Evaluation of Delamination Area of Composite Materials Based on Multiscale Features of Ultrasonic Lamb Waves and Neural Network

ABSTRACT To address the problem of quantitative analysis for delamination damage in composite materials, a method of evaluating delamination area based on Lamb waves multiscale features is proposed. In this method, the Lamb wave scattering signals are collected from composite plate with delamination defects using finite element simulation, and the multiscale feature vectors of time-frequency domain are extracted by using complete ensemble empirical mode decomposition with adaptive noise algorithm. In addition, the delamination area can be evaluated and predicted through a generalized regression neural network by taking advantage of nonlinear mapping capability. The hyperparameters of the neural network are also optimized using genetic algorithm, and the feature vectors calculated at different scales are assigned to the network for training and verification. The results show that the multiscale features of delamination damage are more accurate and stable for the model. The mean value and the mean square deviation of mean absolute percentage error proposed in this study is 13.35 % and 4.35 %, respectively, indicating that the overall performance is better than using single scale features and traditional signal decomposition methods.