The challenge
The works maintains its pumps and blowers on fixed intervals. Some are opened up while still healthy; others fail between services, usually at night or in a storm, when inflow is highest.
An unplanned failure means emergency call-outs, hired standby pumps, overtime and sometimes a spill that has to be reported. The parts that wear out (bearings, seals, impellers) give warning signs for weeks, but nobody is watching for them.
Our approach
-
Criticality ranking
With the maintenance team, we rank every pump and blower by what its failure costs, so sensors go first on the assets where an early warning pays back fastest.
-
Instrument the assets
Vibration and temperature sensors go on bearings and motors, and energy analysers measure motor current and power. An edge gateway collects this alongside existing SCADA values like flow and pressure.
-
Learn normal behaviour
Models learn each machine’s normal signature across its duty points over several weeks. Rising vibration, creeping temperatures or more power for the same flow then stand out as early signs of wear.
-
Turn alerts into work orders
Each alert names the asset, the likely fault and how urgent it is, and goes into maintenance planning. The team records what technicians actually find, and that feedback tunes the models.
System architecture
How the layers connect, from the people using the system down to the equipment in the field.
- Maintenance planner
- Technicians
- Plant manager (monthly asset health report)
- Machine-learning model per asset
- Health scores
- Alert triage
- Historian
- Existing SCADA values: flow, pressure, run hours
- Edge gateway
- MQTT and OPC UA publishing
- Vibration
- Bearing and winding temperature
- Energy analysers on motor feeds
What we’d deploy
- Vibration and temperature sensors on critical pumps and blowers
- Energy analysers measuring motor current and power
- Edge gateway combining sensor and SCADA data
- Historian storing detailed trends for every asset
- Machine-learning models with a health score per asset
- Alerts linked into the maintenance planning workflow
Target outcomes
Targets we would design and report against, not measured results from a named client. Actual results depend on the baseline established in the audit.
