Optimal design and equivalency of accelerated life testing plans.

摘要

Accelerated Life Testing (ALT) is an efficient approach to obtain failure observations by subjecting the test units to stresses severer than design stresses and utilize the test data to predict reliability at normal operating conditions.;ALT plans under multiple stresses needs to be designed to resemble the normal operating conditions and obtain useful failure observations for accurate reliability prediction. However, to date there is little research into the theory of planning ALT for reliability prediction with multiple stresses. Multiple stresses can result in a large number of stress-level combinations which presents a challenge for implementation. We propose an approach for the design of ALT plans with multiple stresses using Latin hypercube design (LHD) and demonstrate the proposed method with examples based on actual tests. The obtained optimal test plans are compared with those based on full factorial design. The comparison shows that ALT based on LHD not only increases the accuracy of reliability prediction significantly but also reduces the test duration dramatically.;ALT under Type-I and Type-II censoring has been extensively investigated. We generalize the one stage censoring to multi-stage progressive censoring, where the surviving test units are removed at intermediate stages other than the final termination of the test. This procedure further minimizes the test time and cost. We also combine the progressive censoring scheme with competing risk when test units experience different failure modes to investigate general, practical and optimal ALT plans.;ALT is usually conducted under constant-stresses which need a long time at low stress levels to yield sufficient failure data. Many stress loadings, such as step-stresses obtain failure times faster than constant-stresses but the accuracy of reliability predictions based on such loadings has not yet been investigated. We develop test plans under different stress applications such that the reliability prediction achieves equivalent statistical precision to that of the constant-stress. The research shows indeed there are such equivalent plans that reduce the test time, minimize the cost and result in the same accuracy of reliability predictions.