The　vibration　signals　generated　by　the　gear　partial　failure　show　non-stationary　and　non-periodic.　The　sideband　spectrum　near　the　gear　mesh　frequency,　the　second　and　the　third　harmonics　of　the　corresponding　frequency　spectrum　all　grow　significantly.　Because　principles　of　prediction　and　update　are　closely　related　to　the　fault　information,　the　vibration　signals　are　first　analyzed　and　preprocessed　by　lift　wavelet,　and　the　wavelet　basis　function　selection　is　replaced　by　designing　the　predictor　and　updater　during　the　signal　decomposition　process,　which　can　significantly　raise　the　feature　extraction　efficiency.　The　high　frequency　signal　is　demodulated　by　Hilbert　transformation,　the　conventional　vibration　components　are　removed　but　only　fault　information　retained　in　its　envelope　spectrum,　and　the　faults　are　located　after　the　fault　feature　frequency　can　be　identified　effectively.　An　illustration　verifies　that　the　Hilbert　modulation　technology　based　on　lifting　wavelet　transform　is　fully　competent　for　gear　fault　diagnosis.