The　sparse　decomposition　based　on　Q-factor　is　an　adaptive　sparse　expression　method　for　signals.　Aiming　at　the　problem　that　the　gearbox　non-stationary　composite　fault　signal　submerged　in　strong　noise　environment　is　difficult　to　be　extracted　and　segmented,　a　fast　independent　component　analysis　method　based　on　the　parallel　dual-Q-factors　is　proposed.　Firstly,　the　wavelet　transform　analysis　method　based　on　the　parallel　dual-Q-factors　is　used　to　perform　the　denoising　and　dimension　raising　of　the　single-channel　mechanical　fault　signal,　many　low-resonance　impact　components　are　obtained　according　to　different　Q-factors,　which　make　up　multi-dimension　signal　and　signal　dimension-raising　is　achieved.　Secondly,　the　fast　independent　component　analysis　method　is　used　to　carry　out　the　blind　separation　of　the　composite　fault　signals.　The　data　analysis　results　of　simulation　signal　and　the　analysis　results　of　the　experiment　data　for　the　roller　bearing　composite　faults　confirm　the　feasibility　and　validity　of　this　method.　The　proposed　method　provides　a　new　idea　for　the　separation　and　extraction　of　the　mechanical　composite　fault　signals　in　strong　noise　environment.