Experimental-design techniques in reliability-growth assessment

Several recent statistical methods, including a Bayesian technique, have been proposed to detect the presence of significant effects in unreplicated factorials. It is recognized that these techniques were developed for s-normally distributed responses; and this may or may not be the case for times between failures. In fact, for homogeneous Poisson processes (HPPs), these times are exponentially distributed. Still, response data transformations can be applied to these times so that, at least approximately, these procedures can be used. It was therefore considered important to determine how well these different techniques performed in terms of power. The results of an extensive Monte Carlo simulation are presented in which the power of techniques is analyzed. The actual details of a fractional factorial design applied in the context of reliability growth are described. Finally, power comparison results are presented