Rigorous verification across every system layer — authentication, real-time data, ML detection, and alert accuracy
The application opens on a secure login screen. Credentials are validated against a hashed JSON credential store. Failed attempts are rejected cleanly with no crash or information leak. Three distinct roles unlock different pages and capabilities.
The Digital Twin page renders an SVG ISS diagram with color-coded health indicators at all 7 module locations. Testing confirmed all indicators update in real time as fault states are injected and cleared via the data generator.
Historical sensor data is plotted as interactive Chart.js line charts. Each parameter shows its nominal range as a reference, and anomaly-flagged readings are visually distinguished from nominal data. Testing confirmed data integrity across full historical queues.
The Quick Look view surfaces all 20 sensor readings across all 7 modules simultaneously. Operators can scan the entire system state at a glance without navigating between locations. Parameter status indicators update in sync with the 1-second WebSocket tick.
When the Isolation Forest + Random Forest pipeline classifies incoming sensor data as an active fault, the system fires a critical alert. Testing validated all 8 fault types trigger correctly with zero false positives during nominal operation.
Zero false alerts raised during 1,000 consecutive nominal samples.
A.U.R.A. distinguishes Critical alerts (active fault confirmed by ML pipeline) from Warning alerts (parameter drifting toward out-of-range). Both tiers were validated independently for threshold accuracy and display correctness.
Active fault. RF confidence above gate. Fires after N consecutive anomalous ticks.
Parameter approaching limit. Pre-fault indicator. Fires before Critical triggers.
Alert thresholds, cooldown windows, and ML gate values are configurable at runtime. Changes persist to disk and take effect on the next server tick without restart.