Verification of human error traps in NPP procedures utilizing syntactic and semantic information extraction

Procedures must be provided in an easy-to-follow manner to ensure human performance quality especially in complex process industries. Principles and practices to enhance human performance of plant personnel are described in a procedure writing manual. However, a coherent review of a large volume of procedures concerning various human performance issues is a challenging task. This paper introduces a novel application of natural language processing for the verification of human error traps in procedures in two stages. First, all the significant syntactic and semantic information are extracted from the input procedure documents, and then the stepwise drilled-down verification results for each human error trap are provided. Case study results of the proposed approach to selected procedures obtained from a U.S. commercial nuclear power plant are introduced showing the various summary analysis data produced by the software tool implemented. Although the number of analyzed procedures was limited, it is shown that the proposed approach could make the process of verifying human-factored procedures systematic and coherent. Increasingly advanced management of procedures could be implemented by integrating more intelligent features into the proposed approach.