Crack-type damage detection and localization of a thick composite sandwich structure based on Convolutional Neural Networks

Efficient and reliable Structural Health Monitoring (SHM) systems are required for structural damage detection of composite materials which are widely used in aeronautics and aerospace. In this work, a numerical approach for crack-type damage detection and localization of a thick composite sandwich structure based on a deep learning algorithm Convolutional Neural Networks (CNN) is reported. An intact composite sandwich plate with a thick honeycomb core and same plates with cracks are modelled in ANSYS. Excitation pulse signals are applied on the plate surface to stimulate vertical displacement. Raw vibration response signals are transformed into 2D images by continuous wavelet transform, which are used as input to CNN for training and test of the network. A two-stage CNN is constructed to detect the occurrence and the localization of the damage. Besides, impact of the number of sensors and their emplacement on the efficiency of the CNN-based SHM is investigated, which may provide a reference for similar studies.