Groth, Katrina, Mosleh, Ali Data-driven modeling of dependencies among influencing factors in human-machine interactions (Inproceedings) Proceedings of the ANS International Topical Meeting on Probabilistic Safety Assessment and Analysis (PSA 2008), American Nuclear Society Knoxville, Tennessee, 2008.

BibTeX

@inproceedings{GrothPSA2008,
title = {Data-driven modeling of dependencies among influencing factors in human-machine interactions},
author = {Katrina Groth and Ali Mosleh},
year = {2008},
date = {2008-09-01},
booktitle = {Proceedings of the ANS International Topical Meeting on Probabilistic Safety Assessment and Analysis (PSA 2008)},
address = {Knoxville, Tennessee},
organization = {American Nuclear Society},
abstract = {Human performance is currently represented in most Human Reliability Analysis (HRA) models by a set of Performance Shaping Factors (PSFs). The majority of HRA methods have proposed relationships that produce human error probabilities based on the state of these PSFs. What current HRA methods are lacking is a model for human performance that includes dependencies between PSFs based on human performance data.

Using data from the Nuclear Regulatory Commission’s Human Events Repository and Analysis (HERA) database and applying iterative principal factor analysis and polychoric correlation, we have developed a methodology to obtain preliminary groupings of PSFs that lead to human errors in specific types of tasks.

The goal is to obtain a greater understanding of how PSFs are interrelated and to determine if certain groups of PSFs can be causally linked to certain types of failures. The final product will be a data-driven methodology to estimate the relationship between PSFs and human error probabilities in nuclear power plants. The methodology will consider systematic dependence between the PSFs and will use Bayesian techniques to integrate different types of data. Application of Bayesian Factor Analysis (BFA) will allow us to fill in gaps in the data and incorporate data types that cannot be used in principal factor analysis. This will enable us to make the best use of the limited amount of data we have available in HRA.

This paper will cover how we have used the data from HERA and will offer preliminary results based on the data currently available. It will also offer a systematic way to define the interrelationships between PSFs in different aspect of human-machine interaction.},
keywords = {Bayesian Networks, causal models, human error, Human Reliability Analysis (HRA), nuclear power, Performance Shaping Factors (PSFs)},
pubstate = {published},
tppubtype = {inproceedings}
}

Abstract

Human performance is currently represented in most Human Reliability Analysis (HRA) models by a set of Performance Shaping Factors (PSFs). The majority of HRA methods have proposed relationships that produce human error probabilities based on the state of these PSFs. What current HRA methods are lacking is a model for human performance that includes dependencies between PSFs based on human performance data.

Using data from the Nuclear Regulatory Commission’s Human Events Repository and Analysis (HERA) database and applying iterative principal factor analysis and polychoric correlation, we have developed a methodology to obtain preliminary groupings of PSFs that lead to human errors in specific types of tasks.

The goal is to obtain a greater understanding of how PSFs are interrelated and to determine if certain groups of PSFs can be causally linked to certain types of failures. The final product will be a data-driven methodology to estimate the relationship between PSFs and human error probabilities in nuclear power plants. The methodology will consider systematic dependence between the PSFs and will use Bayesian techniques to integrate different types of data. Application of Bayesian Factor Analysis (BFA) will allow us to fill in gaps in the data and incorporate data types that cannot be used in principal factor analysis. This will enable us to make the best use of the limited amount of data we have available in HRA.

This paper will cover how we have used the data from HERA and will offer preliminary results based on the data currently available. It will also offer a systematic way to define the interrelationships between PSFs in different aspect of human-machine interaction.