In some environments where prognostics and health management would be beneficial, for example on board U.S. naval vessels, installation location and accessibility to power system must be considered. In this study, we investigate condition based maintenance and fault diagnosis for hydraulic actuators in power constrained environments. The experimental setup for collecting data is outlined, and a data set replicating multiple types of faults is collected. Several types of machine learning classifiers, including random forest and classification trees, are tested on the data set. Prediction accuracy as well as training and testing times are compared, which are used as a surrogate for power consumption in this study. We find that the random forest algorithm provides the lowest error rate of the tested classifiers but has some of the highest training and testing times. Classification trees, on the other hand, provide a better tradeoff between accuracy and computation time.