Use cases

Industrial Automation

Sensor fusion for predictive maintenance, hard-real-time safety interlocks, and operator dashboards — on the factory floor where milliseconds are safety.

On a factory floor, “latency” stops being a UX metric and becomes a safety property. A light curtain that halts a conveyor in 8 ms protects a hand; one that takes 80 ms doesn’t. The Wireless Bobulator was built for exactly this seam — fusing dozens of sensors, making the call deterministically, and proving in the audit log that it made the right one.

The scenario: a packaging cell that protects itself

A single cell has a conveyor, a robotic arm, vibration sensors on two motor bearings, a light curtain at the load point, and an operator HMI. Three jobs run at once — predictive maintenance, a safety interlock, and the live operator view — all from one Bobulator:

rule "bearing-predictive-alert":
  priority: 60
  when:
    - ml.anomaly_score(sensor.vib.motor_a) > 0.82
    - rolling_avg(sensor.temp.motor_a, 5min) > 70
  then:
    - dashboard.alert("Motor A bearing — schedule service", level=WARN)
    - maintenance.ticket.open("motor_a", severity=2)

rule "load-point-interlock":
  priority: 100
  when:
    - sensor.curtain.load == BROKEN
  then:
    - conveyor.halt(reason="curtain")
    - arm.safe_stop()
    - dashboard.alert("E-STOP: load curtain", level=CRIT)

The interlock at priority: 100 outranks everything — production schedules, energy savers, the lot. When the curtain breaks, the conveyor halts and the arm safe-stops on the same engine cycle, under 8 ms, and the event is hash-chained into the audit log for the incident report. Meanwhile the on-device ML model watches Motor A’s vibration signature and opens a maintenance ticket days before the bearing fails — no cloud round-trip, inference running locally on the Cortex-M7.

Why it matters here

  • Hard real time, not best effort. The safety path lives on the deterministic core, isolated from the web stack — a slow dashboard query can never delay an interlock.
  • Predictive, on the edge. Vibration and thermal fusion runs a quantized model on-device; nothing leaves the cell, and there’s no internet dependency for a safety-adjacent function.
  • Proof, not vibes. Every interlock and alert is logged with its trigger, latency, and conflict set — exactly what an auditor or an OSHA reviewer asks for.
  • Survives the network. Run the line as a mesh; lose the uplink and every cell keeps deciding locally.

The same fusion-and-decide loop that lights a court keeps a hand out of a pinch point. Read how the engine ranks a safety rule over a schedule.

copied!