fleet management architecture.
The EU Cyber Resilience Act in manufacturing
The CRA began enforcement in September 2026. It requires manufacturers to maintain certificate-based device identity across the full product lifecycle, from provisioning through decommissioning. A manufacturer running hundreds of PLCs, HMI terminals, and edge gateways across 20 plants accumulates those devices faster than any manual identity management process can track.
Each unenrolled device is an unverifiable node on the production network. An edge platform with certificate-based identity built into the runtime closes that gap as a normal operational property, not as an audit preparation exercise.
When edge computing makes sense in manufacturing
Batch reporting, quality data aggregation, and production scheduling can run in the cloud without consequence. Edge architecture matters for real-time control loops (CNC, robotics, assembly), predictive maintenance on high-value equipment, safety applications where response time is the constraint, and compliance logging where data must stay within the facility.
For multi-site manufacturers, there is a governance argument on top of the latency argument: one platform managing control logic, security, and updates across every plant beats 20 separate implementations that drift apart over time.
Back to the main guide: Industrial edge computing, what it is and how it works
By Helin