Bayesian Aspects of Trouble Shooting Behavior

Abstract

Thirty-nine Navy technician trainees filled out a symptom-malfunction matrix on a blocking oscillator circuit. The technicians then attempted to solve six troubleshooting problems in the same oscillator circuit. The particular sequence of checks used by each man on each problem was combined with his symptom-malfunction matrix, via a Bayesian algorithm, to yield computer estimates of failure likelihoods for each component. The computer program predicted actual parts-replacement behaviors in about half of the cases. Those technicians who start out with valid symptom-malfunction matrices are more likely to resemble the Bayesian processor.