h1

TY  - THES
AU  - Lu, Yao
TI  - Heterogeneous step-stress accelerated life testing
PB  - RWTH Aachen University
VL  - Dissertation
CY  - Aachen
M1  - RWTH-2025-10244
SP  - 1 Online-Ressource : Illustrationen
PY  - 2025
N1  - Veröffentlicht auf dem Publikationsserver der RWTH Aachen University 2026
N1  - Dissertation, RWTH Aachen University, 2025
AB  - Accelerated life testing (ALT) evaluates the lifetime performance of testing units within a limited period of time by imposing stress levels higher than those encountered under normal operating conditions. Step-stress ALT (SSALT) is a special type of ALT where stress is incrementally adjusted at predefined time points during the experiment. Statistical models for SSALT experiments, assuming a homogeneous population and various lifetime distributions, have been extensively discussed in the literature, also taking different censoring schemes into account. However, SSALT for a heterogeneous population, consisting of two or more groups, has received little attention so far, especially in cases where the group membership is unknown and the testing units form disjoint groups. Therefore, this thesis introduces a heterogeneous SSALT (hSSALT) model with two stress levels, and develops associated inferential procedures for both Type-I and Type-II censoring. Procedures are developed for continuously monitored experimental data, as well as, in case of Type-I censoring, for interval monitored data. Assuming the cumulative exposure model and exponentially distributed lifetimes, the hSSALT model accounts for unobserved heterogeneity at the second stress level by employing a two-component exponential mixture. The parameter estimates are obtained via the adapted Expectation-Maximization algorithm, with closed-form update rules derived for handling censored mixture data. Both asymptotic and bootstrap confidence intervals are constructed for interval estimation. In the presence of heterogeneity, the advantage of the hSSALT model over its homogeneous counterpart is demonstrated through theoretical analysis and simulation studies, particularly in yielding more accurate parameter estimates. A segmented log-link function is introduced to adjust lifetime extrapolation at normal operating condition, demonstrating that the homogeneous SSALT model tends to overestimate product lifetimes when heterogeneity is present. Furthermore, to determine the necessity of applying the hSSALT model, a likelihood ratio type test is developed for assessing homogeneity in mixture models under censoring, focusing on two-component exponential distributions. The test evaluates homogeneity against a two-subpopulation alternative and accommodates for both Type-I and Type-II censoring. The proposed test modifies the test statistic in a data-driven manner, offering a practial solution for homogeneity testing in censored mixture models. The associated critical values are obtained through simulation studies. In addition, power analyses under various scenarios demonstrate the effectiveness of the proposed test and confirm its practical applicability using real-world datasets from the literature.
LB  - PUB:(DE-HGF)11
DO  - DOI:10.18154/RWTH-2025-10244
UR  - https://publications.rwth-aachen.de/record/1022767
ER  -