Text mining for procedure-level primitives in human reliability analysis

Abstract

The classification of nuclear power plant procedures at the sub-task level can be accomplished via text mining. This method can inform dynamic human reliability calculations without manual coding. Several approaches to text classification are considered with results provided. When a discrete discriminant analysis is applied to the text, this results in clear identification procedure primitive greater than 88% of the time. Other analysis methods considered are Euclidian difference, principal component analysis, and single value decomposition. The text mining approach automatically decomposes procedure steps as Procedure Level Primitives, which are mapped to task level primitives in the Goals, Operation, Methods, and Section Rules (GOMS) human reliability analysis (HRA) method. The GOMS-HRA method is used as the basis for estimating operator timing and error probability. This approach also provides a tool that may be incorporated in dynamic HRA methods such as the Human Unimodel for Nuclear Technology to Enhance Reliability (HUNTER) framework.