Anomaly Detection

Find what's wrong before anyone reports it — including what no rule covers.

Most monitoring tools detect anomalies you already know to look for — by threshold, by rule. PULSAR goes further. By learning normal behavior from the Dynamic Twin, it detects three classes of anomalies: known failure patterns, rare statistical outliers, and behavioral drift that would only become visible in a late integration cycle — or in the field. The earlier a problem surfaces, the cheaper it is to fix.

Visual: anomaly learning over time (e.g. SDVDiag-style figure) or per-domain health traffic lights.

Key aspects

Detection scope diagram placeholder.
  • AI-driven model selection: picks the right anomaly model per signal type.
  • Improves over time: false positives shrink as more data is observed.
  • System-wide: across the full dependency graph — not per-ECU silos.
  • Three classes: known patterns · rare outliers · invisible drift.

Outcomes & KPIs

Detection scope Known + hidden drift

Beyond thresholds — behavioral signals tied to the twin.

Integration cycles Earlier discovery

Surface issues before SOP, not only after.

False positives ↓ over time

Continuous improvement with each run.

Target groups: Test & Validation · Heads of Quality · Warranty Managers

Use cases by industry

How teams apply this capability in their sector (more industries coming soon).