On the identifiability of start-up demonstration mixture models (Q2409392)

The authors consider start-up demonstration testing -- a technique for assessing the reliability of a manufactured unit through its performance on successive start-ups. This performance is taken into account and several different types of decision criteria (most of them inspired by the theory of runs and scans) for accepting or rejecting the unit are introduced. The authors study binary start-up demonstration tests, assuming that they have at hand two different types of units with potentially different probabilities of successful start-up. In this case, the waiting time distributions are expressed as two-component mixture models and their identifiability is discussed. An estimation method based on the Expectation-Maximization algorithm (EM algorithm) for the model parameters is described and some numerical examples are presented.