Survival analysis and maintenance policies for a series system, with highly censored data

This paper considers the problem of estimating the survival function from a large set of sampling data subject to high levels of random censoring on the right. The system under study consists of a series arrangement of four functional subsystems. Each of the functional subsystems consists of a collection of independent components in series. The system does not have redundant components. This study aims to simulate a series arrangement of four unique components and compare the performance of the Kaplan Meier Estimator (KME), the piecewise exponential estimator (PEXE) and the maximum likelihood estimator (MLE) in estimating the survivor functions for the system as well as individual components under high levels of random censorship. Monte Carlo analysis is used to compare total time on test plots and optimal age replacement times determined using the KME and PEXE methods. This study extends the work of Klefsjo and Westberg (1994) by considering the estimation of survivor functions and optimal age replacement periods under higher levels of random censorship (up to 90%). The effect of such high censoring is that both the survivor curve and the optimal replacement time are generally, and sometimes severely, underestimated at the component level but not necessarily at the system level. Further studies will examine the trade-offs in using system level vs. component level data to make maintenance decisions for highly censored samples.