Motor current signature analysis (MCSA) has been successfully used in squirrel cage motor for the diagnosis of incipient broken rotor bar defect. However, when the motor operates in time-varying rotating speed condition, the method fails as the rotating speed and slip vary, thus, the stator current is characterized as a non-stationary signal. This paper proposes a novel fault diagnosis method for the incipient broken rotor bar of squirrel cage motor in time-varying rotating speed operation condition. The core of the method is to conduct discrete wavelet transformation on the 3 phase stator current Park's vector modulus square signal; and according to the waveform evolution spectrum of broken rotor bar fault frequency 2sf characteristic pattern and the wavelet energy changing during the speed continuous variation, it is judged if the broken rotor bar defect occurs. The experiment verification was performed on a 3kw motor experiment test bench; the experiment results prove the effectiveness of the proposed method.