Decomposition of mechanical signals

Time frequency transformations have gained increasing attention for the characterization of non-stationary signals in a broad spectrum of science and engineering applications. Signals encountered in rotary machine systems can be broadly classified as being either stationary or nonstationary. This study evaluates the performance of the traditional method-polynomial fit filtering with Fourier spectrum analysis and the new developed method-empirical mode decomposition with Hilbert transform (EMD+HT), in mechanical signal decomposition. The former method is based on the sense of least squares, thus insensitive to noise. However, it demands a predetermined time scale, which is unchangeable once fixed, while EMD is adaptive with multi-resolution and univocal. One shortcoming of the latter approach-sensitive to noise, is alleviated by the wavelet threshold de-noising method. A synthetic signal as well as a path error signal of precision working table is analyzed using the two methods. Evaluation is made upon the mode mixing phenomenon and illusive components problem included in EMD with proposed indicators, which confirms the validity of this method used in mechanical signal decomposition.