Subsea resident AUV systems have very strong robustness requirements regarding software and hardware, in order to endure the hazards and difficulties posed by the marine environment. In such condition, Fault Diagnosis and Fault Prognosis are essential to ensure reliable operation of the vehicle and enable correct response to failures. The first results of the development of a data-driven online adaptive fault detection system for underwater thrusters will be presented. The goal of this work is to study the occurrence of soft-faults due to vehicle model changing in long term exposition of the vehicle to the environment, and perform component degradation estimation to aid autonomous condition evaluation when the vehicle is submerged. The fault detection method is based on novelty detection applied to the rotational speed and current signals of thrusters. Several regression methods were evaluated for model learning, novelty detection and outlier filtering.