MeterSense is an end-to-end SQL Server–based analytics platform designed for monitoring, diagnosing, and optimizing the performance of smart LPG meters deployed in the field.
It provides insights on:
- Firmware update performance
- Auto-configuration time
- Network reliability
- Failure root causes
- Hybrid link-switching (Cellular/Satellite)
- ML-based failure prediction
- Track upload, flashing, and auto-configuration time
- Compute total update duration
- View success/failure distribution
- Identify slow firmware versions
- Compare performance across releases
- Capture RSSI (signal strength), network type, errors
- Detect weak-signal periods linked to failures
- Understand failure patterns around update windows
- Support for dual connectivity (CELLULAR + SATELLITE)
- Store ML models, versions, metrics, training windows
- Predict firmware-update failure probability
- Recommend link type based on risk
- ML-based decision thresholds on a per-meter basis
- Evaluate model vs real outcomes
Switch between CELLULAR ↔ SATELLITE when:
- Signal RSSI drops below threshold
- Repeated failures occur
- ML predicts high failure risk
- Manual override is applied
All switching events are logged with:
- Reason
- Previous failures
- Previous RSSI
- From/To link type
| Table | Purpose |
|---|---|
Customers |
Customer/site registry |
Meters |
Physical LPG meters |
FirmwareVersions |
Firmware releases |
FirmwareUpdates |
Full update lifecycle logs |
ConnectivityLogs |
Network quality + errors |
ConfigurationEvents |
Auto-config or overrides |
UsageReadings |
Gas, battery, temperature |
MLModels |
Stored ML model metadata |
LinkFailurePredictions |
Failure risk per update |
MeterConnectivityConfig |
Thresholds + ML settings |
LinkTypes |
CELLULAR / SATELLITE |
LinkSwitchEvents |
Connectivity fallback events |
| View Name | Purpose |
|---|---|
vw_FirmwareVersionKPI |
Firmware KPIs (avg time, failure rate) |
vw_ProblemMeters |
Meters with high failures or slow updates |
vw_FailureRootCause |
Network context around update failures |
vw_MLPredictionPerformance |
Prediction accuracy vs actual |
- Slow-performing meters
- Failing firmware versions
- Sites with persistent weak network
- Timeouts caused by poor RSSI
- When satellite fallback is necessary
- Probability a firmware update will fail
- Whether Cellular or Satellite should be used
- Which meters are likely to cause operational delays
- MQTT pipeline for real-time ingestion
- Device-heartbeat monitoring
- Over-the-air config commands
- Add RandomForest/GBM/Neural models
- Rolling model retraining automation
- Per-site adaptive RSSI thresholds
- Multi-link support (Wi-Fi / LoRaWAN / Satellite)
- Predictive switching based on moving RSSI trends
- Cost-optimized routing (cellular vs satellite billing)
- Real-time streaming dashboards
- Predictive analytics panel
- Site-level aggregation (failures by county/district)
Brian Rono
Smart Meter Systems Engineer & Machine Learning Researcher
If this project is helpful, consider giving the repo a ⭐ on GitHub.