Spectrum construction for non stationary vibration: Application to a moving flexible robot

This paper presents a method for constructing spectrum in non-stationary vibration using time series vector autoregressive model. A modal classification criterion and a modal amplification factor are introduced based on the eigen-decomposition of the block data matrix. The classical spectrum computation can be therefore modified to amplify only the physical modes and lighten the computation induced modes. In combination with a sliding window technique, it is shown that the method provides a clear, lightweight spectral representation of non stationary vibration. Application on the vibration signals measured from a moving flexible robot along a given trajectory shows the effectiveness and applicability of the method. The technique is integrated in the online modal surveillance software call STAR (Short Time AutoRegressive).