Braeburn Power

Background

Braeburn operates a fleet of gas turbines and had a digital-twin platform that, by their own admission, had become shelfware — a one-time model that drifted from the real units within weeks of commissioning and that nobody trusted enough to act on.

The challenge

Their twin didn't assimilate live data, so it diverged from the asset and its alarms were noise. The operators wanted a twin that tracked the real machine closely enough to forward-roll degradation and tell a genuine fouling or wear trend apart from a flaky sensor — before a unit tripped, not after.

What we did

• Rebuilt the twin on Aether from the existing thermodynamic model and CAD, then wired live telemetry through data assimilation so it stayed synced to each unit.
• Forward-rolled compressor and hot-section degradation against the physics, with calibrated uncertainty on the time-to-threshold.
• Cross-checked anomalies against the physics twin to separate a real performance trend from sensor drift, instead of alarming on both.
• Surfaced predictions as ranked maintenance actions with the evidence — the simulated trajectory — attached.

The outcome

• The twin flagged a compressor fouling trend on one unit roughly six weeks before it would have forced an unplanned trip, with a wash recommended inside a maintenance window.
• It tracked the real units instead of drifting, so operators trusted it enough to schedule on its output.
• Sensor-fault anomalies were correctly separated from performance trends; no false alarm was acted on during the window.
• The twin now runs continuously across the fleet rather than sitting as a commissioning artifact.

One detail worth calling out

The credibility moment was replaying a historical trip: the twin, fed only the data available before the event, reproduced the degradation trajectory that led to it. Once operations saw it predict an event they remembered, the shelfware label came off and they let it schedule real work.