Localization of low-velocity impact in CFRP plate using time–frequency features of guided wave and convolutional neural network

Abstract

Carbon fiber reinforced polymer (CFRP) could be damaged by low-velocity impacts. Precise localization of impact can increase the efficiency of nondestructive testing by narrowing the scanning in a small area. The conventional localization method is based on the time-of-arrival (ToA) of impact induced guided wave, whose localization accuracy could be influenced by the reflected waves from plate edges. This paper proposes an impact localization method based on time–frequency features of guided wave and convolutional neural network. To test the algorithm, 585 impact experiments are performed at 117 different locations of a CFRP plate. Signals corresponding to 12 randomly selected locations are treated as test set, and the remaining signals are used for training of the network. The results show that the average localization error of the proposed method is 10.2 mm.