Abstract: Based on phase space reconstruction and decomposition, the weighted phase space reconstruction algorithm for noise reduction makes the time series continuation from one dimension to higher dimension, which transforms the feature not easily discriminated in one dimension to the attractor easily discriminated in higher dimension. Distinguishing the different nature and feature of the attractor in higher dimension and pro-jecting the signal from higher dimension to one dimension by adapting the Hanning weighting windows, the natural feature of signal is displayed sufficiently. When the machinery breaks down, their vibration signals often include nonlinear and non-stationary feature. The algorithm is proposed to use for the noise reduction of machinery. This algorithm was analyzed and certificated by numeric value simulation and gear fault diagno-sis. The result makes known that this algorithm is good on noise reduction for this kind of signal.

Key words: Fault diagnoses, Noise reduction, Phase space reconstruction, Time series analysis, microwear, monocrystalline silicon, tribology