Illustrative engagement — describes the type of work we deliver; client details are withheld.
Business challenge
Unplanned equipment stoppages disrupted production schedules, and maintenance was largely reactive.
Existing environment
- Sensor data stored locally on each line
- Maintenance logs in a separate system
- No shared view of equipment health
Technical challenges
- High-frequency data from heterogeneous machines
- Few labelled failure events
- Alerts that operators would trust
Our approach
- 1Streamed sensor data into a central platform
- 2Joined telemetry with maintenance history
- 3Built anomaly and failure-risk models
- 4Delivered alerts and dashboards to maintenance planners
Architecture overview
- EdgeGateways publishing telemetry to Kafka
- StreamingSpark structured streaming and feature store
- ModelsAnomaly detection and failure-risk scoring
- OperationsGrafana dashboards and alert routing
Technologies used
- Kafka
- Spark
- Python
- AWS
- Grafana
Implementation
- Phase 1
Connect
Pilot line instrumented and streaming
- Phase 2
Model
Baselines and risk models validated with engineers
- Phase 3
Operate
Alerts integrated into maintenance planning
- Phase 4
Extend
Rollout to additional lines
Results
- Earlier warning of developing equipment issues
- A single view of equipment health across lines
- Maintenance planned around production instead of interrupting it
Business impact
Maintenance shifts from reactive to planned, improving schedule reliability.