We present the first system, called iCM-Hydraulic, for intelligent condition monitoring (CM) of hydraulic machines that combines statistical, probabilistic and semantic data analysis for fault detection and diagnosis with semantic explanations. The modelling of the domain ontology in OWL2 and the probabilistic domain belief network is based on CM standards and domain expert interviews. Fast fault detection and diagnosis online is performed by the system over a multi-variate sensor data feature stream with statistical fault state classification, semantic symptom detection and diagnosis query answering with C-SPARQL, semantic and probabilistic reasoning in the continously updated belief network. Condition diagnosis queries are also answered offline over the central SwiftOWLIM store with history data. The system prototype was developed for our customer HYDAC Filter Systems GmbH and successfully tested for typical hydraulic aggregates and sensors.