Estimation of component mean lifetimes of a masked system using unclassified system life data (Q2571932)

Consider a sample of \(n\) masked systems with \(J\) components, which have exponential distribution with mean life \(\mu_i\), \(i=1,\dots,J\). Let \(T\) be a system life-length, which occurs when one of \(J\) components fails, and \(f(t| \mu_i)\) be the conditional density of \(T\) given that the \(i\)th component fails. A stochastic approximation procedure is proposed to compute the conditional expectation of \(E(\underline{\mu}| \underline{t}_{n+1})\) as the estimate of component mean lifetime \(\underline{\mu}=(\mu_1,\dots,\mu_J)\) recursively with \(E(\underline{\mu}| \underline{t}_n)\) and new observation \(t_{n+1}\), where \(\underline{t}_n=(t_1,\dots,t_n)\). The results of a simulation study show that the proposed procedure produces a better estimate than the \(K\)-mean algorithm.