Comparative statistical techniques in maintenance management

A common problem in industry is to compare the ‘life to failure’ of two batches of equipment. These batches often represent equipment maintained under two regimes, built by two manufacturers, or in two states of modification. In its general form, this kind of problem is well known in the statistical literature. It is formally referred to as the ‘two-sample problem’, and is mainly concerned with the procedure that has to be followed in order to detect any statistically significant differences between two populations, on the basis of samples drawn at random from them. This paper summarizes part of the work carried out to date on seeking alternative and preferably analytical solutions to the two-sample problem, in the context of reliability or maintenance applications. For this purpose, nonparametric two-sample tests for censored data are introduced, reviewed, and successfully applied to engineering problems. The tests employed include the generalized Gehan-Wilcoxon, the logrank, and the modified Kolmogorov-Smirnov.