Application of machine learning for defect detection in FRP-bonded systems with different defect sizes

Acoustic-laser technique has been developed as a promising method to detect defects in structures by vibrating the target object with an acoustic excitation, especially to identify near-surface defects in fiber-reinforced polymer (FRP)-bonded systems. The vibration characteristics are measured by laser beam to determine the integrity of interfacial bonding in structural systems. The sensitivity of acoustic-laser technique can be affected by several operational parameters. The limitation of data acquisition system and the missing data during measurement can influence the accuracy of defect detection. The defect size can also affect the effectiveness of acoustic-laser technique as the acoustic wave is unable to excite the defect region if the defect size is too small. To efficiently reconstruct acoustic-laser measurement for continuous or random missing data situations, a machine learning approach is proposed considering the effect of defect size. This method is based on K-singular value decomposition (K-SVD) with the orthogonal matching pursuit (OMP) algorithm. In this study, FRP-bonded systems with two different sizes of interfacial defect are adopted in the experimental measurement using acoustic laser technique for defect detection. The results demonstrate the effectiveness of machine learning method in the reconstruction of the missing information for electrical signals. The reconstructed data is more reliable for the cases with smaller defect sizes and random missing data. For further application in a broader range, more measured results of defect size should be considered in the dataset of the proposed method.