Lab/Job sheets
12 job sheets

A problem, the logic, and the action it earns.

Each sheet states what is wrong, which sensing principles are involved, the decision rules in plain language, what the hardware does without asking, and the record that survives the event. Thresholds are examples to be tuned per site.

01

Catch a fire while it is still a smell

Server room

The problem

A rack fan fails, a capacitor overheats, and the first alarm is a smoke detector in the corridor after the room is already lost.

Sensing

The logic

  1. Sample smoke every second; require 3 consecutive frames over 3 %/m, not one spike.
  2. Compare the thermal array against the inlet/outlet delta: a hot pixel 15 °C above its neighbours is a rack-local fault.
  3. Watch the VOC trend over 10 minutes; a slow rise with a flat smoke reading is the precursor window.
  4. If smoke is confirmed OR (hot pixel AND VOC rising), latch a local alarm and cut non-essential rack power via a contactor.

What the system does

Local strobe and buzzer, contactor drops non-critical circuits, a dry contact releases the fire panel, and a message goes out with the rack and the pixel.

What you get

The event is caught in the precursor window, the response is local and immediate, and the log shows which rack and which sensor spoke first.

02

Find water before it finds the servers

Server room

The problem

Condensate from a CRAC unit pools under a raised floor and reaches a PDU before anyone walks in.

Sensing

The logic

  1. Zone the leak rope: each 5 m section is its own digital input, so the alarm names the section.
  2. Treat a flood probe as the confirmation; a wet rope plus a dry probe is a spill, not a flood.
  3. Track dew point against the chilled-water pipe temperature; condensation risk is a computed number, not a guess.
  4. On a wet zone, stop the condensate pump, close the chilled-water valve and start the extract fan.

What the system does

Valve closes, pump stops, local alarm sounds, and the message names the zone and the time the rope first got wet.

What you get

A leak is localised in metres, the source is shut off, and there is a timestamped wet/dry history for the insurance report.

03

Hold the room at temperature when a fan dies

Server room

The problem

One CRAC fan fails on a Friday night; the room climbs 1 °C every eight minutes and nobody knows until Monday.

Sensing

The logic

  1. Read six temperature nodes: inlet, outlet, cold aisle top and bottom, hot aisle, and outside the door.
  2. Derive the room gradient; more than 5 °C top-to-bottom means airflow, not cooling capacity.
  3. Prove each fan is turning with a current clamp: a commanded fan with no current is a failed fan.
  4. Above 28 °C inlet, ramp the standby fan, open the economiser damper and page; above 32 °C, start a controlled shutdown of non-critical hosts.

What the system does

Standby fan ramps, damper opens, alert escalates, and the shutdown order is staged rather than panicked.

What you get

The room rides through a single fan failure, and the trend shows the failure before the temperature reaches the servers.

04

Detect the gas before it is an explosion

Server room

The problem

A battery string starts to off-gas, or a generator test pushes exhaust into the intake, and the room has no nose.

Sensing

The logic

  1. Catalytic sensor watches 0–100 % LEL; alarm at 10 % LEL, trip at 20 % LEL.
  2. Hydrogen at the ceiling confirms battery off-gassing; CO at the intake confirms exhaust ingress.
  3. Oxygen below 19.5 % means displacement, not combustion; treat it as a different emergency.
  4. Above 10 % LEL: force ventilation, inhibit battery charging, drop non-essential power and alarm locally.

What the system does

Ventilation runs at full speed, the charger is inhibited, a dry contact signals the fire panel, and everyone gets a message that names the gas.

What you get

The room distinguishes a battery fault from exhaust ingress, and the first action removes the ignition risk instead of adding to it.

05

Prove the walk-in stayed cold

Cold chain

The problem

A door is left ajar, the compressor is running but not cooling, and the first evidence is spoiled stock.

Sensing

The logic

  1. Two temperature probes: air and product-simulant, because air recovers fast and product does not.
  2. Door contact starts a timer; more than 3 minutes open is a door alarm, not a temperature alarm.
  3. Compressor current proves it is running; a running compressor with a rising temperature is a refrigerant fault.
  4. Log every 60 s with a rolling 30-day record; alarm at 4 °C for 10 minutes or 8 °C for 2 minutes.

What the system does

Alarm at the door and at the phone, with the graph attached; the log is exportable for the audit.

What you get

A door event and a refrigeration fault look different in the record, and the temperature history survives the incident.

06

Interlock the tools to the extraction

Workshop

The problem

The extractor fills up, airflow drops, and people keep cutting because nothing tells them the air is getting worse.

Sensing

The logic

  1. Differential pressure across the filter gives loading; below 80 Pa of draft, the filter is blocked.
  2. Particulate above 150 µg/m³ for 2 minutes with the extractor commanded on means capture has failed.
  3. Current clamp proves the extractor is actually running, not just switched on.
  4. Interlock: remove the run-enable from the tool relay and sound a local horn until airflow is restored.

