Predictive maintenance pilot

A sensor analytics pilot translating vibration and temperature readings from heavy plant equipment into early failure warnings, moving maintenance from a fixed schedule to an evidence-based one.

Timeline: March 2024 - November 2024
Unplanned downtime: Down 34%
Average early warning: 11 days

A tablet showing live sensor readings mounted beside industrial machinery. — Image adapted from the CIDS project image set for the project detail page.

We used to replace parts on a calendar. Now we replace them when the data says to.

ML

Maintenance lead

Resources partner

Recognition and proof

Project signals

Industry partners reviewing equipment maintenance plans together.

Operational impact

Reduced unplanned downtime by a third

Vibration and temperature signals flagged bearing wear an average of 11 days before failure, giving maintenance crews a scheduled window instead of an emergency one.

A production line connected to ongoing monitoring work.

Rollout

Extended to a second plant

After a six-month trial at one site, the partner asked CIDS to extend sensor coverage to a second facility using the same model architecture.

Project overview

From scheduled servicing to condition-based maintenance.

Heavy plant equipment at the partner’s site was serviced on a fixed calendar, regardless of actual wear. That meant some parts were replaced early, wasting spend, while others failed between scheduled visits, causing unplanned downtime.

CIDS fitted a subset of machines with vibration and temperature sensors, then built a model to flag developing faults against a baseline of normal operating signatures. Alerts routed straight to the maintenance team’s existing ticketing system, so no new tooling was required on their side.

The pilot ran for six months against one production line before the partner asked to extend it. Model thresholds are now tuned per-machine rather than using a single site-wide baseline, which cut false positives substantially in the second phase.

A wall-mounted dashboard summarising equipment health across a production line. — Each machine's health score updates hourly, calculated from a rolling window of sensor readings rather than a single instantaneous reading.11-day average early warning

Before and after

2 operational shifts from the project

Predictive Maintenance Pilot

Before

Maintenance approach

Fixed calendar servicing regardless of actual wear.

Unplanned downtime

Failures discovered when equipment stopped.

After

Maintenance approach

Condition-based servicing triggered by sensor signals.

Parts are replaced closer to when they actually need it.

Unplanned downtime

Faults flagged an average of 11 days ahead.

Maintenance crews get a scheduled window instead of an emergency one.