Method Description
The analysis module applies statistical deviation models against rolling baseline windows derived from historical sensor data. Classification trees are executed at the edge layer for low-latency scoring; complex pattern-matching workloads are offloaded to the central processing core. Baseline windows are updated on a configurable schedule without requiring system restarts or manual recalibration.
Data Types
| Type | Source | Preprocessing |
| Time-series telemetry | Sensor arrays · SCADA outputs | Normalisation · Outlier removal · Gap-fill interpolation |
| Categorical event logs | Access control · Manual entries · System events | Label encoding · Frequency aggregation |
| Optical / image metadata | Camera feeds via edge classifier | Feature vector extraction (edge) · Dimensionality reduction |
| Environmental readings | Temperature · Vibration · Pressure sensors | Moving average · Threshold band calculation |
Output Formats
| Output Type | Format | Delivery |
| Anomaly score | Float 0.0–1.0 per asset per interval | REST API · MQTT topic |
| Trend report | Structured JSON · PDF | Scheduled push · On-demand query |
| Baseline snapshot | Compressed JSON archive | SFTP · Local storage |
| Predictive maintenance flag | Boolean + confidence + rationale string | Webhook · Operator dashboard |
Configuration
| Parameter | Options |
| Baseline window | 7 days · 30 days · 90 days · Custom |
| Model update frequency | Daily · Weekly · Manual trigger |
| Confidence threshold | Configurable per asset class (default 0.75) |
| Processing allocation | Edge-only · Core-only · Hybrid (recommended) |