Quantifying Human Performance for Reliability Analysis of Systems

Abstract

A general mathematical model of the probability of errorless human performance was derived and equated to human reliability for time-continuous tasks. The application of this model and the implications of the time-to-first-human-error (TTFHE) concept were tested with data collected using a laboratory vigilance task. The error data were ordered, and through classical inference theory the underlying density functions were isolated and tested for goodness of fit. Weibull, gamma, and log-normal distributions emerged as relevant; normal and exponential distributions were rejected. The relevant distribution parameter values were applied to the general mathematical model, and predictions were made of human performance reliability for the task. It was concluded that this is a feasible and meaningful way to quantify human performance for time-continuous tasks for use in reliability analyses of systems.