Releasedv1.0.0Predictive Maintenance · MLOps
Production ML Platform for Predictive Maintenance
End-to-end predictive maintenance platform demonstrating production ML engineering, evaluation discipline, MLOps, and API deployment.
End-to-end failure-risk scoring: leakage-safe features, chronological evaluation, LightGBM vs baseline, FastAPI serving.
Problem
Unplanned downtime is expensive
Failure signals hide in noisy sensor streams
Notebook models rarely reach production packaging
Hiring managers need evaluation discipline, not accuracy theater
Solution
A production-shaped failure-risk platform
- Validated data → features → baselines → primary model
- Chronological holdout with F1 / PR-AUC focus
- FastAPI + Docker + presentation pack for review
Pipeline
Interactive system flow
Selected stage
AI4I CSV
Load and validate machine telemetry records.
Loading Interactive Demo…
Loading Performance…
Engineering Quality
Production packaging signals
Integrity
Engineering decisions & limitations
- Failure-mode target flags excluded from features (no leakage)
- Chronological split by UDI — not random shuffle
- AI4I 2020 is synthetic — not a live plant warranty
- Rare failures: F1 / PR-AUC prioritized over accuracy
- Optional X-API-Key; disabled when API_KEY=change-me
Loading API Explorer…