Localizing ruptures in block stochastic systems

In the present paper, rupture detection in a possibly complex stochastic system is enhanced by means of a procedure for localizing the actual rupture among various possible sources. A general model is considered which includes a variety oflinear and non-linear innovation based models. The proposed localizing multiple comparison procedure satisfies coherence and consonance and, for a wide class of models, strongly controls the familywise error i.e. the various type I error probabilities involved. The first order autoregressive optimal control model is considered as a popular example. Montecarlo simulations are performed to evaluate asymptotic approximations and diagnostic performances by means of mean time of delay and mean time between false alarms. Some conjectures are then given on the possible structure of these indicators.