Blind source separation for classification and detection of flaws in impact-echo testing

This paper presents a technique for defect detection and material classification based on blind source separation by ICA for application in impact-echo testing. The transfer functions between the impact point and the defects in the material are modelled as “sources” for blind source separation. The sensors located on the material surface measure a convolutive mixture of the contribution of each of the defects. From spectral analysis the dominant resonance frequencies, varying from homogeneous to defective material, are selected. These frequencies are processed by instantaneous ICA in order to obtain more information about the defects using bootstrap resampling to analyze the stability of the solution. Results show that source estimates fit well with the theoretical response of the material. In addition, it has been found that the number of defects can be estimated by ICA in simulations and experiments with various defective parallelepiped-shape materials of aluminium alloy series 2000.