Aimed　at　the　difficulty　to　extract　the　fault　information　of　a　low　speed　shaft　in　a　large　gearbox,　wavelet　analysis　was　used　to　realize　the　local　analysis　of　signals　in　a　time　and　frequency　domain　simultaneously.　The　signals　were　dissembled　to　a　series　of　compactly　supported　wavelet　bases　non-redundantly　and　without　leaking.　The　saltation　coefficient　of　wavelet　analysis　was　regarded　as　a　characteristic　value　to　predict　a　sudden　accident　and　the　changing　trend　of　the　coefficient　was　figured　out.　The　results　showed　that　wavelet　transform　could　capture　the　characteristics　in　a　time　domain　and　the　evolvement　procedure　of　a　fault.　The　trend　graph　of　the　coefficient　could　effectively　predict　the　development　trend　of　a　fault　and　avoid　a　sudden　accident.