An information theory approach to a fault location problem

Abstract

The fault location model under investigation consists of an n-component series system known to have exactly one failed component. Component positions in the system are taken as fixed. A component is either working or failed. Components work or fail independently of each other, with their a prior reliabilities taken as given but not necessarily equal. Group testing to locate the failed component is sequential, binary and dichotomous in nature with certain results. The only costs are the number of tests made. The three solution procedures investigated are (1) a dynamic programming formulation, (2) a sequential halving procedure, and (3) a procedure based on information theory. The criteria for optimality are minimization of the expected number of tests required and minimization of the maximum number of tests required.