In many Navy applications, the detection and classification of changes in systems is critical; examples are mechanical faults (e.g., helicopter gearboxes or in manufacturing processes) and underwater transients. Such systems often require nonstationary models, such as Hidden Markov Models (HMMs), which arise in contexts from modeling of speech to transients in machining. It is also of critical interest to monitor vibrations in mechanical systems which are subject to uncontrolled and unmeasured excitations. The main difficulty is that the measured signals reflect both the nonstationarities due to the surrounding excitation and the nonstationarities due to changes that one desires to detect. Our proposed research is aimed at developing new concepts for dynamic adaptive detection and fault classification systems in nonstationary noisy environments, which can quickly and accurately diagnose even small changes while being robust to changes in the unobserved excitation. We propose to investigate two approaches which seek to integrate the auditory signal representations ( Thrust area I) with dynamic statistical detection techniques: ( B.1) the first approach is based on the asymptotic local approach for change detection and classification, and will build on our work in adaptive algorithms [.Marcus 1990.]; ( B.2) the second involves extension of the stationary models for detection to nonstationary conditions using piecewise-stationary models.