Six things we learned deploying 1,200 IoT sensors in a factory

The architecture diagram looked excellent. Clean lines from sensor to gateway to cloud. Then we walked onto the factory floor.

1. The environment will break everything you tested indoors

Vibration is pervasive and relentless in a factory. We spent two weeks characterising the environmental noise profile of each machine zone before we could distinguish normal from failure signatures.

2. Test RF propagation under production conditions

Wi-Fi at 2.4GHz and metal machinery are not friends. The interference pattern changed completely when the production line was running. Always do your wireless site survey under production conditions, not idle ones.

3. MQTT QoS levels are a tradeoff

QoS 0 is fast but lossy. QoS 2 is reliable but slow. In a 1kHz sensor network, QoS 0 is usually right for telemetry. For threshold alerts, QoS 1 is mandatory.

4. Edge processing is load management

Streaming all raw data to the cloud: 1,200 sensors × 1kHz × three channels = infrastructure costs that killed the project economics. Edge processing on gateways cut data volume by 95% while preserving everything the anomaly detection models needed.

5. Alerts without context create alert fatigue

The alerts that get acted on include: what crossed the threshold, by how much, the trend over 24 hours, and a suggested action. Build the context in from the start.

6. Plan OTA firmware updates before you need them

Updating 1,200 devices post-deployment required staged rollouts, automatic rollback, and canary deployment to 5% of devices first. None of that was in the original scope.