Abstract: Feature extraction of vibration signal as a part of condition monitoring and fault diagnosis of mechanical equipment is a fundamental and key issue and has been proven to be highly effective. Focusing on the defects of different signal representation, an adaptive signal decomposition technique is presented to extract fault feature waveform from vibration sig-nals. This method is a modified algorithm of matching pursuit (MP) which decomposes the observed signal into series of ex-pansion of waveforms, these waveforms are calculated by the nonparametric waveform estimation (NWE) and used to best match the signal local structures. So, the construction of gen-eral basis function described by some parameters is unneces-sary. A prior information about the observed signal is no more required for the template signal in the NWE owing to applying the adaptive template signal, which felicitates the method for a wide variety of applications. Both computer simulation and experimental results verify this approach is practicable and effective, even if the noise frequency band coinsides with that of the feature waveform of the signal.

Key words: Feature extraction, Matching pursuit, Nonparametric waveform estimation