For easy administration, commonly used constant-stress ALT test plans consist of equally spaced test stresses, each with the same number of test specimens. Such standard plans are usually inefficient for estimating product reliability at the design stress. Earlier work by other researchers proposed statistically optimal plans for constant stress ALT which involved only two stress levels. Such plans lack robustness as the assumed stress-life relationship cannot be validated.

In recently years, there are some revived interests in extending the existing plan under other test constraints or allowing non-constant shape parameter. These plans advocated the use of a compromised plan for 3 stress levels without optimizing the middle stress and allocations of samples. As a result, the variance of the estimate at design stress is much higher for these compromised plans as compared to the 2-stress optimum plan. In this paper, we briefly describe a 3-stress CSALT plan that optimises both low and middle stress and their respective allocations. Our plan also limits the probability that the resulting probability plots of the failure data result in non-parallel lines under the usual assumption of constant shape parameter. In this way, one could infer conclusively that the shape parameters are indeed different when the data argue overwhelmingly that inconsistency exists.

In designing a test plan for 2-step SSALT, the high stress level is usually chosen to be the maximum stress condition under which the assumed stress-life relation holds. However, the selection of the low stress is not obvious. Choosing a low stress close to design stress level may result in too few observed failures at low stress level for meaningful statistical inference. Choosing a low stress close to high stress level may result in too much extrapolation of the stress-life relation. We have proposed a sequential approach to planning a multi-step step-stress ALT with Type-I censoring so as to achieve a pre-specified acceleration factor.

An optimal test plan of a step-stress ALT (SSALT) gives more precise estimates and yet uses the same number of test specimens and test time by specifying the optimal step-stress pattern. A mathematical programming approach to planning a constant stress ALT (CSALT) has been developed by assuming failure time following a Weibull distribution. We also developed a constrained non-linear program for solving for an optimal plan of 2-step SSALT with a target acceleration factor for exponential failure time.

We observe that while the proposed plan is almost as good as the optimum plan (Table 1) the Meeker-Hahn plan yields a standardized variance that is 15% higher than the optimum plan. The figures in brackets show the relative improvements that our plan has over both the Meeker-Hahn. It can be seen that the proposed plan works better under the higher acceleration factor.

(.) denotes percentage reduction in variance that the proposed plan made over other plans