Many Autonomous Underwater Vehicles (AUV) have high rates of false-alarms because their health management relies on user-generated rules. We suggest that the high false-alarm rate could be substantially lowered if fault-detection were based on actual actuator performance instead of heuristics. We collected performance data on a critical AUV actuator, a mass-shifter, in order to develop an unsupervised fault detector. We found that a small number of features were sufficient to detect known and novel faults with a high probability of detection and a low false alarm rate. We also found that n-point false-alarm reduction schemes performed poorly due to correlation during startup.