Back to AI Research

AI Research

Agentic Root Cause Analysis through Evidence-Ground... | AI Research

Key Takeaways

  • Agentic Root Cause Analysis through Evidence-Grounded Reasoning Industrial systems are complex, and when they behave abnormally, identifying the specific phy...
  • Diagnosing the root cause of anomalies is essential for safe industrial operation.
  • Despite extensive sensor instrumentation, formulating hypotheses and gathering evidence remains a manual process, creating a major operational bottleneck.
  • To address this gap, we introduce AgentRCA, a zero-shot agentic framework for evidence-grounded root cause analysis.
  • Rather than learning fault-specific mappings, AgentRCA performs inference-time reasoning by combining a data-driven digital twin (modeling normal system dynamics) with a tool-augmented large language model.
Paper AbstractExpand

Diagnosing the root cause of anomalies is essential for safe industrial operation. Despite extensive sensor instrumentation, formulating hypotheses and gathering evidence remains a manual process, creating a major operational bottleneck. While existing data-driven approaches aim to automate this, two critical limitations restrict their deployment: their operate as black boxes unable to justify their diagnosis, and they require scarce labeled examples of faulty operation. To address this gap, we introduce AgentRCA, a zero-shot agentic framework for evidence-grounded root cause analysis. Rather than learning fault-specific mappings, AgentRCA performs inference-time reasoning by combining a data-driven digital twin (modeling normal system dynamics) with a tool-augmented large language model. The agent iteratively gathers statistical evidence, evaluates competing hypotheses, and identifies the physical fault that best explains the observed behavior. Evaluated on a real-world multiphase-flow facility and a large-scale chemical plant, AgentRCA achieves diagnostic performance competitive with fully supervised baselines without relying on fault-specific training. Crucially, it produces transparent reasoning traces that explicitly link observed symptoms to their underlying physical causes. These results establish autonomous hypothesis-driven reasoning as a practical foundation for scalable industrial root cause analysis.

Agentic Root Cause Analysis through Evidence-Grounded Reasoning
Industrial systems are complex, and when they behave abnormally, identifying the specific physical fault is a difficult, manual task. Current automated solutions often struggle because they either act as "black boxes" that cannot explain their reasoning or require large amounts of historical data from past failures—data that is often rare or impossible to collect. This paper introduces AgentRCA, an autonomous framework that uses an "agentic" approach to diagnose industrial faults. Instead of relying on pre-labeled examples of failures, it uses a combination of a digital twin and a large language model to reason through evidence in real-time, mirroring the diagnostic process of a human engineer.

How AgentRCA Works

The framework functions by pairing a data-driven digital twin with a tool-augmented large language model. The digital twin is trained exclusively on data from normal system operations, allowing it to understand how the system should behave under various conditions. When an anomaly occurs, the agent acts as an investigator: it gathers statistical evidence—such as shifts in sensor correlations, mean values, or reconstruction errors—and compares these findings against natural-language descriptions of potential faults. By iteratively testing hypotheses and ruling out inconsistencies, the agent arrives at a diagnosis that is backed by clear, transparent reasoning.

Performance and Reliability

The researchers tested AgentRCA on two challenging industrial benchmarks: a multiphase-flow facility and a large-scale chemical plant. Even without being trained on any specific fault data, the system achieved diagnostic accuracy competitive with traditional, fully supervised models that require extensive historical failure records. A key finding is that the agent’s performance improves significantly when it is forced to structure its reasoning, such as by maintaining a ranked table of candidate faults. This structured approach helps the agent organize evidence and avoid jumping to premature conclusions when faced with complex, overlapping symptoms.

The Importance of Context

A critical component of the framework is its ability to interpret data relative to the system's current operating state. The study found that matching test data to the correct "normal" operating regime is essential; without this context, diagnostic accuracy drops significantly. By grounding its analysis in these specific baselines and providing a step-by-step trace of its logic, AgentRCA offers a more reliable and interpretable alternative to purely statistical methods. This allows engineers to see exactly which sensor deviations led the agent to its conclusion, making it a practical tool for real-world industrial maintenance.

Key Considerations

While the framework is robust across different language models, the researchers noted that the complexity of the reasoning task matters. Smaller models may struggle when asked to maintain and revise multiple competing hypotheses simultaneously. Additionally, the system’s effectiveness relies on the quality of the diagnostic tools provided to the agent and the clarity of the task-specific guidance in the system prompts. These results suggest that by combining traditional causal reasoning with modern AI, it is possible to build scalable, autonomous diagnostic systems that do not depend on the scarcity of historical failure data.

Comments (0)

No comments yet

Be the first to share your thoughts!