Published January 1, 2026 | Version v1

Explainable AI For Detecting Safety Barrier Degradation From Near Miss Reports

Authors/Creators

Description

- Industrial maintenance generates near miss narratives that may reveal weaknesses in energy isolation, equipment integrity, permit coordination and other safety barriers. However, interpreting these reports requires contextual judgement, and an automated summary can conceal missing evidence or produce unjustified conclusions. This conceptual paper proposes an evidence-linked artificial intelligence framework for identifying reported barrier degradation and supporting accountable human review. The framework combines a controlled barrier taxonomy, source-linked extraction, retrieval from approved site documents, uncertainty handling and verification of corrective actions. Its central research proposition is that the value of AI should be evaluated through the quality of barrier decisions and the completeness of supporting evidence, rather than classification accuracy alone. A staged mixed methods protocol is specified, including expert annotation, temporal and site-separated testing, comparison with simpler methods, a randomised reviewer study and a prospective shadow deployment. Primary outcomes concern detection of expert-adjudicated barrier concerns, unsupported assertions and reviewer decision quality. Incident reduction is reserved for later investigation and is not inferred from report processing performance. No field dataset, implemented prototype or empirical results are presented. The contribution is a testable research design that connects explainable AI with practical safety assurance while retaining operational authority with competent personnel

Files

ICSEMT.V14_issue5_107.pdf

Files (900.2 kB)

Name Size
md5:a3d70bfd0dbd41b242ce66e466b4d241
900.2 kB Preview Download

Additional details