Validating the separability of independent components from NDE sonic signals

Recently, independent component analysis (ICA) has been introduced in non-destructive evaluation (NDE) by impact-echo. Vibrations induced by the hammer impact are sensed through a multichannel system as a convolutive mixture between the excitation signal and the internal defects of the tested material. It is modelled that each defect, as internal acoustic focuses, originates a component or components to be separated. This paper assesses the separability of the sources estimated by ICA in NDE sonic signals by means of simulations and experiments. To evaluate the statistical reliability of the estimated independent components a resampling approach is applied using two different implementations based on separability matrix and clustering visualization. Results from finite element models and experiments on parallelepiped-shape material of aluminium alloy with up to three defects are presented. From simulations and experiments it is shown that estimated components are separable and from simulations it is shown that those components are originated from the material defects