Damage identification of CFRP laminate based on cascaded wavelet scattering transform and 1D convolutional neural network

Accurate identification of damage in CFRP structures under limited sample conditions is critical for ensuring the safety of in-service structures. To detect damage of CFRP structures, a cascaded wavelet scattering transform and 1D convolutional neural network (WST-1DCNN) method was proposed in this paper. First, to capture the primary damage features in Lamb wave signals, a wavelet scattering transform stage was designed. On this basis, a 1D convolutional neural network stage was designed to identify damage in CFRP structures. This stage could autonomously extract multi-scale hierarchical features through its layered architecture. Finally, the proposed method was verified on a finite element simulation data and an authoritative public CFRP composites data set of NASA prognostics data repository. The results demonstrated that the proposed WST-1DCNN method achieved better performance on test samples containing damage types absent in the training set. This paper provides a feasible method for identifying damage in CFRP structures under limited sample conditions.