Models for Human Error in Structural Reliability

Abstract

Probabilistic models are proposed for the occurrence, detection, and consequences of human errors in structures. The models are developed in an overall framework of decision theory applied to the problem of allocating control efforts that would narrow the gap between the estimated and actual rates of structural failure. Two models for error occurrence are presented: (1) A binomial distribution for errors in discrete tasks; and (2) a Poisson process for errors occurring in a continuous production interval. Checking is modeled as a sequence of Bernoulli trials. The restrictive independence assumptions implied in the classical use of the preceding distributions are replaced by the weaker and more appropriate assumption of exchangeability, by accounting for the uncertainty regarding distribution parameters. A bivariate probability model is also developed which correlates the performance of a structure to its error content. The aforementioned probabilistic models provide an outline of areas where prior information is needed and a specific format for the collection of additional data. An application to decision‐making in the area of nondestructive testing is also presented.