Choosing the right architecture for an industrial IoT setup

Written by Davide Ferraretto

In an industrial IoT project, the most important question isn't about the cloud or the final dashboard, but how data actually gets in from the field: which sensors, which protocol, which connectivity. The choices made at this stage determine the cost and reliability of the system far more than the technology chosen downstream.

Start from constraints, not from technology

Before choosing a protocol or a cloud platform, it pays to analyze the actual constraints of the plant: how many measurement points, at what frequency, in what environment (temperature, interference, presence of metal), and whether wired, local wireless or cellular connectivity is needed. A plant with hundreds of low-frequency sensors has very different needs from a handful of high-frequency sensors used for process control.

Choosing the right protocol

MQTT remains the most common choice thanks to its lightweight publish/subscribe model, but it isn't always the best fit: for network device diagnostics, SNMP is often already integrated into existing supervision tools, while in older industrial environments it's still common to have to talk to legacy serial protocols through dedicated gateways.

Edge computing: when it makes sense

Not all data needs to reach the cloud raw. Filtering, aggregating or making critical decisions directly on-machine (edge) reduces the bandwidth required and keeps the system running even with unstable connectivity, a common scenario on industrial sites. The choice between edge and pure cloud should be made case by case, based on required latency and control criticality.

A well-designed industrial IoT project starts from careful field analysis, not from a sensor catalog: truly understanding the production process before writing a single line of firmware or dashboard code is what reduces cost and risk in the later stages.