Speaker
Description
Modern particle accelerators are highly complex instruments where maximizing beam availability is both an operational priority and economic necessity. However, intricate subsystem connections make rapid fault diagnosis challenging.
This work presents a distributed, multi-agent diagnostic framework combining continuous anomaly detection with deterministic root cause analysis.The system uses a scalable network of autonomous agents, each monitoring a specific subsystem (e.g., BPMs, magnets, vacuum profiles). Agents operate independently, using continuous unsupervised learning to detect deviations while adapting to long-term system changes. When an anomaly is detected, the agent's status feeds into a global logical model for reasoning.A key challenge is the asymmetry of fault propagation: a root cause necessarily produces certain downstream effects, but observing a symptom only possibly indicates a specific cause. Conventional approaches cannot capture this distinction, leading to overly conservative diagnoses or false positives. We employ modal logic with necessity and possibility operators to overcome this. This formalism deterministically encodes causality's directional nature, avoiding limitations of statistical methods like ambiguity and lack of explainability.
We demonstrate this at the Advanced Light Source (ALS), showing the process is deterministic, mathematically correct, and rapidly isolates root causes from cascading symptoms.
| In which format do you inted to submit your paper? | LaTeX |
|---|---|
| Preprint marking on your proceeding paper | I wish my paper to be marked as preprint. |