A study on dependent-failure detection for 2-unit parallel systems by Kolmogorov-Smirnov test
This thesis proposes a statistical test method for 2-unit parallel systems, which discriminates whether the units(e.g. components, devices, etc.) failed independently or dependently. The method which makes a system redundant by combining several units is often used in order to construct a highly-reliable system. However, if the units break down dependently, the reliability of the system would decrease. We present the statistical test scheme based on the failure occurrence times of the units. It is assumed that the life time distribution of the survival unit changes if one unit fails. There-fore our scheme can ﬁnd dependent failure occurrences by identifying the change of the distribution. The performance of the proposed method is shown by simulation studies.