Abstract: As the gear-box non-stationary fault information submerged in strong noise environment is a mixture of sustained oscillations and non-oscillatory transients, which are frequency-overlapped and cannot be processed effectively based on frequency, this paper proposes a method that constructs composite Q-factor bases (high-Q factor base and low-Q factor base) adaptively to process the fault signal, and a favorable effect was achieved in the extraction of the impulse signal characteristics of gearbox faults. The analysis result of simulation data and application examples confirm the feasibility and validity of this method, which provides a separation of mechanical fault signals in strong noise environment.