The operation of gearbox is under a varied working condition with the change of speed and load,and the vibration signal is non-linear and multi-components,so it’ s difficult to realization of gearbox fault diagnose in varied working condition. Firstly,The original vibration signal interpolated by uniform angle re-sampling time through order tracking based on numerical differentiation,which transforms non-stationary vibration signals into angle domain stationary signal. Subsequently,the shock component,harmonic component and noise component are separated from angular domain stationary signal through morphological component analysis（MCA）,which can extract the fault feature from non-linear and multi-component signal in gearbox.Then,the angle domain averaging of shock component highlight the fault characteristic. Finally,by using instantaneous power spectrum（IPS） to identify failure of gearbox. The experimental results show that,using this method,the distribution of IPS in domain of angle and order can identify gear faults,and further testing the method also apply to gearbox fault diagnose under the condition of variable condition.