Integrating Data and Causal Reasoning to Automate Root Cause Analysis

This paper presents a structured, data driven framework for automating Root Cause Analysis (RCA) in complex nuclear power plant environments.
Traditional RCA depends on manual interpretation of heterogeneous data sources (i.e., telemetry, maintenance records, and system documentation) making the process time-consuming, expertise dependent, and vulnerable to
inconsistencies. These challenges are amplified by system interdependencies, incomplete or ambiguous data, and unclear temporal relationships between causes and effects. The proposed framework integrates plant telemetry, historical records, and system architecture into a unified ``hypothesize-test-compare-decide'' workflow. It combines three complementary reasoning dimensions: structural plausibility, temporal coherence, and historical evidence. A knowledge graph derived from model based systems engineering constrains hypotheses to feasible components and failure modes, while a vector based retrieval system links candidate explanations to relevant documentation and supporting information.
The method generates, evaluates, and ranks multiple causal hypotheses, explicitly representing
uncertainty and alternatives. Evidence is categorized and used to refine rankings,
ensuring traceability and engineering interpretability throughout the RCA process.
A representative case study demonstrates how jointly analyzing timing, system structure, and
supporting evidence improves diagnostic rigor, consistency, and transparency.
The results illustrate how integrated causal reasoning can enhance RCA efficiency while
preserving expert oversight.