Fault　diagnosis　of　offshore　wind　turbines　(WTs)　is　vital　to　the　reliability　and　efficiency　of　wind　power　generation.　Generalized　demodulation　has　been　demonstrated　to　be　effective　in　revealing　the　fault-related　frequencies,　but　there　are　intricate　interfering　components　in　the　demodulation　spectra　due　to　the　iterative　demodulation　operation　and　the　complex　modulation　nature　of　vibration　signals　of　WT.　This　paper　develops　a　novel　generalized　Vold-Kalman　filtering　approach.　To　separate　the　signal　components,　Vold-Kalman　filter　is　adopted　to　the　amplitude　envelope　of　vibration　signal　by　exploiting　its　merit　in　non-component　decomposition,　in　which　a　frequency　set　is　constructed　by　the　modulation　nature　of　machinery　vibration.　The　phase　functions　of　the　frequencies　in　the　frequency　set　are　constructed,　by　which　the　variable　components　are　transformed　into　constant　ones.　The　proposed　method　is　tested　via　the　data　of　rolling　bearings　and　a　WT　planetary　gearbox.