AISaudi Arabia2024
Predictive-maintenance ML for rotating equipment
Streaming vibration and process models that call bearing and pump faults weeks ahead.
- Client
- nybl
- Region
- Saudi Arabia
- Sector
- Industrial predictive-maintenance ML
- Engagement
- 2024 · 21 weeks
- Team
- 2 ML · 1 data · 1 MLOps
- Status
- In production
The challenge
Unplanned failures on pumps and compressors dominated maintenance cost, and threshold alarms fired too late to prevent secondary damage. nybl needed time-series models that detect degradation early from existing sensor streams, with a pipeline operators could trust.
What we built
The full pipeline, end to end — not a blurb.
- 01Data
Historian integration over OPC UA with a feature pipeline and drift monitoring.
- 02ML
Multivariate anomaly and remaining-useful-life models on vibration and process tags.
- 03MLOps
A retraining pipeline with shadow validation, per-asset thresholds, and a model registry.
- 04Deployment
Alerts into the CMMS with explainable fault attributions.
Results
18 days
average lead time on bearing faults before failure
0.89 / 0.86
precision / recall on labeled historical failures
27%
reduction in unplanned downtime on covered assets
41%
fewer false alarms than the prior threshold system
Feeds the maintenance planning cycle across plants.
Have a constraint like nybl’s? Bring us yours.
Next engagement
Blickfeld
Solid-state LiDAR / silicon