Abstract:Impulsive modulated type signal is the characteristic response of a defected roller bearing. How to extract the impulsive signal from the noised vibration signal becomes the key step for fault diagnosis of bearing. A novel method named Adaptive Multi-scale Morphological Gradient (AMMG) based on the mathematical morphology is proposed for feature extraction of roller bearing fault signal in this study. The AMMG technique has the advantage to depressing the noise as well as keeping the detail components of the signal. Comparing study is done with the most used envelope demodulation and the morphological close method. Results of application to the simulated signal and the real bearing fault signal demonstrated that the proposed AMMG technique can extracts the impulse signal more effectively from the original signal disturbed with strong background noise. Moreover, the computation of AMMG is relatively simple and it has provided one an effective and fast way to feature extraction for fault diagnosis of roller bearing.