Impact-acoustics-based health monitoring of tile-wall bonding integrity using principal component analysis

The use of the acoustic features extracted from the impact sounds for bonding integrity assessment has been extensively investigated. Nonetheless, considering the practical implementation of tile-wall non-destructive evaluation (NDE), the traditional defects classification method based directly on frequency-domain features has been of limited application because of the overlapping feature patterns corresponding to different classes whenever there is physical surface irregularity. The purpose of this paper is to explore the clustering and classification ability of principal component analysis (PCA) as applied to the impact-acoustics signature in tile-wall inspection with a view to mitigating the adverse influence of surface non-uniformity. A clustering analysis with signature acquired on sample slabs shows that impact-acoustics signatures of different bonding quality and different surface roughness are well separated into different clusters when using the first two principal components obtained. By adopting as inputs the feature vectors extracted with PCA applied, a multilayer back-propagation artificial neural network (ANN) classifier is developed for automatic health monitoring and defects classification of tile-walls. The inspection results obtained experimentally on the prepared sample slabs are presented and discussed, confirming the utility of the proposed method, particularly in dealing with tile surface irregularity.