Development test programs for 1-shot systems: 2-state reliability and binary development-test results

This paper considers efficient development testing for one-shot systems (e.g., missiles) that are destroyed in testing or first normal use, where there is the possibility of reliability growth of the basic system design as a result of redesigns following failed development tests. They consider a situation where the cost of redesign is negligible, each development test produces a binary (success-failure) outcome, and there is a fixed procurement budget covering both system development and purchase. For a two-state model of system reliability, dynamic programming is used to identify test-plans that are optimal, viz, maximize the mean number of effective systems (of the final design) that can be purchased with the remaining budget when development testing is stopped. Several reasonable and easily implemented suboptimal rules are also considered, and their performances are compared to that of the optimal rule for a variety of combinations of model parameters. Optimal tests plans are easily computable, even for problems where the initial budget is large, and for some combinations of model parameters offer important improvements over more naive test heuristics. The qualitative character of the present results is anticipated to extend to more complicated and realistic models for this problem.