Crankshaft bearing early wear fault feature is weak and non-stationary time-varying. A polar coordinate enhancement method of the non-stationary cycling feature is proposed. Using the continuous wavelet transform to achieve signal time- frequency representation for the engine vibration signals, and then each work cycle data is re-sampled according to engine working process with gas-phase relationship, signal feature in the time-frequency is mapped to the polar coordinate and is synchronously enhanced. Finally, feature parameters are classified by the fuzzy C-means clustering. Simulated signal is used to testify the effectiveness of this method. The results show that the method can weaken the noise disturbance and strengthen the signal feature. Using this method to diagnose crankshaft bearing early wear fault for a certain type engine, effectively extract the crankshaft bearing fault feature information and accurately identify different crankshaft bearing wear faults.