Satellite Reliability: Statistical Data Analysis and Modeling

Abstract

The technical literature has long recognized the importance of satellite reliability, but a statistical analysis of expansive on-orbit failure data is still lacking. As a result, inconsistencies persist in the literature due to the absence of an empirical basis for settling the issues, for example, regarding the existence or absence of satellite infant mortality. Weibull distributions with a shape parameter larger than one are commonly used to model satellite reliability. This choice for the shape parameter fails to capture infantmortality and is shown here to be flawed. The present work fills a gap in the technical literature by 1) conducting a thorough nonparametric statistical analysis of recent on-orbit failure data, 2) fitting a parametricmodel to the actual/observed spacecraft reliability, and 3) quantifying the relative contribution of each subsystem to satellite failure and identifying the subsystems that drive spacecraft unreliability. The sample analyzed in this study consists of 1584 Earth-orbiting satellites successfully launched between January 1990 andOctober 2008. The results presented here should prove useful for the space industry, for example, in redesigning satellite (and subsystem) test and screening programs and in providing an empirical basis for subsystem redundancy allocation and reliability growth plans.