The business problem
SkyBridge operates a fleet of regional turboprops. Each engine has thousands of hours of sensor history (temperatures, pressures, vibrations, oil chemistry). Maintenance is currently scheduled by hours-since-overhaul — a one-size-fits-all rule that's safe but expensive. SkyBridge wants a model that predicts remaining useful life for each engine so they can pull engines for service when the data says so, not when the calendar says so.
This case applies Decision 2 (Regression) at industrial scale, with the BizML lifecycle for thinking about deployment.
The cost asymmetry matters
Predict failure too early → waste good engine life. Predict too late → catastrophic in-flight failure. The model isn't optimized for accuracy alone; it's optimized for the dollar (and life) cost of being wrong.
Case kit
Everything you need
- Engine Maintenance DatasetSensor histories from a fleet of engines, with known failure events.
- BizML Lifecycle NotebookFull BizML pipeline: business framing, data, modeling, deployment, monitoring.
- Google Sheets / Excel ScorerScoring tool for line maintenance to use without writing code.
- Feature Names (JSON)Feature schema for re-implementation in another platform.
Sample presentation
- BizML Lifecycle — Reference DeckHow to talk about the project across the lifecycle — framing, modeling, deployment, monitoring.
Topics you'll be applying
- Decision 1 — Data Prep & EDA — sensor data needs heavy preprocessing
- Decision 2 — Regression — predict remaining useful life (a number)
- Decision 5 — Anomaly Detection — early-warning features can come from anomaly scores
- Decision 7 — Time Series — sensor history has a strong time dimension