Karma Electronics
All projects

Industrial IoT

Predictive-maintenance sensor gateway

A proof-of-concept cellular gateway that fits sensors to commercial fan packs and streams the data to the cloud, where it is combined with operating data to spot failed fans and predict the next failure.

Industry
Commercial fan systems
Our role
Architecture, sensors, firmware, cloud data
Connectivity
LTE Cat-M1 cellular
Cloud
AWS backend and analytics

  1. Sensors

    Vibration, temperature, current, air pressure

  2. Gateway

    Carrier board and cellular module

  3. AWS cloud

    Sensor data joined with run hours and power use

  4. Dashboards

    Fan health and failure alerts

The challenge

The client wanted to know when a fan had failed, and ideally before it failed. Speed to deployment mattered most, so the design leaned on certified off-the-shelf modules rather than new radio hardware.

Hardware

A gateway built on an FCC-certified module with Cat-M1 cellular, Wi-Fi and BLE, plus a carrier board adding a dedicated processor for sensor I/O, signal conditioning and power. Sensors mount in IP65 boxes with panel connectors and an I2C extender for long cable runs.

Firmware

High-level APIs in MicroPython to shorten development, sampling accelerometers in the kilohertz range and slower channels every five minutes.

Cloud and software

Sensor data sent over cellular to an AWS backend and combined with the client's own power-consumption and run-hour data to detect failed fans and predict failing ones.

Key decisions

  • LTE Cat-M1
  • MicroPython
  • ADXL372
  • SPI and I2C
  • Signal conditioning
  • AWS

Client name and confidential details are withheld under NDA.

  • Three platforms comparedA certified cellular module, a Raspberry Pi with a cellular HAT and an Orange Pi 4G board were scored on antenna options, parts supply, BOM cost, example code and development time. The certified module won on speed and supply.
  • Sampling fast enough to see a faultFans spin at about 25 Hz, so vibration is sampled at 3.2 kHz with 12-bit, ±200 g accelerometers, up to eight per gateway over SPI.
  • Ruling out ultrasonicUltrasonic sensors work above 20 kHz, outside the useful range for this fault signature, so they were dropped.
  • Slow channels for contextTemperature, AC current and differential air pressure (±500 Pa) are sampled every five minutes.

Fan fleet dashboard

A fleet view of one gateway: vibration for each fan, a developing fault on fan 6, and the message the gateway sends up.

Mockup with simulated data
Fan fleet dashboardUpdates every few seconds

This dashboard is an illustration built for this site. Every value is generated in your browser; no client data is shown.

Building something similar?

Tell us about your product, from the board to the dashboard. We'll reply with next steps and a rough plan.

Request a meeting Email us