What the system does

The tool relay drops, the horn sounds, and the dashboard shows filter loading rather than a vague 'air quality' score.

What you get

Dust exposure is controlled by the machine state, not by a person remembering to check a gauge.

07

Water on demand, not on a timer

Growing

The problem

A fixed schedule over-waters after rain and under-waters during a heatwave, and the pump runs dry when the tank is low.

Sensing

The logic

  1. Irrigate when volumetric water content falls below the crop's threshold, not at a clock time.
  2. Scale the dose by light integral and vapour-pressure deficit: a bright, dry day needs more water.
  3. Prove flow with the turbine; a running pump with no flow is a blocked line or a dry tank.
  4. Float switch on the tank inhibits the pump below 10 %; rain gauge suspends the schedule for 6 hours after 3 mm.

What the system does

Zone valves open in sequence, the pump is protected, and the log records litres per zone per day.

What you get

Water use tracks the plant, the pump never runs dry, and every zone has a litres-per-day record.

08

Hear a bearing fail weeks early

Machines

The problem

A motor bearing degrades slowly, then seizes, and the line stops with no warning.

Sensing

The logic

  1. Sample vibration at 1–3 kHz for 2 seconds every minute; compute RMS and the bearing-frequency band energy.
  2. Compare against a baseline learned over the first week of running, per speed.
  3. Trend current and winding temperature: a rising current at constant load is mechanical, not electrical.
  4. Alert when the bearing band rises 6 dB over baseline for 3 consecutive hours; trip at 12 dB or on a shock event.

What the system does

A maintenance ticket with the trend, the speed and the last 24 hours of data; the trip is optional and staged.

What you get

The repair is planned in a window instead of an outage, and the baseline proves the machine actually improved afterwards.

09

One door, two factors, and a room that knows

Access

The problem

A key can be copied, a PIN can be shared, and nobody knows who left the plant room door open.

Sensing

The logic

  1. Require card and PIN within 10 seconds; either alone does not release the lock.
  2. Reed contact confirms the door actually opened after release; a door that does not open logs a forced or stuck event.
  3. PIR confirms a person entered; a card read with no motion is a tailgate or a test.
  4. On exit, start a 60-second re-lock timer; a door held longer alarms and the HVAC setback is suspended.

What the system does

The lock releases, the event is logged with the person and the door, and the room stays in occupied mode until the door closes.

What you get

Entry is attributable, tailgating has a signal, and the environmental log shows exactly when the room was open.

10

Never run a pump dry

Water

The problem

A transfer pump runs with a closed valve or an empty sump and destroys its seal in minutes.

Sensing

The logic

  1. Start only with the low float wet and the discharge pressure below the dead-head value.
  2. Prove flow within 8 seconds of start; no flow means stop and latch a fault.
  3. Watch current against the learned running band: low current is dry running, high current is a jammed impeller.
  4. Stop on the high float, on 3 consecutive no-flow samples, or on any current outside the band for 5 seconds.

What the system does

The contactor opens, the fault latches, and the message names dry-run, blockage or over-current.

What you get

The pump is protected by evidence from three independent signals, not by a single float.

11

Keep the pressure cascade honest

Labs

The problem

A lab door is propped open and the corridor pressure flips, pulling aerosol out of the room.

Sensing

The logic

  1. Measure room-to-corridor differential pressure every second; alarm if it is outside the target band for 10 seconds.
  2. Door contact distinguishes an open door from a failed fan or a blocked filter.
  3. Particulate in the corridor confirms a containment failure, not just a pressure anomaly.
  4. On a sustained loss, sound the local alarm, increase supply air and notify the responsible person.

What the system does

Local alarm plus a message with the pressure trend and the door state; the log shows how long containment was lost.

What you get

The room proves containment continuously, and the record separates human behaviour from equipment failure.

12

Make a power event explain itself

Power

The problem

A server reboots, the UPS beeps once, and by the time anyone looks the evidence is gone.

Sensing

The logic

  1. Sample mains voltage every cycle; record sags, swells and dropouts with a timestamp and a 200 ms pre-trigger buffer.
  2. Log every UPS transition and the battery voltage under load.
  3. Correlate a reboot with a voltage event within 2 seconds rather than blaming the server.
  4. Run an automatic monthly transfer test and record the whole curve, not just pass/fail.

What the system does

A power-event report with the waveform window, the UPS state and the affected circuits.

What you get

The next reboot has a cause attached, and the UPS battery trend is visible before the next outage.

Have a different problem?

The shape is always the same: a condition, a confirmation rule, a local action and a record. Describe the condition and the environment.

Start a brief Sensor atlas