Skip to content

Scenario 6: Long-Term Fuel Integrity Monitoring

Reactor: Generic SMR. Pattern: Pattern 9 + KG + adaptive. Primary concepts: KG grounding, simulator-coupling.

Description

Long-term monitoring of fuel integrity across an operating cycle (18-24 months). Unlike transient-focused scenarios, this tests AI advisory over extended timescales where gradual trends develop slowly and context spans weeks to months.

Agent Architecture

  • Fuel performance analyst: Monitors burnup, cladding integrity, fission gas release. Grounded in KG encoding fuel design limits and vendor guidelines
  • Projection agent: Coupled to FRAPCON. Periodically computes forward projections of fuel conditions. LLM specifies boundary conditions; physics from simulation code
  • Chemistry analyst: Tracks primary coolant chemistry (iodine, caesium ratios) for early fuel degradation indicators
  • Operating history agent: Maintains episodic memory of fuel-relevant events across the full cycle

KG Grounding

The knowledge graph encodes fuel design limits, Tech Spec surveillance requirements, parameter-to-performance relationships, and historical fuel failure precursor patterns. KG guardrails prevent reassuring assessments when parameters have exceeded design limits.

Adaptive Architecture

Reduced mode during normal operations (single analyst + periodic FRAPCON projections), escalating to full deployment when indicators approach limits or anomalous chemistry detected.

Demonstrated Principles

Tests simulator-coupling for Level 3 SA over long timescales. Tests KG grounding against verified fuel design knowledge. Tests long-term episodic memory management across months without unbounded context growth.