Combining infinitesimal perturbation analysis (IPA) with stochastic approxi
mation gives identification algorithms to estimate the optimal threshold va
lue for failure-prone manufacturing systems consisting of one machine produ
cing one part type. Two adaptive control schemes are proposed. The adaptive
control schemes do not require the knowledge of the distribution functions
of the up and down times. Under some appropriate conditions, the strong co
nsistency, as well as the convergence rates, of the identification algorith
ms and the cost function is established for the adaptive control schemes. I
n particular, it is shown that central limit theorems hold for the identifi
cation algorithms.