Onnes: A Physics-Grounded Multi-Agent LLM Simulator for Cryogenic Fault Diagnosis in Quantum Computing Infrastructure

2026-07-07 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Data Science & Analytics · Depth: Expert, quick

Summary

Onnes, a physics-grounded multi-agent LLM simulator, addresses the challenge of fault diagnosis in dilution refrigerators, critical infrastructure for superconducting quantum computers. The simulator integrates a real dilution-cooling floor, a noise-and-correlation fingerprint learned from BlueFors logs, and six physics-grounded fault classes. In a 1000-turn evaluation, a zero-shot LLM agent panel initially matched a supervised ML classifier on fault detection but trailed on classification. However, curated contrastive few-shot demonstrations and self-consistency voting significantly boosted classification accuracy from 0.685 to 0.990, matching the supervised classifier's 0.985 with only six labeled demonstrations. A sim-to-real check demonstrated a 6.4% real-hardware false-alarm rate and 100% recall on injected physics faults.

Key takeaway

For AI Scientists developing fault diagnosis systems for complex infrastructure like quantum computers, you should consider integrating physics-grounded digital twins with multi-agent LLM panels. This approach, demonstrated by Onnes, achieves high classification accuracy (0.990) with minimal labeled data (six demonstrations), outperforming supervised ML. Explore few-shot contrastive demonstrations and self-consistency voting to significantly enhance your LLM diagnostic capabilities and reduce false alarms.

Key insights

Onnes leverages physics-grounded digital twins and few-shot multi-agent LLMs for highly accurate cryogenic fault diagnosis.

Principles

• Digital twins enhance LLM-based diagnostic accuracy.
• Few-shot demonstrations significantly boost LLM classification.
• Self-consistency voting improves diagnostic reliability.

Method

Onnes couples a real dilution-cooling floor, a learned noise fingerprint, and physics-grounded fault classes to drive a live multi-agent LLM operations layer for fault diagnosis.

In practice

• Integrate digital twins for complex system simulation.
• Use few-shot contrastive demonstrations for LLM accuracy.
• Apply self-consistency voting to improve LLM classification.

Topics

• Multi-Agent LLMs
• Cryogenic Fault Diagnosis
• Quantum Computing Infrastructure
• Digital Twin Simulation
• Few-Shot Learning

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Evaluating Multi-Agent LLM Architectures for Rare Disease Diagnosis — Multi-agent LLM architectures do not guarantee improved diagnostic accuracy; topology selection is critical.
• Think like a Scientist: Physics-guided LLM Agent for Equation Discovery — KeplerAgent improves equation discovery by integrating scientific reasoning and physics-guided priors into LLM agents.
• Exploring LLM-based Verilog Code Generation with Data-Efficient Fine-Tuning and Testbench Automation — Multi-agent LLMs can automate testbench generation, creating high-quality data for efficient Verilog code generation.
• ASyMOB: Algebraic Symbolic Mathematical Operations Benchmark — LLMs struggle with symbolic math perturbations, indicating pattern memorization, though frontier models show surprising robustness.
• MARCH: Multi-Agent Reinforced Self-Check for LLM Hallucination — MARCH uses multi-agent reinforcement learning and information asymmetry to reduce LLM hallucination in RAG systems.
• Multi-Modal Agents for Power Distribution Defect Detection: An Evaluation of Foundation Models — Multi-Modal Agents require integrated perception, reasoning, and tool usage for autonomous industrial defect detection.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.