The Challenge
Each mining site streamed data from hundreds of sensors — equipment health, throughput, environmental readings — into local systems that never spoke to each other. Leadership had no live operational picture, and problems were typically discovered only after they caused downtime. The operator needed a single command center that could turn high-volume sensor noise into timely decisions.
The Solution
We built a streaming telemetry pipeline on the Baalvion Operating System that ingests sensor data at scale, stores it as time series, and feeds a real-time command center. Operators see live equipment status across every site, with anomaly detection surfacing developing problems before they become outages.
- High-throughput ingestion sized for hundreds of sensors per site.
- Time-series storage optimised for fast range queries.
- Live dashboards giving one operational picture across all sites.
- Anomaly and threshold alerting that flags trends, not just breaches.
Architecture
Sensors publish to a high-throughput ingestion layer that normalises readings into a strict data contract. A stream processor computes rolling aggregates and runs anomaly detection, writing both raw and derived series to a time-series store. The command center reads from there for live dashboards, and the alerting service notifies operators the moment a trend crosses a learned bound. The approach mirrors observability for distributed systems.
Technology Stack
Stream ingestion, a time-series database, real-time dashboards, anomaly detection, and an alerting service — delivered through our automation and cloud solutions practices for the manufacturing sector.
Results
Telemetry now reaches operators in seconds rather than after the fact. Unplanned downtime dropped by roughly 28% as developing equipment issues were caught early, and every site is visible in a single operational view.
Lessons Learned
Defining the sensor data contract before building the dashboard prevented a swamp of inconsistent readings. Alerting on trends rather than raw thresholds cut alert fatigue dramatically. And operators only trusted the system once it could explain *why* it raised an alert — explainability mattered as much as accuracy.