System Configuration & Key Components
- Core Technologies
- PyTorch-based segmentation model for region extraction and level calculation
- TensorFlow-based image classification model for emulsion detection
- Hardware
- Hanwha Vision QNB-6000 IP Camera, Infotrol–APC Server
- Interval
- Video frames captured every 5 seconds and stored in InfluxDB
- Output
- Predicted fluid level, emulsion status, anomaly detection, and connection status → recorded as tags
- Operating Conditions
- Real-time prediction, storage monitoring, and DCS-based alert system for anomaly detection
Benefits & Differentiation
- Stable level detection without operator intervention → eliminates repetitive manual checks
- Sensorless level prediction via visual input → applicable in environments where sensor installation is not feasible
- Robust to visual disturbances such as glare, vibration, or changes in lighting → built-in exception handling logic
- High reliability compared to manual reading → validated through comparison and InfluxDB-based monitoring
- Continuous data accumulation supports trend tracking and predictive maintenance
- Real-time DCS integration enables early warning for unexpected conditions or process abnormalities
