Hybrid time-frequency methods for non-stationary mechanical signal analysis

This paper presents a family of hybrid time-frequency methods to be used as a tool for transient signal analysis. These methods are based on autoregressive models of the signal, and on the maximum likelihood method. A classical Wigner-Ville distribution and a parametric time-frequency methods are used for comparison purposes. The characteristics, advantages and disadvantages, of non-parametric, parametric and hybrid methods are discussed and their performances are compared by analysing actual data. The superiority of time-frequency resolution of the hybrid methods is pointed out over that of the parametric and non-parametric ones. Three practical examples of transient signals illustrate the discussion: the vibration signal issued from a damaged gearbox, a signal from a pendular scratch test presenting beating phenomenon and the vibration signal measured in an electrical motor during start-up. (C) 2004 Elsevier Ltd. All rights reserved.