DIAGNOSTIC ENTROPY - A QUANTITATIVE MEASURE OF THE EFFECTS OF SIGNAL INCOMPLETENESS ON SYSTEM DIAGNOSIS

Abstract

A set of measures quantifying the effects of signal incompleteness upon system diagnosis are defined and investigated. The most important is diagnostic entropy, a new quantitative measure of the effects of signal incompleteness upon system uncertainty. Diagnostic entropy is defined as the average uncertainty of a system when the system is indicated to be in an undesired state. This measure appears to be more useful for quantifying the difficulty of system diagnosis than conventional system entropy or conditional system entropy measures due to its relevance to the difficulty of human diagnosis of the system when it is in an undesired state. The magnitude of the diagnostic entropy is shown to be usually larger than that of the conditional entropy for typical highly reliable systems. This means that the uncertainty of the system is larger than usual when a reliable system is in an undesired state. We also suspect, but have not identified, the existence of a relationship between the diagnostic entropy of a system and the average physiological stress of human operators in diagnosing the system.