Modeling reliability growth late in development

Reliability is often overlooked during early developmental testing when the focus is on improving system performance. As a result, rapid reliability growth programs are frequently initiated late in development, just before the system enters production. The widely-used AMSAA reliability growth model is difficult to apply on such programs because it is a cumulative function of test time and the number of failures observed; therefore, it usually projects that an excessive amount of testing is required to overcome a period of low reliability, even at high growth rates. This paper formulates an alternative model for projecting reliability growth late in development that does not suffer from this problem. The model also provides a framework for reliability growth on system acquisition programs that utilize short development cycles, such as spiral development. An example application of the model on the AWACS RSIP reliability growth program is presented and discussed.